diff --git a/.agents/skills/setup-matt-pocock-skills/triage-labels.md b/.agents/skills/setup-matt-pocock-skills/triage-labels.md
index b716855d..376e1e4b 100644
--- a/.agents/skills/setup-matt-pocock-skills/triage-labels.md
+++ b/.agents/skills/setup-matt-pocock-skills/triage-labels.md
@@ -2,13 +2,13 @@
 
 The skills speak in terms of five canonical triage roles. This file maps those roles to the actual label strings used in this repo's issue tracker.
 
-| Label in mattpocock/skills | Label in our tracker | Meaning                                  |
-| -------------------------- | -------------------- | ---------------------------------------- |
-| `needs-triage`             | `needs-triage`       | Maintainer needs to evaluate this issue  |
-| `needs-info`               | `needs-info`         | Waiting on reporter for more information |
-| `ready-for-agent`          | `ready-for-agent`    | Fully specified, ready for an AFK agent  |
-| `ready-for-human`          | `ready-for-human`    | Requires human implementation            |
-| `wontfix`                  | `wontfix`            | Will not be actioned                     |
+| Label in our tracker | Meaning                                  |
+| -------------------- | ---------------------------------------- |
+| `needs-triage`       | Maintainer needs to evaluate this issue  |
+| `needs-info`         | Waiting on reporter for more information |
+| `ready-for-agent`    | Fully specified, ready for an AFK agent  |
+| `ready-for-human`    | Requires human implementation            |
+| `wontfix`            | Will not be actioned                     |
 
 When a skill mentions a role (e.g. "apply the AFK-ready triage label"), use the corresponding label string from this table.
 
diff --git a/.archon/commands/.gitkeep b/.archon/commands/.gitkeep
new file mode 100644
index 00000000..e69de29b
diff --git a/.archon/commands/unic-dlc-build.md b/.archon/commands/unic-dlc-build.md
new file mode 100644
index 00000000..95ef3f38
--- /dev/null
+++ b/.archon/commands/unic-dlc-build.md
@@ -0,0 +1,59 @@
+# /unic-dlc-build
+
+Execute the TDD build workflow for a planning session, enforcing red → green per issue.
+
+## Usage
+
+```
+/unic-dlc-build <slug>
+```
+
+Where `<slug>` is the session identifier used when you ran `/unic-dlc-plan`. The generated
+`.archon/workflows/build-<slug>.yaml` must already exist (produced by the `yaml-gen` node in plan).
+
+## What this command does
+
+1. **Slopcheck gate** — before any implementation begins, scans `package.json` (and other
+   manifest files) for packages introduced since the last commit. Each new package is checked
+   against the npm registry. Packages that fail the check are flagged `[ASSUMED]` and require
+   explicit human approval before the build can continue.
+
+   Registry check strategy (in order of preference):
+
+   - Python `slopcheck` tool (GSD's slopsquatting gate) if available on `PATH`
+   - npm registry HEAD request fallback
+   - If neither is available: all new packages are treated as `[ASSUMED]` (strict default)
+
+   To bypass slopcheck for known-safe cases: `SLOPCHECK_BYPASS=1 /unic-dlc-build <slug>`
+
+2. **Generated build workflow** — executes `.archon/workflows/build-<slug>.yaml` which was
+   produced by the `yaml-gen` node in `/unic-dlc-plan`. The generated workflow:
+   - Issues a `code-red` node per issue: writes FAILING acceptance tests
+   - Issues a `code-green` node per issue: writes minimum implementation to pass those tests
+   - Enforces `code-red` before `code-green` within each issue via `depends_on` edges
+   - Runs independent issues' `code-red` (and `code-green`) phases in parallel
+
+## Prerequisites
+
+- `/unic-dlc-plan <slug>` must have been run and the plan PR approved
+- `.archon/workflows/build-<slug>.yaml` must exist
+- `.archon/unic-dlc.config.json` must be present (created by the install hook)
+
+## TDD contract
+
+Every issue in the build goes through:
+
+```
+RED:   code-red-<id>   → write failing acceptance tests
+GREEN: code-green-<id> → write minimum implementation to pass those tests
+```
+
+No `code-green` node may run before its corresponding `code-red` node completes.
+Independent issues run their `code-red` and `code-green` phases in parallel,
+giving the fastest possible end-to-end build time.
+
+## Runs
+
+```
+archon workflow run unic-dlc-build --input slug=<slug>
+```
diff --git a/.archon/commands/unic-dlc-cleanup.md b/.archon/commands/unic-dlc-cleanup.md
new file mode 100644
index 00000000..36497ac2
--- /dev/null
+++ b/.archon/commands/unic-dlc-cleanup.md
@@ -0,0 +1,61 @@
+# /unic-dlc-cleanup
+
+Post-merge cleanup for a planning session: architecture review, ADR consolidation, and a
+triage pass to refresh project state.
+
+## Usage
+
+```
+/unic-dlc-cleanup <slug>
+```
+
+Where `<slug>` is the session identifier used throughout the plan → build → qa cycle.
+
+## What this command does
+
+1. **`arch-review` node** — reads `docs/workflow/<slug>/PRD.md` (intent) and
+   `docs/workflow/<slug>/report.md` (technical outcome) alongside the changed code.
+   Detects three categories of drift:
+
+   - **Technical drift**: too-shallow modules, tight coupling, leaky abstractions.
+   - **Intent drift**: delivered behaviour that diverges from the PRD; silently dropped
+     acceptance criteria; scope creep added during build.
+   - **Deepening opportunities**: modules that could hide more complexity behind their
+     current interface.
+
+   Output: `docs/workflow/<slug>/arch-review.md`
+
+2. **`adr-consolidation` interactive node** — presents each proposed ADR individually for
+   human approval. Sources:
+
+   - "Decisions Made" section of `report.md`
+   - "Accept as ADR" items from `arch-review.md`
+
+   Each ADR is shown with Context, Decision, and Consequences. The user accepts (A),
+   rejects (R), or edits (E) each candidate. Only accepted ADRs are written to `docs/adr/`.
+
+3. **`run-triage` node** — invokes the shared triage workflow by name via
+   `archon workflow run unic-dlc-triage`. No triage logic is duplicated in the cleanup
+   workflow. This produces `HANDOFF.md` and updates `docs/workflow/ROADMAP.md` exactly
+   once per cleanup run.
+
+## Prerequisites
+
+- The build and QA cycles for `<slug>` must be complete (build PR merged, QA approved)
+- `docs/workflow/<slug>/PRD.md` and `docs/workflow/<slug>/report.md` must exist
+- `.archon/unic-dlc.config.json` must be present
+
+## Outputs
+
+| File                                  | Description                                     |
+| ------------------------------------- | ----------------------------------------------- |
+| `docs/workflow/<slug>/arch-review.md` | Architecture review with drift findings         |
+| `docs/adr/NNNN-*.md`                  | Accepted ADRs (one file per accepted candidate) |
+| `HANDOFF.md`                          | Updated handoff snapshot (from triage)          |
+| `docs/workflow/ROADMAP.md`            | Updated project phase (from triage)             |
+
+## Runs
+
+```
+archon workflow run unic-dlc-cleanup --input slug=<slug>
+```
diff --git a/.archon/commands/unic-dlc-explore.md b/.archon/commands/unic-dlc-explore.md
new file mode 100644
index 00000000..29ba263e
--- /dev/null
+++ b/.archon/commands/unic-dlc-explore.md
@@ -0,0 +1,54 @@
+---
+description: Run the unic-archon-dlc explore workflow — parallel research across stack, features, architecture, and pitfalls, then synthesize into docs/workflow/<slug>/findings.md
+---
+
+# /unic-dlc-explore
+
+Runs the `explore` workflow: four parallel research agents investigate the project from different angles, then a synthesize node combines their findings into a single `findings.md` document.
+
+## When to use
+
+- At the start of a new feature or investigation to build a shared mental model.
+- When onboarding to an unfamiliar project or module.
+- Before writing a spec, to surface pitfalls and architecture constraints early.
+- Any time you want a structured snapshot of what the project is, what it does, and where the risks are.
+
+## What it produces
+
+- **`docs/workflow/<slug>/findings.md`** — five sections:
+  - **Stack** — runtime, package manager, toolchain
+  - **Features** — shipped capabilities, in-progress work, planned features
+  - **Architecture** — directory structure, design patterns, ADR decisions, cross-platform constraints
+  - **Pitfalls** — TODOs, untested modules, CI issues, HANDOFF.md blockers
+  - **Integrated Brief** — 3-5 sentence synthesis with the highest-impact next step
+
+## Usage
+
+```sh
+archon workflow run unic-dlc-explore --input slug=<slug>
+```
+
+Or invoke from Claude Code:
+
+```
+/unic-dlc-explore <slug>
+```
+
+Replace `<slug>` with a short identifier for this exploration (e.g. `auth-refactor`, `v2-planning`).
+
+## Workflow structure
+
+```
+research-stack  ──┐
+research-features ─┤
+                   ├──▶  synthesize  ──▶  findings.md
+research-architecture ─┤
+research-pitfalls ──┘
+```
+
+All four research nodes are independent and run in parallel. The `synthesize` node depends on all four and writes the final output.
+
+## Inspiration
+
+- Parallel research pattern from the Archon workflow specification.
+- This plugin's `lib/findings-writer.mjs` handles idempotent directory and file creation.
diff --git a/.archon/commands/unic-dlc-plan.md b/.archon/commands/unic-dlc-plan.md
new file mode 100644
index 00000000..a659f25e
--- /dev/null
+++ b/.archon/commands/unic-dlc-plan.md
@@ -0,0 +1,61 @@
+---
+description: Run the unic-archon-dlc plan workflow — adversarial spec interview, PRD synthesis, and human PR gate
+---
+
+# /unic-dlc-plan
+
+Runs the `plan` workflow: loads project context and prior research, runs an adversarial interview to surface requirements and decisions, synthesises the transcript into a structured PRD, and opens a PR for human review before proceeding.
+
+## When to use
+
+- After `/unic-dlc-explore` to turn research findings into a formal PRD.
+- When starting a new feature and you want to stress-test assumptions before writing specs.
+- Any time you need a structured `PRD.md` that captures problem, solution, stories, decisions, and scope.
+
+## What it produces
+
+- **`docs/workflow/<slug>/PRD.md`** — seven sections: Problem Statement, Solution, User Stories, Implementation Decisions, Testing Decisions, Out of Scope, Further Notes.
+- **`docs/adr/NNNN-<slug>.md`** — one or more ADR files for non-obvious decisions surfaced during the interview (written live, confirmed by user).
+- **PR** targeting `develop` — pauses until approved. Rejection returns control to the adversarial interview for refinement.
+
+## Usage
+
+```sh
+archon workflow run unic-dlc-plan --input slug=<slug>
+```
+
+Or invoke from Claude Code:
+
+```
+/unic-dlc-plan <slug>
+```
+
+Replace `<slug>` with a short identifier for this planning session (e.g. `auth-refactor`, `v2-planning`).
+
+## Options
+
+Set `workflow.discuss_mode` in `.archon/unic-dlc.config.json` to control the interview style:
+
+| Value                 | Behaviour                                                                |
+| --------------------- | ------------------------------------------------------------------------ |
+| `interview` (default) | One focused, adversarial question per turn; probes deeply                |
+| `assumptions`         | Surfaces all implicit assumptions upfront, then confirms or refutes each |
+
+## Workflow structure
+
+```
+load-context ──▶ specs (loop) ──▶ to-prd ──▶ prd-gate (interactive)
+                     ▲                              │
+                     └──────── rejected ────────────┘
+```
+
+- `load-context` reads CONTEXT.md, CONTEXT-MAP.md, all ADRs, and findings.md (if present).
+- `specs` runs the adversarial interview; writes ADRs live for confirmed decisions.
+- `to-prd` synthesises the interview transcript into PRD.md.
+- `prd-gate` validates all 7 sections, opens a PR, and waits for human approval.
+
+## Inspiration
+
+- The `grill-with-docs` skill — adversarial interview pattern against the domain model.
+- The `to-prd` skill — structured PRD synthesis from conversation context.
+- Archon PR-gate pattern for human-in-the-loop workflow checkpoints.
diff --git a/.archon/commands/unic-dlc-qa.md b/.archon/commands/unic-dlc-qa.md
new file mode 100644
index 00000000..f44bfbee
--- /dev/null
+++ b/.archon/commands/unic-dlc-qa.md
@@ -0,0 +1,78 @@
+---
+description: Run the unic-archon-dlc QA workflow — e2e tests, coverage gate, UAT checklist, and merge
+---
+
+# /unic-dlc-qa
+
+Runs the `qa` workflow: executes end-to-end tests, enforces the coverage threshold, walks through
+a UAT checklist derived from the PRD's acceptance criteria, and merges the PR once the human approves.
+
+## When to use
+
+- After `/unic-dlc-build <slug>` is complete and the build PR has been reviewed and approved.
+- When you want a structured QA gate before merging a feature branch.
+- Any time you need to confirm that acceptance criteria are met before shipping.
+
+## What it produces
+
+- **e2e results** — pass/fail output from the command configured in `e2e_command`.
+- **Coverage report** — pass/fail against the `coverage_threshold` set in `.archon/unic-dlc.config.json`.
+- **UAT checklist** — numbered acceptance criteria from `docs/workflow/<slug>/PRD.md`, presented alongside build coverage evidence.
+- **Merged PR** — once UAT is approved, the PR is merged via the configured tracker CLI and the feature branch is cleaned up (Gitflow only).
+
+## Usage
+
+```sh
+archon workflow run unic-dlc-qa --input slug=<slug>
+```
+
+Or invoke from Claude Code:
+
+```
+/unic-dlc-qa <slug>
+```
+
+Replace `<slug>` with the same identifier used in `/unic-dlc-plan` and `/unic-dlc-build`.
+
+## Prerequisites
+
+- `/unic-dlc-build <slug>` must have been run and the build PR approved.
+- `.archon/unic-dlc.config.json` must be present (created by the install hook).
+- `e2e_command` must be set in the config. If not, run `archon install --reconfigure` to set it.
+- The current branch must have an open PR targeting `develop` (for the merge step).
+
+## Workflow structure
+
+```
+e2e  ──▶  coverage-gate  ──▶  uat-gate (interactive)  ──▶  merge
+```
+
+- `e2e` — reads `e2e_command` from config; fails fast with a helpful message if it is not set.
+- `coverage-gate` — runs `pnpm test --coverage`; parses the output to compare against `coverage_threshold`. Skipped if no threshold is configured.
+- `uat-gate` — presents the numbered UAT checklist; waits for APPROVE or REJECT from the human.
+- `merge` — merges the PR via the tracker CLI (gh / az / manual instructions for jira and local-markdown); deletes the feature branch on Gitflow.
+
+## Configuration reference
+
+All settings are read from `.archon/unic-dlc.config.json`:
+
+| Field                | Type   | Default     | Description                                                              |
+| -------------------- | ------ | ----------- | ------------------------------------------------------------------------ |
+| `e2e_command`        | string | —           | Shell command to run e2e tests (e.g. `"pnpm test:e2e"`)                  |
+| `coverage_threshold` | number | —           | Minimum coverage % required (e.g. `80`). Omit to skip the gate.          |
+| `tracker`            | string | `"github"`  | Issue tracker: `github`, `ado`, `jira`, `local-markdown`                 |
+| `branching`          | string | `"gitflow"` | Branch strategy: `gitflow` (delete feature branch) or `github-flow`      |
+| `pr_strategy`        | string | `"squash"`  | GitHub merge style: `squash` or `merge` (used for `github` tracker only) |
+
+## UAT interaction
+
+The `uat-gate` node pauses and asks for your decision:
+
+- Type **`APPROVE`** to proceed to merge.
+- Type **`REJECT AC-N`** (e.g. `REJECT AC-3`) to flag a failing criterion. The workflow halts and you are returned to address the issue before re-running.
+
+## Runs
+
+```
+archon workflow run unic-dlc-qa --input slug=<slug>
+```
diff --git a/.archon/commands/unic-dlc-review.md b/.archon/commands/unic-dlc-review.md
new file mode 100644
index 00000000..3f016f28
--- /dev/null
+++ b/.archon/commands/unic-dlc-review.md
@@ -0,0 +1,82 @@
+---
+description: Run the unic-archon-dlc self-contained code review workflow — four aspects, structured findings, single PR comment, idempotent re-run
+---
+
+# /unic-dlc-review
+
+Runs the `review` workflow: analyses the current PR diff across four review aspects
+(code quality, test coverage, silent failures, type design), then posts — or updates — a
+single structured comment on the PR.
+
+**No runtime dependency on pr-review-toolkit or any other plugin.**
+All review logic is self-contained in `.archon/workflows/unic-dlc-review.yaml`.
+
+## When to use
+
+- After opening a PR and before requesting human review.
+- As a final quality gate before merging.
+- Re-run it after addressing feedback — it will **update** the prior comment rather than
+  posting a duplicate.
+
+## What it produces
+
+A single structured PR comment (or updated comment on re-run) with four sections:
+
+1. **Code quality** — readability, naming, function length, project-convention adherence
+2. **Test coverage adequacy** — exercises public interfaces, output-focused assertions, no
+   internal-collaborator mocks
+3. **Silent failure patterns** — swallowed exceptions, empty catch blocks, fallbacks that
+   hide bugs
+4. **Type design quality** — encapsulation, invariants in types, illegal-state-unrepresentable
+   patterns
+
+Each section lists specific findings with `file:line` references, or an explicit
+"No findings." line if nothing was detected.
+
+## Usage
+
+```sh
+archon workflow run unic-dlc-review
+```
+
+Or invoke from Claude Code:
+
+```
+/unic-dlc-review
+```
+
+No arguments needed. The workflow reads `.archon/unic-dlc.config.json` to determine the
+tracker and the current open PR.
+
+## Workflow structure
+
+```
+code-review  ──▶  structured comment posted / updated on PR
+```
+
+Single-node DAG. The node reads CLAUDE.md / AGENTS.md for project conventions, runs the
+four review aspects, and calls the tracker adapter to post or update the comment.
+
+## Re-run behaviour
+
+When you run `/unic-dlc-review` again on the same PR:
+
+- The workflow searches for a prior comment whose body contains the sentinel marker
+  `<!-- unic-dlc-review -->`.
+- If found, it **updates** that comment in-place.
+- If not found, it creates a new comment.
+
+This prevents duplicate review threads accumulating over multiple runs.
+
+## Inspiration
+
+- `apps/claude-code/pr-review/` — multi-aspect analysis structure, compact finding
+  schema (`severity / filePath / startLine / endLine / title / body`), re-review
+  detection via sentinel, and the discipline of posting a single summary comment rather
+  than many inline threads.
+- Matt Pocock's review skill
+  (https://github.com/mattpocock/skills/blob/main/skills/in-progress/review/SKILL.md) —
+  aspect-driven review structure, explicit "no findings" lines, and the principle that
+  every run should produce actionable output even when everything looks good.
+- This plugin's `lib/tracker-adapter.mjs` — used to post or update the comment via the
+  configured tracker backend (github / ado / jira / local-markdown).
diff --git a/.archon/commands/unic-dlc-triage.md b/.archon/commands/unic-dlc-triage.md
new file mode 100644
index 00000000..e0eb8ad2
--- /dev/null
+++ b/.archon/commands/unic-dlc-triage.md
@@ -0,0 +1,35 @@
+---
+description: Run the unic-archon-dlc triage workflow — produces HANDOFF.md and updates ROADMAP.md
+---
+
+# /unic-dlc-triage
+
+Runs the `triage` workflow: reads current issue states from the configured tracker, reconciles them against `docs/workflow/ROADMAP.md`, and produces `HANDOFF.md` at the repo root.
+
+## When to use
+
+- After a session boundary to resume cleanly in a fresh context.
+- As the final step of the `/unic-dlc-cleanup` workflow (reused by the cleanup DAG).
+- Any time you want a status snapshot of open issues, blockers, and recent decisions.
+
+## What it produces
+
+- **`HANDOFF.md`** (repo root) — four sections: current phase, open issues by state, blockers, recent decisions.
+- **`docs/workflow/ROADMAP.md`** — created on first run; phase status updated on each run. Human-edited sections are preserved.
+
+## Usage
+
+```sh
+archon workflow run unic-dlc-triage
+```
+
+Or invoke from Claude Code:
+
+```
+/unic-dlc-triage
+```
+
+## Inspiration
+
+- Matt Pocock skills repo — `/triage` skill pattern.
+- This plugin's `docs/agents/` files (written by the install hook) provide the domain context consumed by the triage prompt nodes.
diff --git a/.archon/unic-dlc.config.json b/.archon/unic-dlc.config.json
new file mode 100644
index 00000000..4db29997
--- /dev/null
+++ b/.archon/unic-dlc.config.json
@@ -0,0 +1,36 @@
+{
+  "model_profile": "balanced",
+  "tdd_mode": true,
+  "nyquist_validation": true,
+  "slopsquatting_gate": true,
+  "tracker": "github",
+  "pr_strategy": "merge",
+  "branching": "gitflow",
+  "e2e_command": null,
+  "repo_layout": "multi-context",
+  "labels": {
+    "state": {
+      "needs-triage": "needs-triage",
+      "needs-info": "needs-info",
+      "needs-specs": "needs-specs",
+      "ready-for-agent": "ready-for-agent",
+      "ready-for-human": "ready-for-human",
+      "resolved": "resolved",
+      "closed": "closed",
+      "rejected": "rejected"
+    },
+    "type": {
+      "feature": "feature",
+      "bug": "bug",
+      "spike": "spike",
+      "tech-debt": "tech-debt",
+      "docs": "docs"
+    },
+    "priority": {
+      "p0": "p0",
+      "p1": "p1",
+      "p2": "p2",
+      "p3": "p3"
+    }
+  }
+}
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/build.yaml b/.archon/workflows/unic-dlc-build.yaml
similarity index 98%
rename from apps/claude-code/unic-archon-dlc/.archon/workflows/build.yaml
rename to .archon/workflows/unic-dlc-build.yaml
index da2433a9..37bc35fc 100644
--- a/apps/claude-code/unic-archon-dlc/.archon/workflows/build.yaml
+++ b/.archon/workflows/unic-dlc-build.yaml
@@ -13,6 +13,7 @@ nodes:
   - id: slopcheck
     name: Slopcheck — Verify New Packages
     type: bash
+    runtime: bun
     depends_on: []
     script: |
       #!/usr/bin/env node
@@ -130,8 +131,8 @@ nodes:
            "build-{{ inputs.slug }}.yaml not found. Run /unic-dlc-plan {{ inputs.slug }} first to generate it."
          and halt.
 
-      2. Invoke the generated workflow via Archon:
-         Run: archon run .archon/workflows/build-{{ inputs.slug }}.yaml --input slug={{ inputs.slug }}
+      2. Invoke the generated workflow via Archon by name:
+         Run: archon workflow run unic-dlc-build-{{ inputs.slug }} --input slug={{ inputs.slug }}
 
          The generated workflow contains:
            - code-red-<id> nodes: write FAILING acceptance tests per issue
@@ -152,6 +153,7 @@ nodes:
   - id: verification
     name: Verification — Test Suite, Stubs, Wiring
     type: bash
+    runtime: bun
     depends_on: [run-build]
     script: |
       #!/usr/bin/env node
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/cleanup.yaml b/.archon/workflows/unic-dlc-cleanup.yaml
similarity index 97%
rename from apps/claude-code/unic-archon-dlc/.archon/workflows/cleanup.yaml
rename to .archon/workflows/unic-dlc-cleanup.yaml
index eb6d54b6..7cbc946e 100644
--- a/apps/claude-code/unic-archon-dlc/.archon/workflows/cleanup.yaml
+++ b/.archon/workflows/unic-dlc-cleanup.yaml
@@ -139,6 +139,7 @@ nodes:
   - id: run-triage
     name: Run Triage — Refresh Project State
     type: bash
+    runtime: bun
     depends_on: [adr-consolidation]
     script: |
       #!/usr/bin/env node
@@ -147,11 +148,11 @@ nodes:
       import { execFileSync } from 'node:child_process'
 
       try {
-        execFileSync('archon', ['run', '.archon/workflows/triage.yaml'], {
+        execFileSync('archon', ['workflow', 'run', 'unic-dlc-triage'], {
           stdio: 'inherit',
           env: { ...process.env },
         })
       } catch (err) {
-        console.error('run-triage: archon run failed:', err.message)
+        console.error('run-triage: archon workflow run failed:', err.message)
         process.exit(1)
       }
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/explore.yaml b/.archon/workflows/unic-dlc-explore.yaml
similarity index 100%
rename from apps/claude-code/unic-archon-dlc/.archon/workflows/explore.yaml
rename to .archon/workflows/unic-dlc-explore.yaml
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/plan.yaml b/.archon/workflows/unic-dlc-plan.yaml
similarity index 99%
rename from apps/claude-code/unic-archon-dlc/.archon/workflows/plan.yaml
rename to .archon/workflows/unic-dlc-plan.yaml
index 6c127069..5c79b155 100644
--- a/apps/claude-code/unic-archon-dlc/.archon/workflows/plan.yaml
+++ b/.archon/workflows/unic-dlc-plan.yaml
@@ -344,6 +344,7 @@ nodes:
   - id: yaml-gen
     name: YAML Gen — Build Workflow Generator
     type: bash
+    runtime: bun
     depends_on: [plan-checker]
     script: |
       #!/usr/bin/env node
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/qa.yaml b/.archon/workflows/unic-dlc-qa.yaml
similarity index 99%
rename from apps/claude-code/unic-archon-dlc/.archon/workflows/qa.yaml
rename to .archon/workflows/unic-dlc-qa.yaml
index 3146594b..dcf50f18 100644
--- a/apps/claude-code/unic-archon-dlc/.archon/workflows/qa.yaml
+++ b/.archon/workflows/unic-dlc-qa.yaml
@@ -12,6 +12,7 @@ nodes:
   - id: e2e
     name: End-to-End Tests
     type: bash
+    runtime: bun
     depends_on: []
     script: |
       #!/usr/bin/env node
@@ -61,6 +62,7 @@ nodes:
   - id: coverage-gate
     name: Coverage Gate
     type: bash
+    runtime: bun
     depends_on: [e2e]
     script: |
       #!/usr/bin/env node
@@ -191,6 +193,7 @@ nodes:
   - id: verify-pr-base
     name: Verify PR Base Branch
     type: bash
+    runtime: bun
     depends_on: [uat-gate]
     script: |
       #!/usr/bin/env node
@@ -253,6 +256,7 @@ nodes:
   - id: merge
     name: Merge PR
     type: bash
+    runtime: bun
     depends_on: [verify-pr-base]
     script: |
       #!/usr/bin/env node
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/review.yaml b/.archon/workflows/unic-dlc-review.yaml
similarity index 100%
rename from apps/claude-code/unic-archon-dlc/.archon/workflows/review.yaml
rename to .archon/workflows/unic-dlc-review.yaml
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/triage.yaml b/.archon/workflows/unic-dlc-triage.yaml
similarity index 100%
rename from apps/claude-code/unic-archon-dlc/.archon/workflows/triage.yaml
rename to .archon/workflows/unic-dlc-triage.yaml
diff --git a/.claude/commands/inbox.md b/.claude/commands/inbox.md
deleted file mode 100644
index 9b0d9608..00000000
--- a/.claude/commands/inbox.md
+++ /dev/null
@@ -1,66 +0,0 @@
----
-allowed-tools: [Write, Bash]
-argument-hint: '<one-liner idea>'
-description: 'Capture a raw idea into docs/inbox/ without interrupting the current conversation'
----
-
-# Inbox capture
-
-**Arguments:** "$ARGUMENTS"
-
-Capture the idea in `$ARGUMENTS` as a file in `docs/inbox/`. Act immediately — no follow-up
-questions.
-
-## Step 1 — Generate a slug
-
-From the idea text:
-
-1. Lowercase everything
-2. Replace spaces and punctuation with hyphens
-3. Collapse multiple hyphens into one
-4. Strip leading/trailing hyphens
-5. Truncate to 50 characters at a word boundary
-
-Examples:
-
-- `"pr-review should support GitLab!"` → `pr-review-should-support-gitlab`
-- `"add dark mode support to the dashboard component"` → `add-dark-mode-support-to-the-dashboard-component`
-- `"implement user authentication with OAuth2 and role-based access control"` → `implement-user-authentication-with-oauth2-and` (truncated at word boundary before 50 chars)
-
-## Step 2 — Check for collisions
-
-```bash
-node -e "
-  const root = require('child_process').execSync('git rev-parse --show-toplevel').toString().trim();
-  const p = require('path').join(root, 'docs', 'inbox', '<slug>.md');
-  process.exit(require('fs').existsSync(p) ? 1 : 0);
-"
-```
-
-If the file exists, append `-2` (then `-3`, etc.) until the path is free.
-
-## Step 3 — Write the file
-
-Resolve the repo root with `git rev-parse --show-toplevel`, then create `<root>/docs/inbox/<slug>.md`:
-
-```markdown
----
-title: <original idea text, trimmed>
-created: <today's date as YYYY-MM-DD>
----
-
-<original idea text, trimmed>
-```
-
-Use the `Write` tool. Do not create any other files.
-
-## Step 4 — Confirm
-
-Reply with exactly one line:
-
-```txt
-Captured → docs/inbox/<slug>.md
-```
-
-Nothing else. No summary, no next steps, no follow-up questions. The user is in the middle
-of something else.
diff --git a/.claude/settings.json b/.claude/settings.json
index 5e66f824..32909885 100644
--- a/.claude/settings.json
+++ b/.claude/settings.json
@@ -1,23 +1,35 @@
 {
+  "permissions": {
+    "allow": [
+      "Bash(pnpm --filter * test)",
+      "Bash(pnpm bump *)",
+      "Bash(pnpm check *)",
+      "Bash(pnpm ci:check)",
+      "Bash(pnpm format *)",
+      "Bash(pnpm typecheck *)",
+      "Skill(archon)",
+      "Skill(commit-commands:commit)"
+    ]
+  },
   "hooks": {
-    "PreToolUse": [
+    "PostToolUse": [
       {
         "matcher": "Write|Edit|MultiEdit",
         "hooks": [
           {
             "type": "command",
-            "command": "node .claude/hooks/block-lockfile.mjs"
+            "command": "node .claude/hooks/test-on-edit.mjs"
           }
         ]
       }
     ],
-    "PostToolUse": [
+    "PreToolUse": [
       {
         "matcher": "Write|Edit|MultiEdit",
         "hooks": [
           {
             "type": "command",
-            "command": "node .claude/hooks/test-on-edit.mjs"
+            "command": "node .claude/hooks/block-lockfile.mjs"
           }
         ]
       }
diff --git a/.claude/skills/implement-feature/SKILL.md b/.claude/skills/implement-feature/SKILL.md
deleted file mode 100644
index 28ef8ab3..00000000
--- a/.claude/skills/implement-feature/SKILL.md
+++ /dev/null
@@ -1,189 +0,0 @@
----
-name: implement-feature
-argument-hint: '[slug]'
-description: This skill should be used when the user asks to "implement a feature", "run the Feature Runner", "/implement-feature", "implement all issues for <slug>", or "drain the issue queue overnight". Automates the implementation side of the AI-development cycle for one Feature: creates an isolated worktree and branch, runs /tdd on every ready-for-agent issue in dependency order, and opens a PR when done.
----
-
-# Implement Feature
-
-Automate the implementation side of the AI-development cycle for one Feature. Takes a slug, creates an isolated branch, runs `/tdd` on every `ready-for-agent` issue in dependency order, and marks each `resolved` on success.
-
-**Invocation:** `/implement-feature [slug]`
-
-## Quick start
-
-- **Named run** — `/implement-feature pr-review-doc-context-enrichment` — targets a specific Feature slug directly; creates a worktree, runs all `ready-for-agent` issues, opens a PR.
-- **Auto-select** — `/implement-feature` with no argument — scans `docs/issues/` and picks the first Feature alphabetically that has at least one `ready-for-agent` issue and no unprepped issues (`needs-triage`, `needs-info`, `needs-specs`). Picks up partial features after a failure fix automatically.
-- **Overnight loop** — `/loop /implement-feature` — drains the queue unattended; the runner emits `LOOP_COMPLETE` when no qualifying Feature remains, which terminates the loop.
-- **Safe to interrupt** — Ctrl+C during any issue leaves that issue at `ready-for-agent`; re-running resumes from the first unresolved issue. Note: a **/tdd failure** (as opposed to a Ctrl+C interrupt) sets the failing issue to `needs-info`, preventing auto-select from retrying until the developer intervenes.
-
-## Steps
-
-### 0. Resolve the slug
-
-**If a slug argument was provided**, use it directly and proceed to step 1.
-
-**If no argument was provided**, scan `docs/issues/` for qualifying features:
-
-1. Use the Bash tool to list immediate subdirectories of `docs/issues/`:
-
-```
-ls -d docs/issues/*/
-```
-
-2. For each subdirectory (potential feature slug), use the Bash tool to list its `NN-*.md` files and use the Read tool to check the `**Status:**` line of each one. A feature **qualifies** if:
-
-   - At least one `NN-*.md` file has status `ready-for-agent`, **and**
-   - Every `NN-*.md` file has a status in `{ready-for-agent, resolved, closed, rejected, ready-for-human}`.
-
-   Any file with status `needs-triage`, `needs-info`, `needs-specs`, or any unrecognised state **disqualifies the whole feature** — it is not fully prepped for autonomous execution. Features where every file is `resolved`, `closed`, or `rejected` (nothing left to run) are also skipped.
-
-3. Sort the qualifying slugs alphabetically and select the first one.
-
-4. **If no qualifying feature exists**, emit the **LOOP_COMPLETE signal** (see `references/runner-output-formats.md`) on its own line and exit cleanly (no error). Do not output anything after it.
-
-5. **If a qualifying feature was found**, set the slug to that feature's directory name and continue to step 1.
-
-### 1. Resolve the feature directory and assemble the static context bundle
-
-The slug argument maps directly to `docs/issues/<slug>/`. Use the Bash tool to confirm:
-
-1. The directory exists (e.g. `ls docs/issues/<slug>/`).
-2. It contains at least one `NN-*.md` file (e.g. `ls docs/issues/<slug>/[0-9]*.md`).
-3. At least one of those files has `**Status:** ready-for-agent` (use the Read tool on each file and inspect the `**Status:**` line).
-
-If any of these checks fails, stop and report the specific reason to the user. Do not create a worktree.
-
-**Read the PRD:** `docs/issues/<slug>/PRD.md`. Scan its content for references matching `apps/claude-code/<plugin>/` (any path that starts with that prefix). This determines the ADR scope:
-
-- **Plugin feature** — one or more `apps/claude-code/<plugin>/` references found → use that plugin's `apps/claude-code/<plugin>/CONTEXT.md` and `apps/claude-code/<plugin>/docs/adr/`. Do **not** also inject root ADRs.
-- **Repo/tooling feature** — no such references found → use root `CONTEXT.md` and root `docs/adr/`.
-
-Also extract the `title:` field from the PRD's YAML frontmatter (the value between the opening `---` and closing `---` at the top of the file). Retain it for use in step 7's PR title derivation.
-
-**Read the scoped CONTEXT.md** using the Read tool.
-
-**Read all ADR files** in the scoped ADR directory: list `*.md` files using the Bash tool, then read each one using the Read tool.
-
-**Get the last 5 git commits** using the Bash tool:
-
-```
-git log --oneline -5
-```
-
-These items (PRD content + title, CONTEXT.md, ADRs, recent commits) are static — gather them once before the issue loop begins.
-
-### 2. Create the worktree and branch
-
-First, check whether a worktree from a prior run already exists using the Bash tool:
-
-```
-ls .claude/worktrees/<slug>
-```
-
-- **Exists** — reuse it. The branch `feature/afk/<slug>` already contains the committed work from the previous run. Skip `git worktree add`.
-- **Does not exist** — create it using the Bash tool:
-
-```
-git worktree add .claude/worktrees/<slug> -b feature/afk/<slug> develop
-```
-
-The worktree lands at `.claude/worktrees/<slug>`. All subsequent implementation work happens inside that worktree.
-
-### 3. Collect issues, build the dependency graph, and derive execution order
-
-Use the Bash tool to list **all** `NN-*.md` files in `docs/issues/<slug>/` (including `resolved` and `closed` — they are needed for graph completeness):
-
-```
-ls docs/issues/<slug>/[0-9]*.md
-```
-
-Use the Read tool to read each file. For every file record:
-
-- Its **numeric prefix** (the `NN` integer from the filename).
-- Its **status** (`**Status:**` line).
-- Its **`## Blocked by`** list — the filenames or paths referenced there. `## Blocked by: None`, `## Blocked by: None — can start immediately`, or a missing `## Blocked by` section all mean no predecessors.
-
-**Missing-blocker check — halt before executing anything if violated:**
-
-For each `## Blocked by` reference, verify the referenced filename actually exists in `docs/issues/<slug>/`. If a referenced blocker file is missing (typo, renamed, deleted), halt immediately with the **missing blocker error** (see `references/runner-output-formats.md`), naming the issue and the unresolvable reference. Do not silently treat it as satisfied.
-
-**Conflict check — halt before executing anything if violated:**
-
-For each issue A that lists issue B in `## Blocked by`: if B's numeric prefix is greater than A's numeric prefix, the dependency contradicts numerical convention. Halt immediately with the **dependency conflict error** (see `references/runner-output-formats.md`), naming both issues.
-
-**Build the execution queue:**
-
-From the dependency graph, compute a topological order over all issues (using `## Blocked by` edges). Filter the topological sequence to only `ready-for-agent` issues — `resolved`, `closed`, and `rejected` issues are satisfied and act as satisfied dependency nodes, not as items to execute.
-
-**Unsatisfied dependency check — halt before executing anything if violated:**
-
-For each `ready-for-agent` issue in the execution queue, inspect its `## Blocked by` list. If any listed blocker has status `ready-for-human`, halt immediately with the **unsatisfied dependency error** (see `references/runner-output-formats.md`), naming both issues.
-
-This ordered list is the execution queue. Record M = number of items in the queue (frozen at this moment — do not recount mid-run).
-
-### 4. Implement each issue via `/tdd`
-
-For each issue file in queue order (N = 1, 2, … M), before invoking `/tdd`, emit the **progress line** (see `references/runner-output-formats.md`) substituting N, M, and the issue title (first `# Heading` line of the issue file).
-
-Invoke the sub-agent using the Agent tool with `subagent_type: general-purpose` — the only stock type with access to both the Skill tool (to load `/tdd`) and the Edit/Write tools (to write code). The issue's `## Acceptance criteria` replaces the interactive planning phase — pass it as the pre-approved plan so the agent skips confirmation and proceeds directly to implementation.
-
-Before constructing the prompt, use the Read tool to read all sibling issue files (`docs/issues/<slug>/[0-9]*.md` except the current issue) at their current state — this gives the sub-agent visibility into what is already resolved and what is still pending.
-
-Construct the prompt using the template in `references/tdd-prompt-template.md`, substituting all `<placeholder>` values at runtime. Pass the constructed prompt to the Agent tool. Wait for the agent to return before continuing.
-
-**On failure:** If the Agent call signals failure (throws, returns an error, or explicitly reports it could not complete the issue):
-
-1. Using the Edit tool, change the `**Status:** ready-for-agent` line to `**Status:** needs-info`. This prevents auto-select from picking up this Feature on subsequent loop iterations.
-
-2. Append the **failure note** (see `references/runner-output-formats.md`) to the issue file using the Edit tool, substituting `<slug>`.
-
-3. Stop the runner immediately. Do not execute any subsequent issues — they may depend on a foundation this issue was meant to lay.
-
-4. Report to the user: which issue failed, that the worktree is at `.claude/worktrees/<slug>` on branch `feature/afk/<slug>`, and that no subsequent issues were run.
-
-### 5. Mark each issue resolved
-
-After the Agent call for an issue returns **successfully**, update the issue file using the Edit tool: change the `**Status:** ready-for-agent` line to `**Status:** resolved`.
-
-### 6. Continue until queue is empty
-
-Repeat steps 4–5 for every issue in the queue. When the last issue is resolved, proceed to step 7.
-
-### 7. Open a pull request and clean up
-
-**Push the branch:**
-
-```
-git -C .claude/worktrees/<slug> push -u origin feature/afk/<slug>
-```
-
-**Derive the PR title** from the PRD's `title` frontmatter field (already read in step 1) and the slug:
-
-```
-feat(<slug>): <PRD title>
-```
-
-**List the resolved issues** — all `NN-*.md` files in `docs/issues/<slug>/` whose status is now `resolved` (every issue the runner just processed, in numerical order).
-
-**Open the PR** using the Bash tool. The exact `gh pr create` invocation with the heredoc-wrapped body lives in `references/runner-output-formats.md` under "PR body template" — use that form verbatim, substituting `<slug>`, `<PRD title>`, and the resolved-issue list before running it. Run from inside the worktree (`git -C .claude/worktrees/<slug>`) or pass `--repo` if needed. **Do not run the snippet without substitution** — angle-bracket placeholders are not valid shell.
-
-**Failure handling for steps 7a and 7b:** If `git push` or `gh pr create` fails (non-zero exit, network error, permission denied, branch protection, etc.):
-
-1. Stop immediately. Do **not** remove the worktree.
-2. Report to the user: which command failed, the error output, that the worktree is at `.claude/worktrees/<slug>` on branch `feature/afk/<slug>`, and that all issues are still `resolved` (the failure is post-implementation).
-3. Suggest re-running `git push` / `gh pr create` manually once the cause is resolved.
-
-**Remove the worktree** only after the PR is opened successfully (PR URL returned by `gh`):
-
-```
-git worktree remove .claude/worktrees/<slug>
-```
-
-Report the PR URL and the list of resolved issues to the user.
-
-## Supporting Documentation
-
-- **`references/runner-output-formats.md`** — verbatim strings for the progress line, dependency conflict error, unsatisfied dependency error, failure note, PR body template, and `LOOP_COMPLETE` signal.
-- **`references/tdd-prompt-template.md`** — the AFK prompt template passed to the Agent tool for each `/tdd` sub-agent invocation; substitute all `<placeholder>` values at runtime.
-- **`docs/agents/feature-runner.md`** — human-facing reference: lifecycle diagram, dependency-satisfaction matrix, ADR scope, historical cleanup convention. Maintained in parallel; consult for broader context, not for runtime instructions.
diff --git a/.claude/skills/implement-feature/references/runner-output-formats.md b/.claude/skills/implement-feature/references/runner-output-formats.md
deleted file mode 100644
index 23ae43fd..00000000
--- a/.claude/skills/implement-feature/references/runner-output-formats.md
+++ /dev/null
@@ -1,106 +0,0 @@
-# Runner output formats
-
-Verbatim strings emitted or embedded by the Feature Runner. `SKILL.md` references this file by name rather than repeating these strings inline.
-
----
-
-## Progress line
-
-Emitted to the user before each `/tdd` invocation.
-
-```
-Implementing issue N of M: <issue title>
-```
-
----
-
-## Missing blocker error
-
-Emitted when a `## Blocked by` reference points to a filename that does not exist in `docs/issues/<slug>/`. The runner halts before executing any issue.
-
-```
-Feature Runner error: missing blocker reference.
-  Issue NN-<A> lists "<reference>" in ## Blocked by, but that file does not exist in docs/issues/<slug>/.
-  Check for a typo, a rename, or a deleted issue. Resolve the reference manually before re-running.
-```
-
----
-
-## Dependency conflict error
-
-Emitted when a `## Blocked by` reference points to an issue with a higher numeric prefix than the issue being blocked. The runner halts before executing any issue.
-
-```
-Feature Runner error: dependency conflict detected.
-  Issue NN-<A> is blocked by NN-<B>, but NN-<B> has a higher number than NN-<A>.
-  This conflicts with the numerical ordering convention. Resolve the ordering manually before re-running.
-```
-
----
-
-## Unsatisfied dependency error
-
-Emitted when a `ready-for-agent` issue in the execution queue has a `## Blocked by` dependency whose status is `ready-for-human`. The runner halts before executing any issue.
-
-```
-Feature Runner error: unsatisfied dependency.
-  Issue NN-<A> is blocked by NN-<B>, but NN-<B> has status ready-for-human.
-  Complete issue NN-<B> manually (or update its status) before re-running /implement-feature <slug>.
-```
-
----
-
-## Failure note
-
-Appended to a failing issue file under `## Comments` when `/tdd` cannot complete an issue. The `**Status:**` line is also changed to `needs-info` (see SKILL.md step 4).
-
-```markdown
-## Comments
-
-> _This was generated by AI during triage._
-
-**Feature Runner failure** — `/tdd` could not complete this issue. Status has been set to `needs-info`.
-
-The worktree at `.claude/worktrees/<slug>` has been left in place for inspection. Once the issue is resolved manually, restore `**Status:** ready-for-agent` and re-run `/implement-feature <slug>` to resume. Alternatively, close or reject the issue if it should not be retried.
-```
-
----
-
-## PR body template
-
-The body passed to `gh pr create` after all issues in a feature are resolved.
-
-```markdown
-## Feature
-
-`docs/issues/<slug>/PRD.md`
-
-## Resolved issues
-
-- `docs/issues/<slug>/NN-<name>.md`
-- ...
-
-🤖 Implemented by the Feature Runner via `/implement-feature <slug>`
-```
-
-Pass via a bash heredoc:
-
-```
-gh pr create \
-  --base develop \
-  --title "feat(<slug>): <PRD title>" \
-  --body "$(cat <<'EOF'
-<substituted body content>
-EOF
-)"
-```
-
----
-
-## LOOP_COMPLETE signal
-
-Emitted on its own line when no qualifying feature exists during auto-selection. Must appear alone — nothing before or after it on the same line. This is the stop signal that `/loop /implement-feature` uses to terminate an overnight draining run.
-
-```
-LOOP_COMPLETE
-```
diff --git a/.claude/skills/implement-feature/references/tdd-prompt-template.md b/.claude/skills/implement-feature/references/tdd-prompt-template.md
deleted file mode 100644
index ff4a8d9d..00000000
--- a/.claude/skills/implement-feature/references/tdd-prompt-template.md
+++ /dev/null
@@ -1,27 +0,0 @@
-# /tdd AFK prompt template
-
-Passed to the Agent tool for each `/tdd` sub-agent invocation. All `<placeholder>` values are substituted at runtime before the prompt is sent.
-
-```
-You are running `/tdd` in AFK mode. The interactive planning phase is complete — do not ask for confirmation. Begin by invoking the `tdd` skill via the Skill tool to load its full procedural guidance (red→green→refactor, vertical-slice rule, deep-modules / interface-design / refactoring sub-references), then follow it using the acceptance criteria below as the pre-approved plan and proceed directly to the red→green→refactor loop.
-
-Working directory: .claude/worktrees/<slug>
-
---- ISSUE ---
-<full content of the current issue file>
-
---- PRD (parent context) ---
-<full content of the PRD file at docs/issues/{slug}/PRD.md>
-
---- SIBLING ISSUES ---
-<full content of each sibling issue file, separated by the filename as a header>
-
---- CONTEXT.md ---
-<full content of the scoped CONTEXT.md>
-
---- ADRs ---
-<full content of each scoped ADR file, separated by the filename as a header>
-
---- RECENT COMMITS (last 5) ---
-<output of git log --oneline -5>
-```
diff --git a/.claude/skills/new-plugin/SKILL.md b/.claude/skills/new-plugin/SKILL.md
index 6ac564c8..932e45d8 100644
--- a/.claude/skills/new-plugin/SKILL.md
+++ b/.claude/skills/new-plugin/SKILL.md
@@ -82,18 +82,12 @@ Only include if the plugin registers hooks.
 }
 ```
 
-**`ralph.yml`** — copy from `apps/claude-code/auto-format/ralph.yml`, replace the comment header plugin name.
-
-**`PROMPT.md`** — minimal stub: `# <Plugin Name> — Ralph Prompt\n\nDescribe the task here.`
-
 **`CHANGELOG.md`** — use the exact format from `apps/claude-code/auto-format/CHANGELOG.md` as template. Include only `## [Unreleased]` with empty Breaking/Added/Fixed subsections — no historical version entry yet.
 
 **`README.md`** — one-paragraph description of what the plugin does.
 
 **`CLAUDE.md`** — plugin-specific guidance (not a symlink at the plugin level; see `auto-format/CLAUDE.md` for structure).
 
-**`docs/plans/README.md`** — stub with: `# Spec Roadmap\n\nSpec files for this plugin go here.`
-
 **`scripts/`** and **`tests/`** — empty directories, only if the plugin has scripts or tests.
 
 ## Bundled resources
diff --git a/.claude/skills/new-plugin/references/package-json-template.md b/.claude/skills/new-plugin/references/package-json-template.md
index 64b77ba4..188fc03f 100644
--- a/.claude/skills/new-plugin/references/package-json-template.md
+++ b/.claude/skills/new-plugin/references/package-json-template.md
@@ -19,11 +19,9 @@ Use this as the starting shape for a new plugin's `package.json`. Copy `packageM
     "bump": "unic-bump",
     "sync-version": "unic-sync-version",
     "tag": "unic-tag",
-    "verify:changelog": "unic-verify-changelog",
-    "ralph": "ralph run -c ralph.yml -H builtin:code-assist"
+    "verify:changelog": "unic-verify-changelog"
   },
   "devDependencies": {
-    "@ralph-orchestrator/ralph-cli": "catalog:",
     "@types/node": "catalog:",
     "@unic/release-tools": "workspace:*",
     "@unic/tsconfig": "workspace:*",
@@ -51,11 +49,9 @@ Omit `test`, `typecheck`, and the `@types/node`/`@unic/tsconfig`/`typescript` de
     "bump": "unic-bump",
     "sync-version": "unic-sync-version",
     "tag": "unic-tag",
-    "verify:changelog": "unic-verify-changelog",
-    "ralph": "ralph run -c ralph.yml -H builtin:code-assist"
+    "verify:changelog": "unic-verify-changelog"
   },
   "devDependencies": {
-    "@ralph-orchestrator/ralph-cli": "catalog:",
     "@unic/release-tools": "workspace:*"
   }
 }
diff --git a/.claude/skills/verify-spec/SKILL.md b/.claude/skills/verify-spec/SKILL.md
index 9a5a4eea..e121f3ab 100644
--- a/.claude/skills/verify-spec/SKILL.md
+++ b/.claude/skills/verify-spec/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: verify-spec
-description: This skill should be used when the user asks to "verify spec 15", "check acceptance criteria for spec N", "is spec 14 ready to merge", "run verify-spec on 08", "check if spec 15 is done", or invokes /verify-spec with a spec number or filename. Locates the matching spec file and checks every acceptance criterion against the current codebase.
-argument-hint: '<spec-number-or-filename> (e.g. 15 or 15-release-tools-tests.md)'
+description: This skill should be used when the user asks to "verify spec auto-format-config", "check acceptance criteria for pr-review-rereview", "is pr-review-ado-fetcher-reliability ready to merge", "run verify-spec on unic-archon-dlc", or invokes /verify-spec with an issue slug or partial name. Locates the matching PRD file in docs/issues/ and checks every acceptance criterion against the current codebase.
+argument-hint: '<issue-slug-or-partial> (e.g. auto-format-config or pr-review-rereview)'
 user-invocable: true
 ---
 
@@ -13,24 +13,26 @@ Verify that all acceptance criteria in a spec file are satisfied by the current
 
 `$ARGUMENTS` is one of:
 
-- A bare spec number: `15`
-- A full filename: `15-release-tools-tests.md`
-- A partial name: `release-tools`
+- A full issue slug: `auto-format-config`
+- A partial slug: `auto-format` (substring match)
+- A done sub-task filename: `01-end-to-end-notice-pipeline.md`
 
 ## Locating the spec file
 
+Spec files live in `docs/issues/` as `<slug>/PRD.md`. Completed vertical-slice sub-tasks live under `docs/issues/<slug>/done/NN-title.md`.
+
 Search in order:
 
-1. `docs/plans/` at the repo root
-2. `apps/claude-code/<any-plugin>/docs/plans/`
+1. `docs/issues/` at the repo root — match `<slug>/PRD.md`
+2. `apps/claude-code/<any-plugin>/docs/issues/` — same structure
 
 Matching rules:
 
-- **Bare number** (`15`, `1`): interpret as an exact numeric prefix. `15` matches `15-*` only; `1` matches `01-*` only (left-zero-pad to two digits). It does NOT substring-match `10-`, `11-`, etc.
-- **Partial name** (`release-tools`): substring match against the filename.
-- **Full filename** (`15-release-tools-tests.md`): exact match.
+- **Full slug** (`auto-format-config`): look for `docs/issues/auto-format-config/PRD.md`.
+- **Partial slug** (`auto-format`): substring match against slug directory names; list all matches and ask which to use if more than one.
+- **Done sub-task filename** (`01-end-to-end-notice-pipeline.md` or bare `01`): search `done/` sub-directories under matching slugs.
 
-If more than one file matches, list the candidates and ask which to use. If no file matches, report that clearly and stop.
+If no file matches, report that clearly and stop.
 
 ## Spec file format
 
diff --git a/.claude/skills/verify-spec/references/spec-format.md b/.claude/skills/verify-spec/references/spec-format.md
index db317bfe..011e9beb 100644
--- a/.claude/skills/verify-spec/references/spec-format.md
+++ b/.claude/skills/verify-spec/references/spec-format.md
@@ -1,15 +1,24 @@
 # Spec File Format Reference
 
-All spec files under `docs/plans/` (root or plugin-level) follow this template.
+Spec files live under `docs/issues/` (root or plugin-level) and follow this structure.
 
-## File naming
+## File layout
+
+### Top-level PRD (primary intake format)
+
+```
+docs/issues/<slug>/PRD.md
+```
+
+`<slug>` is a kebab-case identifier matching the GitHub Issue title. Example: `auto-format-config/PRD.md`.
+
+### Done sub-tasks (vertical slices completed within a PRD)
 
 ```
-NN-kebab-case-title.md
+docs/issues/<slug>/done/NN-kebab-case-title.md
 ```
 
-`NN` is a zero-padded two-digit or three-digit number (e.g. `00`, `09`, `15`). The title slug is
-kebab-case and matches the spec title. Example: `15-release-tools-tests.md`.
+`NN` is a zero-padded two-digit number. Example: `done/01-end-to-end-notice-pipeline.md`.
 
 ## Full template
 
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 5f3f3ed0..452d883a 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -52,7 +52,8 @@ jobs:
       needs.changes.outputs.auto-format == 'true' ||
       needs.changes.outputs.unic-confluence == 'true' ||
       needs.changes.outputs.release-tools == 'true' ||
-      needs.changes.outputs.unic-archon-dlc == 'true'
+      needs.changes.outputs.unic-archon-dlc == 'true' ||
+      needs.changes.outputs.pr-review == 'true'
     strategy:
       fail-fast: false
       matrix:
@@ -67,6 +68,8 @@ jobs:
             changed: ${{ needs.changes.outputs.release-tools }}
           - name: unic-archon-dlc
             changed: ${{ needs.changes.outputs.unic-archon-dlc }}
+          - name: pr-review
+            changed: ${{ needs.changes.outputs.pr-review }}
         exclude:
           - package:
               changed: 'false'
@@ -84,6 +87,14 @@ jobs:
         run: pnpm install --frozen-lockfile
       - if: runner.os == 'Windows'
         run: pnpm install --frozen-lockfile --ignore-scripts
+      # Install Azure CLI + azure-devops extension only on Linux/Node 24 so the
+      # pr-review ADO CLI smoke test (tests/ado-cli-smoke.test.mjs) actually
+      # invokes `--help` against each inventory entry on at least one matrix
+      # cell. Other cells fall through the t.skip path cleanly.
+      - if: matrix.package.name == 'pr-review' && matrix.os == 'ubuntu-latest' && matrix.node == '24'
+        run: |
+          curl -sL https://aka.ms/InstallAzureCLIDeb | sudo bash
+          az extension add --name azure-devops
       - run: pnpm --filter ${{ matrix.package.name }} --if-present test
       - run: pnpm --filter ${{ matrix.package.name }} --if-present typecheck
       - if: github.event_name == 'pull_request'
diff --git a/.idea/conventionalCommit.xml b/.idea/conventionalCommit.xml
new file mode 100644
index 00000000..a09c8599
--- /dev/null
+++ b/.idea/conventionalCommit.xml
@@ -0,0 +1,10 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="general">
+    <commitScopes>
+      <entry providerId="37415b03-9388-4c55-b949-8c4526f6934d" order="2" />
+      <entry providerId="e9ce9acf-f4a6-4b36-b43c-531169556c29" order="2" />
+      <entry providerId="e9d4e8de-79a0-48b8-b1ba-b4161e2572c0" order="1" />
+    </commitScopes>
+  </component>
+</project>
\ No newline at end of file
diff --git a/.idea/vcs.xml b/.idea/vcs.xml
index aae8b7c8..4c6280eb 100644
--- a/.idea/vcs.xml
+++ b/.idea/vcs.xml
@@ -9,4 +9,4 @@
   <component name="VcsDirectoryMappings">
     <mapping directory="$PROJECT_DIR$" vcs="Git" />
   </component>
-</project>
+</project>
\ No newline at end of file
diff --git a/AGENTS.md b/AGENTS.md
index 80de4255..55b7278b 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -21,16 +21,20 @@ packages/
 ├── tsconfig/                 # @unic/tsconfig
 └── release-tools/            # @unic/release-tools (bump / sync-version / tag / verify-changelog)
 docs/
-└── plans/                    # Monorepo-level Spec Runner roadmap (specs 00–14)
+├── adr/                      # Architectural Decision Records
+├── agents/                   # Agent skill documentation
+├── inbox/                    # Retired idea-capture notes (historical)
+├── issues/                   # Grilled and scoped feature issues
+├── plans/                    # Retired spec files (historical)
+├── process/                  # Process and workflow guides
+└── research/                 # Research notes and explorations
 ```
 
-Each plugin under `apps/<agent>/` also has its own `docs/plans/` for plugin-specific future work.
-
 ## Navigation
 
 - Plugin manifests: `apps/<agent>/<plugin>/.claude-plugin/plugin.json` and `marketplace.json`
 - Shared release scripts: `packages/release-tools/scripts/`
-- Monorepo roadmap: `docs/plans/`
+- Architectural decisions: `docs/adr/`
 - Process templates: `docs/process/`
 
 ## Commands
@@ -42,12 +46,10 @@ pnpm format                             # Biome + Prettier fix (whole tree)
 pnpm ci:check                           # same as check, non-interactive (for CI)
 pnpm test                               # run tests across all packages
 pnpm typecheck                          # type-check across all packages
-pnpm ralph                              # run the monorepo Spec Runner loop (specs 00–14)
 
-# Per-plugin operations (after spec 03 sets up release-tools)
+# Per-plugin operations
 pnpm --filter <name> bump patch         # bump plugin version
 pnpm --filter <name> verify:changelog   # check changelog
-pnpm --filter <name> ralph              # run that plugin's own Spec Runner loop
 ```
 
 ## Tech stack
@@ -79,7 +81,7 @@ Tag scheme: `<plugin-name>@<version>` (e.g. `auto-format@0.5.5`).
 
 ## Conventional commits
 
-Use package scope: `feat(auto-format): …`, `fix(pr-review): …`, `chore(release-tools): …`, `chore(spec-NN): …`.
+Use package scope: `feat(auto-format): …`, `fix(pr-review): …`, `chore(release-tools): …`, `chore(unic-archon-dlc): …`.
 
 ## Git branching (Gitflow)
 
@@ -110,40 +112,25 @@ To ship a new plugin version:
 | Push to `develop` | ✓           | ✓ (changed packages) | —                  |
 | Push to `main`    | ✓           | ✓ (changed packages) | —                  |
 
-## Spec-driven development
-
-All work starts with a spec file under `docs/plans/`. Specs follow this format:
-
-```
-# NN. Title
-**Priority:** / **Effort:** / **Version impact:** / **Depends on:** / **Touches:**
-## Context
-## Current behaviour
-## Target behaviour
-## Affected files
-## Implementation steps
-## Verification
-## Acceptance criteria
-## Out of scope
-```
+## Feature-driven development
 
-`pnpm ralph` runs the Spec Runner (currently `ralph-orchestrator`), which implements specs one at a time in a loop.
+New work enters through the issue tracker as Features. Use `/tdd` to implement individual issues, or `unic-dlc-build` (via `unic-archon-dlc`) once the harness is set up in this repo.
 
 ## Do not add
 
 - External runtime deps to plugins unless truly essential (`auto-format` has zero; that's the bar)
 - Turborepo or other build orchestrators — plain pnpm workspaces is the current choice
-- Features not described in an open spec file — open a spec first
+- Features not tracked in the issue tracker — open a Feature first
 
 ## LICENSE files
 
-**Never create, copy, or delete `LICENSE` files.** The maintainer manages these manually in every package and plugin directory. If a spec step or acceptance criterion requires a `LICENSE` file to exist, warn the maintainer to add it themselves before continuing.
+**Never create, copy, or delete `LICENSE` files.** The maintainer manages these manually in every package and plugin directory. If an acceptance criterion requires a `LICENSE` file to exist, warn the maintainer to add it themselves before continuing.
 
-## Agent skills
+## Skill summary
 
 ### Issue tracker
 
-Issues live as local markdown files under `docs/issues/`. See `docs/agents/issue-tracker.md`.
+Issues live in GitHub Issues at `unic/unic-agents-plugins` (planned migration to Azure DevOps). See `docs/agents/issue-tracker.md`.
 
 ### Triage labels
 
@@ -153,6 +140,14 @@ Issues live as local markdown files under `docs/issues/`. See `docs/agents/issue
 
 Multi-context repo: Per-plugin `CONTEXT.md` files live under `apps/claude-code/<plugin>/`. Root `CONTEXT-MAP.md` at repo root. See `docs/agents/domain.md`.
 
-### Inbox
+## Agent skills
+
+<!-- unic-archon-dlc:begin -->
+<!-- AUTO-GENERATED by unic-archon-dlc — do not edit; re-run `/unic-archon-dlc:setup` to regenerate. -->
 
-Raw ideas, bugs, and stray thoughts that haven't been grilled or scoped yet live in `docs/inbox/`. Use `/inbox <one-liner>` to capture without interrupting the current flow. Items graduate to `docs/issues/<slug>/` after grilling. See `docs/inbox/README.md`.
+- [issue-tracker.md](docs/agents/issue-tracker.md) — issue tracker backend, CLI, create/update conventions
+- [labels.md](docs/agents/labels.md) — three-tier label taxonomy: state, type, priority
+- [branching.md](docs/agents/branching.md) — branching strategy, branch names, PR targets
+- [domain.md](docs/agents/domain.md) — single-context vs multi-context, CONTEXT.md and ADR locations
+- [workflow.md](docs/agents/workflow.md) — seven workflow phases, artifact outputs, docs/workflow/ paths
+<!-- unic-archon-dlc:end -->
diff --git a/CONTEXT-MAP.md b/CONTEXT-MAP.md
index c6ebe22d..3c28dbaa 100644
--- a/CONTEXT-MAP.md
+++ b/CONTEXT-MAP.md
@@ -2,7 +2,7 @@
 
 ## Shared vocabulary
 
-- [Monorepo](./CONTEXT.md) — Plugin, Workspace Package, Release, Spec, Spec Runner, Consumer
+- [Monorepo](./CONTEXT.md) — Plugin, Workspace Package, Release, Feature, Feature Runner, Consumer
 
 ## Plugin contexts
 
@@ -18,3 +18,4 @@
 - **unic-confluence** can be installed as a git dependency for use outside Claude Code.
 - **pr-review** has a soft dependency on the `pr-review-toolkit` plugin from `anthropics/claude-plugins-official`
 - **unic-archon-dlc** requires the Archon workflow engine (version ≥ 0.10) in the target project; it has no runtime dependencies on any other plugin in this repo
+- Architectural decisions are split by scope: monorepo-wide decisions live in root `docs/adr/`; decisions scoped to a single context live in that context's own `docs/adr/`.
diff --git a/CONTEXT.md b/CONTEXT.md
index d66161eb..ca762a57 100644
--- a/CONTEXT.md
+++ b/CONTEXT.md
@@ -20,20 +20,12 @@ _Avoid_: package (too generic), library, module
 A versioned, tagged snapshot of a single Plugin published to `main`. A per-Plugin event — there is no monorepo-wide release.
 _Avoid_: deploy, publish, version bump
 
-**Spec**:
-A structured markdown file under `docs/plans/` that defines a unit of work with context, target behaviour, and acceptance criteria. The atomic input to the Spec Runner.
-_Avoid_: ticket, task, issue, story
-
-**Spec Runner**:
-The agent automation that reads Specs and implements them one at a time. Currently backed by `ralph-orchestrator` but the concept is tool-agnostic.
-_Avoid_: Ralph, Ralph Orchestrator (tool-specific, not domain terms)
-
 **Feature**:
 A self-contained unit of work tracked as a directory under `docs/issues/<slug>/`, containing a PRD and numbered implementation issues. The atomic input to the Feature Runner.
 _Avoid_: ticket, epic, story
 
 **Feature Runner**:
-The skill that implements a Feature's issues end-to-end in one worktree, branch, and pull request. Parallel concept to Spec Runner, but driven by the issue tracker rather than `docs/plans/`.
+The skill that implements a Feature's issues end-to-end in one worktree, branch, and pull request. Backed by `unic-dlc-build`.
 _Avoid_: issue runner, queue runner
 
 **Consumer**:
@@ -45,8 +37,7 @@ _Avoid_: client, user repo, target repo, host repo
 - A **Plugin** belongs to one agent ecosystem (e.g., Claude Code) and has zero or more **Releases**
 - A **Claude Code Plugin** is a **Plugin** — the inverse is not always true
 - A **Workspace Package** supports **Plugin** development but is not itself a **Plugin**
-- A **Spec** drives exactly one **Spec Runner** iteration
-- A **Feature** drives one **Feature Runner** execution — a Feature is to the Feature Runner what a Spec is to the Spec Runner
+- A **Feature** drives one **Feature Runner** execution
 - A **Consumer** installs one or more **Plugins**
 
 ## Example dialogue
@@ -56,6 +47,3 @@ _Avoid_: client, user repo, target repo, host repo
 
 > **Dev:** "A Consumer reported the auto-format Plugin failing on Windows."
 > **Domain expert:** "That's a cross-platform gap. Plugins must use Node.js APIs — no shell commands or POSIX paths."
-
-> **Dev:** "Can I use Ralph to implement this Spec?"
-> **Domain expert:** "You mean the Spec Runner — Ralph is just the current tool backing it."
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index ed3f1d09..87d6331e 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -1,8 +1,8 @@
 # Contributing
 
-This monorepo uses a spec-driven development workflow. For the full lifecycle — from idea capture through grilling, PRD, issues, execution, and QA — see **[`docs/process/development-workflow.md`](docs/process/development-workflow.md)**.
+This monorepo uses a Feature-driven development workflow. For the full lifecycle — from idea capture through grilling, PRD, issues, execution, and QA — see **[`docs/process/development-workflow.md`](docs/process/development-workflow.md)**.
 
-New features and fixes are described in self-contained spec files under `docs/plans/`. Implementation is automated with the **Spec Runner** (`pnpm ralph`, currently backed by `ralph-orchestrator`) or done by hand following the same steps.
+New features and fixes are tracked as Features in the issue tracker under `docs/issues/<slug>/`. Implementation is done manually using `/tdd` for individual issues; the **Feature Runner** (`unic-dlc-build`, shipped by `unic-archon-dlc`) is the long-term AFK runner once it is wired into this repo (see [ADR-0009](apps/claude-code/unic-archon-dlc/docs/adr/0009-retire-ralph-adopt-archon-runner.md)).
 
 ## Cross-cutting standards
 
@@ -48,7 +48,7 @@ Enforced by Biome (`pnpm ci:check`) for `.mjs`/`.js`/`.ts`/`.json`/`.css` and Pr
 
 - Zero external runtime deps unless essential (`auto-format` ships zero; `unic-confluence` has `marked` — that's the bar).
 - Zero-config from the user's perspective: no configuration files users must create beyond credentials.
-- Every plugin uses the spec-driven workflow: `docs/plans/` + `ralph.yml` + `PROMPT.md` per the template in `docs/process/`.
+- New plugin work enters through the issue tracker as Features — open a Feature before starting implementation.
 
 ### License
 
@@ -56,87 +56,29 @@ LGPL-3.0-or-later for all packages in this monorepo.
 
 ## Prerequisites
 
-| Tool            | Version                                         | How to get it                                                                    |
-| --------------- | ----------------------------------------------- | -------------------------------------------------------------------------------- |
-| Node.js         | ≥ 22 (see `.nvmrc` for the recommended version) | [nodejs.org](https://nodejs.org)                                                 |
-| pnpm            | ≥ 10                                            | `npm install -g pnpm`                                                            |
-| Claude Code CLI | latest                                          | [claude.ai/code](https://claude.ai/code) — required as the Spec Runner's backend |
+| Tool            | Version                                         | How to get it                            |
+| --------------- | ----------------------------------------------- | ---------------------------------------- |
+| Node.js         | ≥ 22 (see `.nvmrc` for the recommended version) | [nodejs.org](https://nodejs.org)         |
+| pnpm            | ≥ 10                                            | `npm install -g pnpm`                    |
+| Claude Code CLI | latest                                          | [claude.ai/code](https://claude.ai/code) |
 
-Everything else (ralph-orchestrator, Biome, Prettier, TypeScript) is a workspace devDependency and installs with:
+Everything else (Biome, Prettier, TypeScript) is a workspace devDependency and installs with:
 
 ```sh
 pnpm install
 ```
 
-**ralph-loop** is a Claude Code plugin and must be installed once globally:
+## Starting new work
 
-```sh
-claude plugins install anthropics/claude-plugins-official/plugins/ralph-loop
-```
-
-## Writing a spec
-
-All work starts with a spec file. See the existing specs under `docs/plans/` for examples.
-
-### 1. Pick a number
-
-Check `docs/plans/README.md` for the current highest spec number. Create `docs/plans/NN-short-slug.md`.
-
-### 2. Required metadata
-
-```markdown
-# NN. Title
-
-**Priority:** P0 | P1 | P2
-**Effort:** XS | S | M | L
-**Version impact:** none | patch (plugin: <name>) | minor (plugin: <name>) | major (plugin: <name>)
-**Depends on:** (spec numbers, or "none")
-**Touches:** (comma-separated files/dirs)
-```
-
-### 3. Required sections
-
-| Section                   | What to write                            |
-| ------------------------- | ---------------------------------------- |
-| `## Context`              | Why this change is needed                |
-| `## Current behaviour`    | Exact code/behaviour _before_ the change |
-| `## Target behaviour`     | What it should look like _after_         |
-| `## Affected files`       | Table: path → Create / Modify / Delete   |
-| `## Implementation steps` | Numbered steps with before → after diffs |
-| `## Verification`         | Shell commands + expected output         |
-| `## Acceptance criteria`  | Checkbox list                            |
-| `## Out of scope`         | Explicit list of things NOT to change    |
-
-Good specs are **self-contained** — the Spec Runner has no memory of prior runs. Include actual code snapshots.
+All work enters through the issue tracker as a Feature. The recommended flow:
 
-### 4. Register the spec
+1. Capture the idea — open a GitHub Issue directly, or run `/triage` to walk it through the state machine.
+2. Grill the design with `/grill-me` or `/grill-with-docs` until the problem and solution are clear.
+3. Create a PRD and issues with `/to-prd` → `/to-issues`.
+4. Implement each issue manually with `/tdd`. (AFK execution via `unic-dlc-build` will replace this step once the harness is wired in — see [ADR-0009](apps/claude-code/unic-archon-dlc/docs/adr/0009-retire-ralph-adopt-archon-runner.md) and [ADR-0010](apps/claude-code/unic-archon-dlc/docs/adr/0010-retire-implement-feature-skill.md).)
+5. Open a PR targeting `develop`.
 
-Add a row to the execution order table in `docs/plans/README.md`.
-
-## Running the Spec Runner
-
-```sh
-pnpm ralph
-```
-
-The Spec Runner reads `ralph.yml`, which points to `PROMPT.md`. Each iteration implements one spec, commits, and stops. Run `pnpm ralph` again for the next spec.
-
-- **Stop early**: `Ctrl+C`
-- **Resume**: `pnpm ralph` again — it finds the first unfinished spec
-- **Paused for review**: a `## Questions` section in a spec means the Spec Runner emitted `LOOP_COMPLETE` — answer the question, remove the section, re-run
-
-## Running manually (without the Spec Runner)
-
-Follow the steps in `PROMPT.md` yourself, replacing the Spec Runner's role.
-
-## Running per-plugin Spec Runner loops
-
-Each plugin also has its own `ralph.yml` + `PROMPT.md` for plugin-specific development. Run from inside the plugin directory:
-
-```sh
-cd apps/claude-code/pr-review
-pnpm ralph
-```
+See [`docs/process/development-workflow.md`](docs/process/development-workflow.md) for the full 8-phase lifecycle.
 
 ## Verification commands
 
diff --git a/PROMPT.md b/PROMPT.md
deleted file mode 100644
index 735fa6dc..00000000
--- a/PROMPT.md
+++ /dev/null
@@ -1,106 +0,0 @@
-# Ralph Orchestrator — unic-agents-plugins Roadmap
-
-You are implementing the roadmap for the `unic-agents-plugins` monorepo, one spec at a time.
-
-## Step 1 — Determine what's next
-
-Check the execution order in `docs/plans/README.md`. Then scan spec files in order (00 → 13) and find the first file that does NOT contain the string `**Status: done`.
-
-If **all** specs contain `**Status: done`, output exactly:
-
-```
-<promise>LOOP_COMPLETE</promise>
-```
-
-…and stop. Do nothing else.
-
-## Step 2 — Read the spec completely
-
-Read the entire spec file before writing any code. Pay special attention to:
-
-- **Current behaviour** — verify the code actually works this way before starting; if it doesn't, add a `## Deviations` section at the bottom of the spec file documenting the discrepancy
-- **Out of scope** — list of things you must NOT change in this iteration
-- **Depends on** — if a dependency spec is not yet marked done, skip this spec and move to the next one that can be done
-
-## Step 3 — Implement
-
-Check the spec's `**Version impact:**` line to choose the implementation approach:
-
-**`none` (workspace/infrastructure spec)** — implement directly. Follow the "Implementation steps" exactly. If a step's "before" snapshot doesn't match the current file, consult the "Deviations" section (if you wrote one) or document the discrepancy and adapt minimally.
-
-**`patch` / `minor` / `major` (plugin spec)** — use `/tdd` to drive implementation. Treat the spec's "Implementation steps" as guidance (key files to touch, rough order), not a recipe. The spec's **Acceptance criteria** are the target; the red-green-refactor cycle drives the path there.
-
-Ground rules (from `docs/plans/README.md`):
-
-- Use `pnpm` for all package operations (not `npm`)
-- Tabs for indentation in `.mjs`/`.js`/`.ts` files, spaces for `.json`/`.yml`/`.yaml` (per `.editorconfig`)
-- Conventional commits: `feat(scope): description`, `fix(scope): description`, `chore(scope): description`
-- Cross-platform: use Node.js APIs (`node:path`, `node:fs`, `node:os`) instead of shell commands; no bash/sh assumptions
-- If something can't be followed as written: document it in `## Deviations`, don't silently deviate
-
-**Supporting skill:** `/diagnose` — if a failure persists and the root cause appears to be outside the spec's scope (an existing module, tooling, or dependency the spec doesn't touch), use `/diagnose` to isolate it before continuing.
-
-## Step 4 — Verify
-
-Run the exact commands in the spec's **Verification** section. Fix any failures before proceeding.
-
-Check every item in **Acceptance criteria**. If any item fails, fix it.
-
-Then run the always-on repo hygiene checks regardless of spec type:
-
-```sh
-pnpm check   # Biome + Prettier — fix any failures before proceeding
-```
-
-## Step 4.5 — Version bump (conditional)
-
-Check the spec's `**Version impact:**` line:
-
-**`none` (workspace/infrastructure spec)** — skip this step entirely. No CHANGELOG update. Proceed to Step 5.
-
-**`patch` / `minor` / `major` (plugin: `<name>`)** — bump the specified plugin:
-
-1. Append one bullet under the matching subsection of `## [Unreleased]` in `apps/claude-code/<name>/CHANGELOG.md`:
-
-   - `### Breaking` — CLI flag change, exit-code change, on-disk schema change
-   - `### Added` — new flag, subcommand, or user-visible feature
-   - `### Fixed` — bug fix, refactor, docs, internal tooling
-
-2. Run: `pnpm --filter <name> bump <patch|minor|major>`
-
-3. Run: `pnpm --filter <name> verify:changelog`
-
-## Step 5 — Mark done and commit
-
-1. Add this line immediately after the spec's `# NN. Title` heading (as the second line):
-
-```
-**Status: done — YYYY-MM-DD**
-```
-
-Replace `YYYY-MM-DD` with today's date.
-
-2. Stage and commit all changes using the appropriate prefix:
-
-- **Workspace specs** (version impact: none): `git commit -m "chore(spec-NN): <short description>"`
-- **Plugin specs** (version impact: patch/minor/major): `git commit -m "feat(spec-NN): <short description> (vX.Y.Z)"`
-- **Migration specs** (05–07): the `git merge --allow-unrelated-histories` creates a first commit; any post-merge fixups go in a second commit: `git commit -m "feat(spec-NN): finalize <plugin> migration"`
-
-3. **Do not push.** Commits only.
-
-## Step 6 — Stop for this iteration
-
-Output a brief summary of what was implemented and committed. Then stop — Ralph will feed this prompt again for the next spec.
-
----
-
-## Important constraints
-
-- Implement **one spec per iteration**. Do not implement multiple specs in a single run.
-- If a spec requires human judgment (e.g. choosing between two design approaches not covered by the spec), stop, document the question in a `## Questions` section at the bottom of the spec file, and output:
-  ```
-  <promise>LOOP_COMPLETE</promise>
-  ```
-  This pauses the loop for human review.
-- Do not create any files outside of what the spec describes.
-- Do not modify any spec files except to add `**Status: done**`, `## Deviations`, or `## Questions` sections.
diff --git a/apps/claude-code/auto-format/AGENTS.md b/apps/claude-code/auto-format/AGENTS.md
new file mode 100644
index 00000000..e6e16226
--- /dev/null
+++ b/apps/claude-code/auto-format/AGENTS.md
@@ -0,0 +1,74 @@
+# AGENTS.md — auto-format
+
+Guidance for any AI agent working inside this Plugin directory. `CLAUDE.md` in this directory is a symlink to this file.
+
+## What this Plugin is
+
+`auto-format` is a Claude Code Plugin in the [`unic-agents-plugins`](../../../AGENTS.md) monorepo. It registers a `PostToolUse` hook that auto-formats and lints files after `Write` / `Edit` / `MultiEdit` / `NotebookEdit` events. The hook script (`scripts/format-hook.mjs`) uses only Node built-ins and shells out to whichever Formatter the Consumer project has configured locally. See [`CONTEXT.md`](CONTEXT.md) for the domain vocabulary (Format Hook, Formatter, Consumer).
+
+## Where to start
+
+Root docs (monorepo-wide conventions live here — pnpm scripts, Gitflow, SemVer, Conventional Commits, code conventions, LICENSE policy, cross-platform requirement):
+
+- [Root AGENTS.md](../../../AGENTS.md) — source of truth for cross-cutting rules
+- [Root CONTEXT.md](../../../CONTEXT.md) — monorepo-wide vocabulary (Plugin, Workspace Package, Release, Feature, Consumer)
+- [Root CONTEXT-MAP.md](../../../CONTEXT-MAP.md) — index of all bounded contexts in the repo
+- [Root docs/adr/](../../../docs/adr/) — monorepo-wide architecture decisions
+- [Root docs/process/](../../../docs/process/) — process and workflow guides
+
+This Plugin's own decisions:
+
+- [Plugin docs/adr/](docs/adr/) — Plugin-specific architecture decisions
+
+## Commands
+
+Plugin-specific pnpm scripts (run from this directory or with `pnpm --filter auto-format <script>` from the repo root):
+
+```sh
+pnpm test                 # run node:test smoke tests
+pnpm typecheck            # tsc --checkJs over the hook source
+pnpm bump <patch|minor|major>   # bump plugin.json version + promote CHANGELOG
+pnpm sync-version         # mirror plugin.json version into marketplace.json + package.json
+pnpm tag                  # create the auto-format@<version> git tag locally
+pnpm verify:changelog     # check CHANGELOG entry for the current version
+```
+
+Monorepo-wide commands (`pnpm install`, `pnpm check`, `pnpm format`, `pnpm ci:check`) are documented in the [root AGENTS.md](../../../AGENTS.md).
+
+## Layout
+
+```tree
+.claude-plugin/           # Plugin manifest (plugin.json) and marketplace listing
+hooks/                    # Hook registration — wires format-hook.mjs to PostToolUse events
+scripts/                  # format-hook.mjs (the hook entrypoint) and helpers
+tests/                    # node:test smoke tests for the hook script
+docs/                     # Plugin-specific documentation
+  adr/                    # Plugin Architecture Decision Records
+```
+
+## Plugin doctrines
+
+Load-bearing invariants. These either originate in a Plugin ADR or are policy decisions that are not obvious from the code.
+
+- **The Format Hook always exits zero.** A failing Formatter must never block Claude Code's tool loop. Diagnostics are surfaced via stderr/log files; exit codes are always `0`. See [ADR-0001](docs/adr/0001-hook-always-exits-zero.md).
+- **Zero runtime dependencies.** The hook script imports only `node:*` builtins. No bundled formatter, linter, or runtime npm package. See [ADR-0002](docs/adr/0002-zero-runtime-dependencies.md).
+- **Consumer owns the Formatters.** The Plugin discovers and invokes whichever formatter binary the Consumer project provides (Prettier, Biome, gofmt, …). It never bundles or installs Formatters. See [ADR-0003](docs/adr/0003-consumer-owns-formatters.md).
+- **Per-project config merge.** Format Hook configuration is merged from user-global → project → file-pattern overrides, with project-level overrides winning. See [ADR-0004](docs/adr/0004-per-project-config-merge.md).
+- **POSIX path normalisation on Windows.** All paths passed to spawned Formatters are POSIX-normalised so cross-platform Formatters behave consistently. See [ADR-0006](docs/adr/0006-posix-path-normalization-windows.md).
+- **`spawnSync` with a timeout and kill signal.** Every Formatter invocation has an upper bound; a runaway Formatter cannot stall Claude Code. See [ADR-0005](docs/adr/0005-spawnsync-timeout-kill-signal.md).
+
+## External dependencies
+
+None. The hook script is pure `node:*`. Formatters themselves are Consumer-provided.
+
+## Do not add
+
+- **Bundled Formatters** (Prettier, ESLint, Biome, …) inside this Plugin — Consumers bring their own. See [ADR-0003](docs/adr/0003-consumer-owns-formatters.md).
+- **MCP servers, agents, or sub-agents.** The Format Hook is a deterministic one-shot — no autonomy is warranted.
+- **Bash hooks or pre-commit hooks.** Those belong in Consumer repos, not in this Plugin.
+- **Blocking Format Hook behaviour.** Any change that lets the hook exit non-zero violates [ADR-0001](docs/adr/0001-hook-always-exits-zero.md).
+- **External npm runtime dependencies.** The zero-deps bar is set by [ADR-0002](docs/adr/0002-zero-runtime-dependencies.md) and is the standard for the monorepo.
+
+## Plugin ADRs
+
+Plugin-specific architecture decisions live in [docs/adr/](docs/adr/).
diff --git a/apps/claude-code/auto-format/CLAUDE.md b/apps/claude-code/auto-format/CLAUDE.md
deleted file mode 100644
index 20e5c2a6..00000000
--- a/apps/claude-code/auto-format/CLAUDE.md
+++ /dev/null
@@ -1,47 +0,0 @@
-# CLAUDE.md
-
-This file provides guidance to Claude Code when working in this repository.
-
-## Project Overview
-
-`unic-claude-code-format` is a Claude Code plugin that auto-formats and lints files after Write/Edit/MultiEdit/NotebookEdit events. The hook script (`scripts/format-hook.mjs`) uses only Node built-ins — no external deps.
-
-## Commands
-
-```sh
-pnpm install              # Install dev dependencies
-pnpm test                 # Run smoke tests (node:test)
-pnpm typecheck            # Type-check plugin source (tsc --checkJs, dev-only)
-pnpm bump patch           # Bump patch version + promote CHANGELOG
-pnpm bump minor           # Bump minor version + promote CHANGELOG
-pnpm bump major           # Bump major version + promote CHANGELOG
-pnpm sync-version         # Propagate plugin.json version to marketplace.json + package.json
-pnpm tag                  # Create git tag v<version> (run git push --follow-tags after)
-pnpm verify:changelog     # Check CHANGELOG structure
-```
-
-## Tech Stack
-
-- **Runtime**: Node.js >=22. `.nvmrc` (currently `24.15.0`) is the recommended local version; CI exercises the matrix on Node 22 and 24.
-- **Package manager**: pnpm (workspace mode, catalog pinning).
-- **Module system**: ESM (`"type": "module"`).
-- **Test runner**: `node:test` built-in. No external framework.
-- **Hook script deps**: zero external. Only `node:child_process`, `node:fs`, `node:path`.
-
-## Code Conventions
-
-- Tabs for indentation in `.js`/`.mjs` files; 2-space for `.json`, `.yml`, `.yaml` (per `.editorconfig`).
-- Conventional commits: `feat(scope): description`, `fix(scope): description`, `chore(scope): description`.
-- SemVer: `pnpm bump <patch|minor|major>`. Never hand-edit `.claude-plugin/marketplace.json`.
-- No `^` or `~` version ranges. Exact pins only (catalog for devDeps).
-
-## Scope Guard — Do Not Add
-
-- MCP servers, skills, agents, or other Claude Code plugin features not described in an existing spec.
-- External npm runtime dependencies to the hook script.
-- Prettier or ESLint bundled inside the plugin — consumers bring their own.
-- Support for tools outside the Claude Code hook contract (e.g. bash hooks, pre-commit hooks — those belong in consumer repos).
-
-## Ralph Orchestrator
-
-This repo uses ralph-orchestrator for iterative development. See `docs/plans/README.md` for the spec roadmap and `PROMPT.md` for the orchestrator loop prompt.
diff --git a/apps/claude-code/auto-format/CLAUDE.md b/apps/claude-code/auto-format/CLAUDE.md
new file mode 120000
index 00000000..47dc3e3d
--- /dev/null
+++ b/apps/claude-code/auto-format/CLAUDE.md
@@ -0,0 +1 @@
+AGENTS.md
\ No newline at end of file
diff --git a/apps/claude-code/auto-format/PROMPT.md b/apps/claude-code/auto-format/PROMPT.md
deleted file mode 100644
index 6f2119a1..00000000
--- a/apps/claude-code/auto-format/PROMPT.md
+++ /dev/null
@@ -1,117 +0,0 @@
-# Ralph Orchestrator — unic-claude-code-format Roadmap
-
-You are implementing the roadmap for the `unic-claude-code-format` Claude Code plugin, one spec at a time.
-
-## Step 1 — Determine what's next
-
-Check the execution order in `docs/plans/README.md`. Then scan spec files in order and find the first file that does NOT contain the string `**Status: done`.
-
-If **all** specs contain `**Status: done**, output exactly:
-
-```
-<promise>LOOP_COMPLETE</promise>
-```
-
-…and stop. Do nothing else.
-
-## Step 2 — Read the spec completely
-
-Read the entire spec file before writing any code. Pay special attention to:
-
-- **Current behaviour** — verify the code actually works this way before starting; if it doesn't, add a `## Deviations` section at the bottom of the spec file documenting the discrepancy
-- **Out of scope** — list of things you must NOT change in this iteration
-- **Depends on** — if a dependency spec is not yet marked done, skip this spec and move to the next one that can be done
-
-## Step 3 — Implement
-
-Check the spec's `**Version impact:**` line to choose the implementation approach:
-
-**`none` (workspace/infrastructure spec)** — implement directly. Follow the "Implementation steps" exactly. If a step's "before" snapshot doesn't match the current file, consult the "Deviations" section (if you wrote one) or document the discrepancy and adapt minimally.
-
-**`patch` / `minor` / `major` (plugin spec)** — use `/tdd` to drive implementation. Treat the spec's "Implementation steps" as guidance (key files to touch, rough order), not a recipe. The spec's **Acceptance criteria** are the target; the red-green-refactor cycle drives the path there.
-
-Ground rules (from `docs/plans/README.md`):
-
-- Use `pnpm` for all package operations
-- Tabs for indentation, LF line endings (per `.editorconfig`)
-- Conventional commits: `feat(scope): description`, `fix(scope): description`, `chore(scope): description`
-- Cross-platform: use Node.js APIs (`node:path`, `node:fs`, `node:os`) instead of shell commands; no bash/sh assumptions
-- **Never hand-edit** `.claude-plugin/marketplace.json` version — use `pnpm bump`
-- If something can't be followed as written: document it in `## Deviations`, don't silently deviate
-
-**Supporting skill:** `/diagnose` — if a failure persists and the root cause appears to be outside the spec's scope (an existing module, tooling, or dependency the spec doesn't touch), use `/diagnose` to isolate it before continuing.
-
-## Step 4 — Verify
-
-Run the exact commands in the spec's **Verification** section. Fix any failures before proceeding.
-
-Check every item in **Acceptance criteria**. If any item fails, fix it.
-
-Then run the always-on repo hygiene checks regardless of spec type:
-
-```sh
-pnpm -w check   # Biome + Prettier (workspace root) — fix any failures before proceeding
-```
-
-## Step 4.5 — Bump version + CHANGELOG
-
-1. Read the spec's `**Version impact:** patch|minor|major` line at the top of the spec file.
-   If the line is absent, infer: breaking CLI/contract change → `major`; new flag/feature → `minor`; bug fix, refactor, docs → `patch`.
-
-2. Append **one bullet** to the matching subsection under `## [Unreleased]` in `CHANGELOG.md`, replacing the `- (none)` placeholder on first use:
-
-   - `### Breaking` — on-disk file schema change, plugin-to-hook contract change
-   - `### Added` — new feature, new configuration option, new extension support
-   - `### Fixed` — bug fix, refactor, docs, internal tooling
-
-3. Run:
-
-   ```sh
-   pnpm bump <patch|minor|major>
-   ```
-
-   This atomically increments `plugin.json` version, mirrors into `marketplace.json`, and promotes `[Unreleased]` → a new dated section.
-
-4. Run `pnpm verify:changelog` to confirm the check passes.
-
-## Step 5 — Mark done and commit
-
-1. Add this line immediately after the spec's `# NN. Title` heading (as the second line):
-
-```
-**Status: done — YYYY-MM-DD**
-```
-
-Replace `YYYY-MM-DD` with today's date.
-
-1. Stage and commit all changes:
-
-```sh
-git add -A
-git commit -m "feat(spec-NN): <short description of what was implemented> (vX.Y.Z)"
-```
-
-Replace `NN` with the spec number (e.g. `00`, `03`) and write a clear description.
-Replace `X.Y.Z` with the version output by `pnpm bump`.
-
-1. **Do not push.** Commits only.
-
-## Step 6 — Stop for this iteration
-
-Output a brief summary of what was implemented and committed. Then stop — Ralph will feed this prompt again for the next spec.
-
----
-
-## Important constraints
-
-- Implement **one spec per iteration**. Do not implement multiple specs in a single run.
-- If a spec requires human judgment (e.g. choosing between two design approaches not covered by the spec), stop, document the question in a `## Questions` section at the bottom of the spec file, and output:
-
-  ```
-  <promise>LOOP_COMPLETE</promise>
-  ```
-
-  This pauses the loop for human review.
-
-- Do not create any files outside of what the spec describes.
-- Do not modify any spec files except to add `**Status: done**`, `## Deviations`, or `## Questions` sections.
diff --git a/apps/claude-code/auto-format/README.md b/apps/claude-code/auto-format/README.md
index 53e2a99d..fed5efa6 100644
--- a/apps/claude-code/auto-format/README.md
+++ b/apps/claude-code/auto-format/README.md
@@ -155,7 +155,7 @@ To disable permanently for a repo: remove the plugin from `.claude/settings.json
 
 ## Contributing
 
-See `docs/plans/README.md` for the ralph-orchestrated implementation roadmap and ground rules.
+See the [project issue tracker](https://github.com/unic/unic-agents-plugins/issues) for the implementation roadmap. Open an issue to propose a change before starting implementation.
 
 ### Developer commands
 
diff --git a/apps/claude-code/auto-format/docs/plans/00-bootstrap-plugin-tooling.md b/apps/claude-code/auto-format/docs/plans/00-bootstrap-plugin-tooling.md
deleted file mode 100644
index 9d6888ce..00000000
--- a/apps/claude-code/auto-format/docs/plans/00-bootstrap-plugin-tooling.md
+++ /dev/null
@@ -1,176 +0,0 @@
-# 00. Bootstrap Plugin Tooling
-**Status: done — 2026-04-27**
-
-**Priority:** P0
-**Effort:** S
-**Version impact:** patch
-**Depends on:** none
-**Touches:** `package.json`, `pnpm-workspace.yaml`, `.npmrc`, `.editorconfig`
-
-## Context
-
-The repo contains only scaffolding files (`README.md`, `CHANGELOG.md`, `LICENSE`, `.gitignore`, `PROMPT.md`, `ralph.yml`). Before implementing any plugin features, we need a proper pnpm workspace with catalog-pinned deps, a Node version pin, engine strictness, and EditorConfig so all files follow the UNIC indentation convention (tabs for code, spaces for JSON/YAML/Markdown).
-
-The plugin's hook script (`scripts/format-hook.mjs`) uses **only Node built-ins** (`node:child_process`, `node:fs`, `node:path`) — no external runtime deps. The `package.json` will have no `dependencies` entry. `devDependencies` (for future test/lint/bump tooling) will use the `catalog:` protocol.
-
-## Current behaviour
-
-The repo root contains:
-- `README.md` — stub
-- `CHANGELOG.md` — skeleton
-- `LICENSE` — LGPL-3.0
-- `.gitignore` — basic
-- `PROMPT.md` — Ralph orchestrator prompt
-- `ralph.yml` — Ralph config
-- `docs/plans/` — this spec and siblings
-
-No `package.json`, no `pnpm-workspace.yaml`, no `.npmrc`, no `.editorconfig`.
-
-## Target behaviour
-
-- `package.json` exists with `name`, `version`, `license`, `type: module`, `packageManager`, `engines`, `private: true`, and a `scripts` section with at least `{"test": "echo 'No tests yet'"}`.
-- `pnpm-workspace.yaml` exists with `catalog:` (empty for now), `useNodeVersion`, `engineStrict: true`, `minimumReleaseAge: 1440`, `preferWorkspacePackages: true`, `strictDepBuilds: true`, `trustPolicy: no-downgrade`.
-- `.npmrc` enforces `engine-strict=true`, `save-exact=true`, `auto-install-peers=true`, `save-prefix=""`.
-- `.editorconfig` exists matching UNIC standard.
-- `pnpm install` succeeds (no deps yet, just generates a minimal lockfile).
-
-## Implementation steps
-
-### Step 1 — Fetch exact pnpm version
-
-Run:
-```sh
-pnpm --version
-```
-
-Record the output (e.g. `10.33.0`). You will use it in `packageManager`.
-
-### Step 2 — Fetch latest Node 24.x LTS patch
-
-Fetch the current Node 24.x LTS version from https://nodejs.org/en/download — look for the "LTS" row with major version 24. Record the full patch version string (e.g. `24.15.0`).
-
-### Step 3 — Create `package.json`
-
-Create `/Users/oriol.torrent/Sites/UNIC/unic-claude-code-format/package.json` with this content, substituting the real pnpm version and Node version from steps 1 and 2:
-
-```json
-{
-	"name": "unic-claude-code-format",
-	"version": "0.1.0",
-	"private": true,
-	"license": "LGPL-3.0-or-later",
-	"type": "module",
-	"packageManager": "pnpm@PNPM_VERSION",
-	"engines": {
-		"node": ">=24",
-		"pnpm": ">=10"
-	},
-	"scripts": {
-		"test": "echo 'No tests yet'"
-	}
-}
-```
-
-Replace `PNPM_VERSION` with the output of step 1 (e.g. `10.33.0`).
-
-### Step 4 — Create `pnpm-workspace.yaml`
-
-Create `/Users/oriol.torrent/Sites/UNIC/unic-claude-code-format/pnpm-workspace.yaml` with this content, substituting the real Node version:
-
-```yaml
-catalog: {}
-
-catalogMode: prefer
-engineStrict: true
-minimumReleaseAge: 1440
-preferWorkspacePackages: true
-strictDepBuilds: true
-trustPolicy: no-downgrade
-useNodeVersion: "NODE_VERSION"
-```
-
-Replace `NODE_VERSION` with the string from step 2 (e.g. `24.15.0`).
-
-### Step 5 — Create `.npmrc`
-
-```
-engine-strict=true
-save-exact=true
-auto-install-peers=true
-save-prefix=""
-```
-
-### Step 6 — Create `.editorconfig`
-
-```
-root = true
-
-[*]
-indent_style = tab
-end_of_line = lf
-charset = utf-8
-trim_trailing_whitespace = true
-insert_final_newline = true
-
-[*.md]
-trim_trailing_whitespace = false
-indent_style = space
-indent_size = 2
-
-[{*.json,*.yml,*.yaml}]
-indent_style = space
-indent_size = 2
-```
-
-### Step 7 — Run pnpm install
-
-```sh
-pnpm install
-```
-
-Confirm `pnpm-lock.yaml` is generated (no errors).
-
-### Step 8 — Commit
-
-```sh
-git add package.json pnpm-workspace.yaml pnpm-lock.yaml .npmrc .editorconfig
-git commit -m "chore(spec-00): bootstrap pnpm workspace and tooling"
-```
-
-## Test cases
-
-| Command | Expected | Exit code |
-|---|---|---|
-| `pnpm install --frozen-lockfile` on a fresh clone | No error; no deps to install | 0 |
-| `node -e "require('./package.json')"` | (Should fail — ESM. Try `import()` or just verify file is valid JSON.) | — |
-| `cat package.json \| python3 -m json.tool` | Valid JSON output | 0 |
-
-## Acceptance criteria
-
-- `package.json` is valid JSON with `name`, `version`, `type: module`, `packageManager`, and `engines` fields.
-- `pnpm-workspace.yaml` exists with `useNodeVersion`, `engineStrict: true`, and an empty `catalog:`.
-- `.npmrc` has `engine-strict=true` and `save-exact=true`.
-- `.editorconfig` has `root = true` and tab indentation for `[*]`, space indentation for JSON/YAML/Markdown.
-- `pnpm install --frozen-lockfile` exits 0.
-
-## Verification
-
-```sh
-# 1. Valid package.json
-cat package.json | python3 -m json.tool > /dev/null && echo "OK: valid JSON"
-
-# 2. pnpm install is clean
-pnpm install --frozen-lockfile && echo "OK: frozen install"
-
-# 3. .editorconfig has root=true
-grep "^root = true" .editorconfig && echo "OK: editorconfig root"
-```
-
-## Out of scope
-
-- No devDependencies yet (added in later specs as tooling is introduced).
-- No test runner setup (spec 09).
-- No CI (spec 10).
-- No `.claude-plugin/` manifest (spec 01).
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/01-plugin-manifest.md b/apps/claude-code/auto-format/docs/plans/01-plugin-manifest.md
deleted file mode 100644
index 389288cd..00000000
--- a/apps/claude-code/auto-format/docs/plans/01-plugin-manifest.md
+++ /dev/null
@@ -1,104 +0,0 @@
-# 01. Plugin Manifest
-**Status: done — 2026-04-27**
-
-**Priority:** P0
-**Effort:** S
-**Version impact:** minor
-**Depends on:** spec-00
-**Touches:** `.claude-plugin/plugin.json`, `.claude-plugin/marketplace.json`
-
-## Context
-
-Claude Code plugins are identified by their `.claude-plugin/plugin.json` manifest. This is the contract that `claude` reads when a user installs or enables the plugin. We also need `marketplace.json` for marketplace listing metadata. Both files must be present before the plugin is installable.
-
-Ref: Claude Code plugin spec (verify exact schema via `/plugin-dev:plugin-structure` skill if in doubt).
-
-## Current behaviour
-
-After spec `00`: `package.json` exists with `"version": "0.1.0"`. No `.claude-plugin/` directory exists.
-
-## Target behaviour
-
-- `.claude-plugin/plugin.json` exists with name, version, description, author, homepage, license, keywords.
-- `.claude-plugin/marketplace.json` exists with display metadata for the Claude Code marketplace.
-- Both files have version `0.1.0` matching `package.json`.
-- The `.claude-plugin/` directory is committed.
-
-## Implementation steps
-
-### Step 1 — Create `.claude-plugin/plugin.json`
-
-Read the current version from `package.json` first (`"version": "0.1.0"`).
-
-Create `.claude-plugin/plugin.json`:
-
-```json
-{
-	"name": "unic-claude-code-format",
-	"version": "0.1.0",
-	"description": "Auto-format and lint files when Claude Code edits them. Runs Prettier + ESLint --fix after Write/Edit/MultiEdit/NotebookEdit.",
-	"author": {
-		"name": "Unic AG",
-		"url": "https://www.unic.com"
-	},
-	"homepage": "https://github.com/unic/unic-claude-code-format",
-	"license": "LGPL-3.0-or-later",
-	"keywords": ["formatter", "prettier", "eslint", "hook", "unic"]
-}
-```
-
-### Step 2 — Create `.claude-plugin/marketplace.json`
-
-```json
-{
-	"name": "unic-claude-code-format",
-	"version": "0.1.0",
-	"displayName": "Unic Auto-Format",
-	"description": "Auto-format and lint files when Claude Code edits them. Uses the consumer repo's Prettier and ESLint — no bundled formatters.",
-	"author": "Unic AG",
-	"homepage": "https://github.com/unic/unic-claude-code-format",
-	"license": "LGPL-3.0-or-later",
-	"keywords": ["formatter", "prettier", "eslint", "hook", "unic"],
-	"tags": ["productivity", "code-quality"]
-}
-```
-
-### Step 3 — Commit
-
-```sh
-git add .claude-plugin/
-git commit -m "feat(spec-01): add plugin manifest and marketplace metadata (v0.1.0)"
-```
-
-## Acceptance criteria
-
-- `.claude-plugin/plugin.json` is valid JSON with `name`, `version`, `description`, `author`, `homepage`, `license`, `keywords`.
-- `.claude-plugin/marketplace.json` is valid JSON with display metadata.
-- Both files have version `0.1.0`.
-- Both files use 2-space indentation (JSON convention per `.editorconfig`).
-
-## Verification
-
-```sh
-# 1. Both files are valid JSON
-cat .claude-plugin/plugin.json | python3 -m json.tool > /dev/null && echo "OK: plugin.json valid"
-cat .claude-plugin/marketplace.json | python3 -m json.tool > /dev/null && echo "OK: marketplace.json valid"
-
-# 2. Versions match package.json
-node -e "
-  import('./package.json', {assert:{type:'json'}}).then(p => {
-    import('./.claude-plugin/plugin.json', {assert:{type:'json'}}).then(pl => {
-      console.assert(p.default.version === pl.default.version, 'version mismatch');
-      console.log('OK: versions match');
-    });
-  });
-"
-```
-
-## Out of scope
-
-- No hooks registration yet (spec 02).
-- No version bump tooling (spec 07).
-- The `marketplace.json` format may vary by Claude Code release; the schema here is based on `unic-claude-code-confluence`. If the `/plugin-dev:plugin-structure` skill reveals a different required schema, document in `## Deviations`.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/02-hook-registration.md b/apps/claude-code/auto-format/docs/plans/02-hook-registration.md
deleted file mode 100644
index 76310f29..00000000
--- a/apps/claude-code/auto-format/docs/plans/02-hook-registration.md
+++ /dev/null
@@ -1,89 +0,0 @@
-# 02. Hook Registration
-**Status: done — 2026-04-27**
-
-**Priority:** P0
-**Effort:** S
-**Version impact:** minor
-**Depends on:** spec-01
-**Touches:** `hooks/hooks.json`
-
-## Context
-
-Claude Code discovers plugin hooks by reading `hooks/hooks.json` in the plugin install directory. This file declares which Claude Code tool events to listen on and which command to run. Without this file, the plugin does nothing even after install.
-
-The hook must match `Write|Edit|MultiEdit|NotebookEdit` — these are the four Claude Code built-in tools that write file content. `MultiEdit` and `NotebookEdit` are commonly omitted by other implementations (e.g. `DXP-Profileservices` only matches `Write|Edit`) but both can be triggered by Claude in normal workflows.
-
-The command invokes the hook script (spec 03) via `node` rather than directly, so it works on systems where `.mjs` files may not have executable permissions (Windows, some CI).
-
-## Current behaviour
-
-After spec `01`: no `hooks/` directory exists. The plugin is not yet wired to any Claude Code events.
-
-## Target behaviour
-
-- `hooks/hooks.json` exists with a `PostToolUse` registration for `Write|Edit|MultiEdit|NotebookEdit`.
-- The command is `node ${CLAUDE_PLUGIN_ROOT}/scripts/format-hook.mjs`.
-- `CLAUDE_PLUGIN_ROOT` is the env var Claude Code sets to the plugin's install path. `CLAUDE_PROJECT_DIR` (used in the script itself) is set to the consumer project's working directory.
-- The hooks directory and file are committed.
-
-## Implementation steps
-
-### Step 1 — Create `hooks/hooks.json`
-
-```json
-{
-	"hooks": {
-		"PostToolUse": [
-			{
-				"matcher": "Write|Edit|MultiEdit|NotebookEdit",
-				"hooks": [
-					{
-						"type": "command",
-						"command": "node ${CLAUDE_PLUGIN_ROOT}/scripts/format-hook.mjs"
-					}
-				]
-			}
-		]
-	}
-}
-```
-
-Note: `${CLAUDE_PLUGIN_ROOT}` is a literal string here — Claude Code performs environment variable substitution when loading the hook. Do NOT expand it yourself.
-
-### Step 2 — Commit
-
-```sh
-git add hooks/
-git commit -m "feat(spec-02): register PostToolUse hook for Write|Edit|MultiEdit|NotebookEdit"
-```
-
-## Acceptance criteria
-
-- `hooks/hooks.json` is valid JSON.
-- `PostToolUse` key is present.
-- Matcher is exactly `"Write|Edit|MultiEdit|NotebookEdit"`.
-- Command contains `${CLAUDE_PLUGIN_ROOT}/scripts/format-hook.mjs`.
-
-## Verification
-
-```sh
-# 1. Valid JSON
-cat hooks/hooks.json | python3 -m json.tool > /dev/null && echo "OK: hooks.json valid"
-
-# 2. Matcher is correct
-node -e "
-  import('./hooks/hooks.json', {assert:{type:'json'}}).then(h => {
-    const m = h.default.hooks.PostToolUse[0].matcher;
-    console.assert(m === 'Write|Edit|MultiEdit|NotebookEdit', 'Wrong matcher: ' + m);
-    console.log('OK: matcher correct:', m);
-  });
-"
-```
-
-## Out of scope
-
-- No hook script yet (spec 03).
-- No PreToolUse hook (not in scope for this plugin).
-- No plugin-level `.claude/settings.json` — that would override user settings. Hooks belong in `hooks/hooks.json`.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/03-format-hook-script.md b/apps/claude-code/auto-format/docs/plans/03-format-hook-script.md
deleted file mode 100644
index 4e8b5122..00000000
--- a/apps/claude-code/auto-format/docs/plans/03-format-hook-script.md
+++ /dev/null
@@ -1,239 +0,0 @@
-# 03. Format Hook Script
-**Status: done — 2026-04-27**
-
-**Priority:** P0
-**Effort:** M
-**Version impact:** minor
-**Depends on:** spec-02
-**Touches:** `scripts/format-hook.mjs`
-
-## Context
-
-The hook script is the core of the plugin. It receives a JSON event from Claude Code on stdin, extracts the path of the file that was just written or edited, and runs Prettier (and optionally ESLint `--fix`) on that file.
-
-Design invariants (borrowed from `DXP-Profileservices` pattern and hardened):
-1. **Always exits 0** — never blocks Claude's tool flow.
-2. **Silent on success** — errors go to stderr only.
-3. **Defensive skipping** — maintains its own `SKIP_PREFIXES` list in addition to relying on `.prettierignore` / `eslint.config.*` ignores; short-circuits before invoking any external tool.
-4. **Path-traversal guard** — rejects any path that resolves outside the consumer project root.
-5. **Pre-flight existence checks** — if `node_modules/.bin/prettier` or `node_modules/.bin/eslint` don't exist, log to stderr and return (consumer repo just hasn't installed yet).
-6. **No external deps** — uses only Node built-ins.
-7. **Extension filtering** — only invokes Prettier/ESLint on file extensions they support.
-
-Note on ESLint exit codes:
-- Status 0: no lint issues (or all fixed).
-- Status 1: lint warnings/errors remain after `--fix`. This is NOT a hook failure — the file was still formatted. The hook treats status 1 as acceptable.
-- Status >1: ESLint crashed or was misconfigured. Log to stderr.
-
-## Current behaviour
-
-After spec `02`: `scripts/` directory does not exist. `hooks/hooks.json` registers `node ${CLAUDE_PLUGIN_ROOT}/scripts/format-hook.mjs` but the file is missing (Claude will log an error when the hook fires).
-
-## Target behaviour
-
-- `scripts/format-hook.mjs` exists and is a valid Node ESM module.
-- The script reads JSON from stdin, extracts `tool_input.file_path` (or `tool_input.notebook_path` for NotebookEdit), validates it, and runs Prettier + ESLint on it.
-- The script exits 0 in all circumstances (including errors).
-- Prettier is invoked as `node ./node_modules/.bin/prettier --write --ignore-unknown --log-level warn <file>` from `CLAUDE_PROJECT_DIR`.
-- ESLint is invoked only for `ALLOWED_ESLINT_EXT` extensions as `node ./node_modules/.bin/eslint --fix --no-error-on-unmatched-pattern <file>` from `CLAUDE_PROJECT_DIR`.
-- Files matching `SKIP_PREFIXES` are skipped before any tool is invoked.
-- ESLint status 1 is treated as acceptable (lint warnings remain but fix ran). Status >1 logs to stderr.
-
-## Implementation steps
-
-### Step 1 — Create `scripts/format-hook.mjs`
-
-Create the file at `scripts/format-hook.mjs` with this exact content:
-
-```js
-#!/usr/bin/env node
-/**
- * unic-claude-code-format — PostToolUse hook
- * Runs Prettier (and ESLint --fix where applicable) on the file
- * Claude just wrote or edited in the consumer project.
- *
- * Invariants:
- *   - Always exits 0 — never blocks Claude's tool flow.
- *   - Silent on success; diagnostics go to stderr only.
- *   - Defensively skips _bmad/, BMad-installed skills, generated dirs,
- *     and any path outside the consumer project root.
- *   - Does not bundle Prettier/ESLint — uses the consumer's node_modules
- *     so each repo keeps its own pinned versions and configs.
- */
-
-import { spawnSync } from 'node:child_process'
-import { existsSync } from 'node:fs'
-import { extname, relative, resolve } from 'node:path'
-
-const PROJECT_DIR = process.env.CLAUDE_PROJECT_DIR || process.cwd()
-
-// Defensive skip list. Kept in sync with the convention in .prettierignore and
-// eslint.config.js so the hook short-circuits before invoking any external tool.
-// _bmad/ is intentionally excluded: BMad source is never modified by end-users.
-const SKIP_PREFIXES = [
-	'_bmad/',
-	'.claude/skills/bmad-',
-	'.claude/worktrees/',
-	'.history/',
-	'.git/',
-	'node_modules/',
-	'dist/',
-	'build/',
-	'.next/',
-	'coverage/',
-]
-
-const ALLOWED_PRETTIER_EXT = new Set([
-	'.md',
-	'.mdx',
-	'.json',
-	'.jsonc',
-	'.yml',
-	'.yaml',
-	'.js',
-	'.mjs',
-	'.cjs',
-	'.ts',
-	'.mts',
-	'.cts',
-	'.tsx',
-	'.feature',
-])
-
-const ALLOWED_ESLINT_EXT = new Set([
-	'.js',
-	'.mjs',
-	'.cjs',
-	'.ts',
-	'.mts',
-	'.cts',
-	'.tsx',
-	'.json',
-	'.jsonc',
-	'.md',
-])
-
-const PRETTIER_BIN = resolve(PROJECT_DIR, 'node_modules/.bin/prettier')
-const ESLINT_BIN = resolve(PROJECT_DIR, 'node_modules/.bin/eslint')
-
-function shouldSkip(rel) {
-	if (rel.startsWith('..')) return true
-	return SKIP_PREFIXES.some((p) => rel.startsWith(p))
-}
-
-function runPrettier(filePath) {
-	if (!existsSync(PRETTIER_BIN)) return
-	const r = spawnSync('node', [PRETTIER_BIN, '--write', '--ignore-unknown', '--log-level', 'warn', filePath], {
-		cwd: PROJECT_DIR,
-		stdio: ['ignore', 'ignore', 'pipe'],
-	})
-	if (r.status !== 0) {
-		process.stderr.write(`unic-format: prettier failed: ${r.stderr?.toString().trim() || 'unknown error'}\n`)
-	}
-}
-
-function runEslint(filePath) {
-	if (!existsSync(ESLINT_BIN)) return
-	const r = spawnSync('node', [ESLINT_BIN, '--fix', '--no-error-on-unmatched-pattern', filePath], {
-		cwd: PROJECT_DIR,
-		stdio: ['ignore', 'ignore', 'pipe'],
-	})
-	// Status 1 = lint warnings/errors remain after --fix (not a hook failure).
-	// Status >1 = ESLint crash or misconfiguration.
-	if (r.status !== 0 && r.status !== 1) {
-		process.stderr.write(`unic-format: eslint failed (exit ${r.status}): ${r.stderr?.toString().trim() || 'unknown error'}\n`)
-	}
-}
-
-async function main() {
-	let buf = ''
-	for await (const chunk of process.stdin) buf += chunk
-	if (!buf.trim()) return
-
-	let event
-	try {
-		event = JSON.parse(buf)
-	} catch {
-		process.stderr.write('unic-format: could not parse hook input as JSON\n')
-		return
-	}
-
-	const filePath = event?.tool_input?.file_path || event?.tool_input?.notebook_path
-	if (!filePath || !existsSync(filePath)) return
-
-	const rel = relative(PROJECT_DIR, filePath)
-	if (shouldSkip(rel)) return
-
-	const ext = extname(rel).toLowerCase()
-	if (!ALLOWED_PRETTIER_EXT.has(ext)) return
-
-	runPrettier(filePath)
-	if (ALLOWED_ESLINT_EXT.has(ext)) runEslint(filePath)
-}
-
-main()
-	.catch((err) => process.stderr.write(`unic-format: unexpected error: ${err?.message || err}\n`))
-	.finally(() => process.exit(0))
-```
-
-### Step 2 — Commit
-
-```sh
-git add scripts/
-git commit -m "feat(spec-03): implement format hook script with prettier and eslint support"
-```
-
-## Test cases
-
-Manual smoke tests (automated tests come in spec 09):
-
-| Scenario | Expected |
-|---|---|
-| Pipe `{}` (empty JSON) to the script | Exits 0, no output |
-| Pipe `{"tool_input":{"file_path":"/nonexistent/file.md"}}` | Exits 0, no output (file doesn't exist guard) |
-| Pipe with `file_path` pointing to `_bmad/foo.md` | Exits 0, no output (SKIP_PREFIXES guard) |
-| Pipe with `file_path` pointing to a real `.md` file outside consumer node_modules | Exits 0; Prettier runs if installed |
-| Pipe with `file_path` pointing to a `.toml` file | Exits 0, no output (extension not in ALLOWED_PRETTIER_EXT) |
-
-Run these manually from a consumer repo:
-```sh
-CLAUDE_PROJECT_DIR="$(pwd)" CLAUDE_PLUGIN_ROOT="/path/to/plugin" \
-  node /path/to/plugin/scripts/format-hook.mjs <<< '{}'
-echo "exit=$?"
-```
-
-## Acceptance criteria
-
-- `scripts/format-hook.mjs` is a valid ESM module (no `require()`, uses `import` and `import.meta`... actually `import()` is not used — it uses `import { ... } from 'node:...'` which is fine).
-- Script exits 0 for all inputs (including invalid JSON, missing file, missing binaries).
-- `SKIP_PREFIXES` includes `_bmad/` as first entry.
-- Prettier is invoked with `--ignore-unknown` so unsupported extensions (TOML, CSV, PY) are silently skipped.
-- ESLint status 1 is not treated as an error.
-- Script has no external npm dependencies (only `node:` built-ins).
-
-## Verification
-
-```sh
-# 1. Valid ESM syntax (quick parse check)
-node --input-type=module <<< "import './scripts/format-hook.mjs'" && echo "OK: valid ESM"
-
-# 2. Exits 0 with empty input
-echo '{}' | CLAUDE_PROJECT_DIR="$(pwd)" node scripts/format-hook.mjs
-echo "exit=$?"
-
-# 3. Skips _bmad/ path
-echo "{\"tool_input\":{\"file_path\":\"$(pwd)/_bmad/test.md\"}}" \
-  | CLAUDE_PROJECT_DIR="$(pwd)" node scripts/format-hook.mjs
-echo "exit=$?"
-
-# 4. No external deps imported
-grep -E "^import" scripts/format-hook.mjs | grep -v "node:" && echo "FAIL: non-built-in import found" || echo "OK: only node: imports"
-```
-
-## Out of scope
-
-- Per-project config support (spec 04).
-- Automated test suite (spec 09).
-- No changes to `hooks/hooks.json` — that was done in spec 02.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/04-per-project-config.md b/apps/claude-code/auto-format/docs/plans/04-per-project-config.md
deleted file mode 100644
index 01e6e165..00000000
--- a/apps/claude-code/auto-format/docs/plans/04-per-project-config.md
+++ /dev/null
@@ -1,173 +0,0 @@
-# 04. Per-Project Config Support
-**Status: done — 2026-04-27**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** minor
-**Depends on:** spec-03
-**Touches:** `scripts/format-hook.mjs`
-
-## Context
-
-Different consumer repos have different skip requirements. `agentic-delivery-model` needs `_bmad/` skipped; another UNIC repo might have a `_legacy/` directory to exclude, or might only want Prettier on `.md` files and not `.json`. Rather than forking the plugin per repo, support a lightweight per-project config file: `.claude/unic-format.json` in the consumer project root.
-
-If the config file is absent, the plugin uses its built-in defaults (spec 03 values). If present, it merges on a key-by-key basis — a consumer can override `skipPrefixes` without touching `eslintExtensions`, etc.
-
-## Current behaviour
-
-After spec `03`: `scripts/format-hook.mjs` has hardcoded `SKIP_PREFIXES`, `ALLOWED_PRETTIER_EXT`, and `ALLOWED_ESLINT_EXT` constants. No config file is read.
-
-## Target behaviour
-
-- The hook script reads `$CLAUDE_PROJECT_DIR/.claude/unic-format.json` if it exists.
-- The config file may override any of: `skipPrefixes` (array of strings), `prettierExtensions` (array of strings), `eslintExtensions` (array of strings).
-- Omitting a key in the config means "use plugin default for that key" (not "empty array").
-- If the config file is present but malformed JSON, log a warning to stderr and fall back to defaults.
-- If the config file is absent, behaviour is identical to spec 03 (pure defaults).
-
-## Implementation steps
-
-### Step 1 — Update `scripts/format-hook.mjs`
-
-Add `readFileSync` to the import from `node:fs`:
-
-Before:
-```js
-import { existsSync } from 'node:fs'
-```
-
-After:
-```js
-import { existsSync, readFileSync } from 'node:fs'
-```
-
-### Step 2 — Extract defaults and add `loadProjectConfig`
-
-After the `PROJECT_DIR` line and before `SKIP_PREFIXES`, insert:
-
-```js
-const DEFAULTS = {
-	skipPrefixes: [
-		'_bmad/',
-		'.claude/skills/bmad-',
-		'.claude/worktrees/',
-		'.history/',
-		'.git/',
-		'node_modules/',
-		'dist/',
-		'build/',
-		'.next/',
-		'coverage/',
-	],
-	prettierExtensions: [
-		'.md',
-		'.mdx',
-		'.json',
-		'.jsonc',
-		'.yml',
-		'.yaml',
-		'.js',
-		'.mjs',
-		'.cjs',
-		'.ts',
-		'.mts',
-		'.cts',
-		'.tsx',
-		'.feature',
-	],
-	eslintExtensions: ['.js', '.mjs', '.cjs', '.ts', '.mts', '.cts', '.tsx', '.json', '.jsonc', '.md'],
-}
-
-function loadProjectConfig() {
-	const configPath = resolve(PROJECT_DIR, '.claude/unic-format.json')
-	if (!existsSync(configPath)) return DEFAULTS
-	try {
-		const cfg = JSON.parse(readFileSync(configPath, 'utf8'))
-		return {
-			skipPrefixes: Array.isArray(cfg.skipPrefixes) ? cfg.skipPrefixes : DEFAULTS.skipPrefixes,
-			prettierExtensions: Array.isArray(cfg.prettierExtensions) ? cfg.prettierExtensions : DEFAULTS.prettierExtensions,
-			eslintExtensions: Array.isArray(cfg.eslintExtensions) ? cfg.eslintExtensions : DEFAULTS.eslintExtensions,
-		}
-	} catch (err) {
-		process.stderr.write(`unic-format: ignoring malformed .claude/unic-format.json: ${err.message}\n`)
-		return DEFAULTS
-	}
-}
-
-const CONFIG = loadProjectConfig()
-```
-
-### Step 3 — Replace hardcoded constants with CONFIG references
-
-Remove the old standalone `SKIP_PREFIXES`, `ALLOWED_PRETTIER_EXT`, `ALLOWED_ESLINT_EXT` constants.
-
-Update `shouldSkip`:
-```js
-function shouldSkip(rel) {
-	if (rel.startsWith('..')) return true
-	return CONFIG.skipPrefixes.some((p) => rel.startsWith(p))
-}
-```
-
-Update the extension checks in `main()`:
-```js
-const ext = extname(rel).toLowerCase()
-if (!new Set(CONFIG.prettierExtensions).has(ext)) return
-
-runPrettier(filePath)
-if (new Set(CONFIG.eslintExtensions).has(ext)) runEslint(filePath)
-```
-
-### Step 4 — Commit
-
-```sh
-git add scripts/format-hook.mjs
-git commit -m "feat(spec-04): add per-project config support via .claude/unic-format.json"
-```
-
-## Test cases
-
-| Scenario | Expected |
-|---|---|
-| No `.claude/unic-format.json` in consumer | Uses defaults; `_bmad/` is skipped |
-| Config with `{"skipPrefixes": ["_legacy/"]}` | `_legacy/` is skipped; `_bmad/` is NOT in defaults since config replaced `skipPrefixes` entirely |
-| Config with `{"prettierExtensions": [".md"]}` only | Only `.md` runs Prettier; other keys use defaults |
-| Config file is invalid JSON | Warning to stderr; falls back to defaults |
-| Config file has `skipPrefixes` as a non-array | Warning implied? No — uses `Array.isArray` check → falls back to default for that key |
-
-## Acceptance criteria
-
-- When `.claude/unic-format.json` is absent, behaviour is identical to spec 03.
-- When present and valid, config values override defaults on a per-key basis.
-- When present but malformed, a warning is written to stderr and defaults are used.
-- `DEFAULTS` object is defined and accurately reflects the spec 03 hardcoded values.
-
-## Verification
-
-```sh
-# 1. Exits 0 without config
-echo '{}' | CLAUDE_PROJECT_DIR="$(pwd)" node scripts/format-hook.mjs
-echo "exit=$?"
-
-# 2. Reads config override (create temp config)
-mkdir -p /tmp/test-consumer/.claude
-echo '{"prettierExtensions":[".md"]}' > /tmp/test-consumer/.claude/unic-format.json
-echo '{"tool_input":{"file_path":"/tmp/test-consumer/README.md"}}' \
-  | CLAUDE_PROJECT_DIR="/tmp/test-consumer" node scripts/format-hook.mjs
-echo "exit=$?"
-rm -rf /tmp/test-consumer
-
-# 3. Malformed config logs warning but exits 0
-mkdir -p /tmp/bad-config/.claude
-echo 'NOT JSON' > /tmp/bad-config/.claude/unic-format.json
-echo '{}' | CLAUDE_PROJECT_DIR="/tmp/bad-config" node scripts/format-hook.mjs 2>&1 | grep "malformed" && echo "OK: warning shown"
-rm -rf /tmp/bad-config
-```
-
-## Out of scope
-
-- No UI for editing the config (consumers edit `.claude/unic-format.json` manually).
-- No JSON Schema validation (array-type check is sufficient for v1).
-- No nested config merging (all or nothing per key).
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/05-readme.md b/apps/claude-code/auto-format/docs/plans/05-readme.md
deleted file mode 100644
index 57ef308e..00000000
--- a/apps/claude-code/auto-format/docs/plans/05-readme.md
+++ /dev/null
@@ -1,168 +0,0 @@
-# 05. Plugin README
-**Status: done — 2026-04-27**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-03, spec-04
-**Touches:** `README.md`
-
-## Context
-
-The `README.md` currently contains only a stub. We need full install + usage documentation so users know how to install the plugin, what it does, and how to configure it per-project.
-
-## Current behaviour
-
-After spec `04`:
-
-```md
-# unic-claude-code-format
-
-A Claude Code plugin that auto-formats and lints files whenever Claude Code writes or edits them. Designed for UNIC repositories used by non-dev users (Product Owners, BMad agents) who won't run `pnpm format` manually.
-
-> **Status:** in development. See `docs/plans/` for the ralph-orchestrated implementation roadmap.
-```
-
-## Target behaviour
-
-A complete README covering:
-1. What it does
-2. Requirements (consumer repo must have Prettier and/or ESLint installed locally)
-3. Install instructions
-4. Which events trigger the hook (Write/Edit/MultiEdit/NotebookEdit)
-5. What is skipped by default (SKIP_PREFIXES)
-6. Per-project configuration (`.claude/unic-format.json`)
-7. How to disable per-session
-8. Invariants (always exits 0, silent on success)
-9. Contributing pointer
-
-## Implementation steps
-
-### Step 1 — Overwrite `README.md`
-
-Replace the full content with:
-
-```md
-# unic-claude-code-format
-
-A Claude Code plugin for UNIC repositories. Auto-formats and lints files whenever Claude Code writes or edits them — using the consumer repo's own Prettier and ESLint installations.
-
-Designed for repositories used by non-dev users (Product Owners, BMad agents) who won't run `pnpm format` manually.
-
-## What it does
-
-After every `Write`, `Edit`, `MultiEdit`, or `NotebookEdit` tool call, the plugin:
-
-1. Extracts the file path from the Claude Code hook event.
-2. Checks the path against a skip list (see below).
-3. Runs `prettier --write --ignore-unknown` on the file (if Prettier is installed in the consumer repo).
-4. Runs `eslint --fix` on the file (if ESLint is installed and the extension is `.js`, `.mjs`, `.cjs`, `.ts`, `.mts`, `.cts`, `.tsx`, `.json`, `.jsonc`, or `.md`).
-
-The hook always exits 0 — it never blocks a Claude edit even if formatting fails.
-
-## Requirements
-
-The consumer repo must have Prettier and/or ESLint installed in its `node_modules`. The plugin discovers them at runtime via `$CLAUDE_PROJECT_DIR/node_modules/.bin/prettier` and `$CLAUDE_PROJECT_DIR/node_modules/.bin/eslint`. If neither is found, the hook is a no-op.
-
-## Installation
-
-```sh
-# In Claude Code, install from the UNIC GitHub org:
-/plugin install github:unic/unic-claude-code-format
-```
-
-Or add it to `.claude/settings.json` in a consumer repo:
-
-```json
-{
-  "enabledPlugins": {
-    "unic-claude-code-format": true
-  }
-}
-```
-
-## Files skipped by default
-
-The following paths are skipped before any tool is invoked:
-
-| Prefix | Reason |
-|---|---|
-| `_bmad/` | BMad source — never user-modified |
-| `.claude/skills/bmad-` | BMad-installed skills |
-| `.claude/worktrees/` | Git worktrees |
-| `.history/` | VS Code local history |
-| `.git/` | Git internals |
-| `node_modules/` | Dependency installs |
-| `dist/`, `build/`, `.next/`, `coverage/` | Generated output |
-
-Files matched by the consumer repo's `.prettierignore` are also skipped (Prettier handles this natively via `--ignore-unknown`).
-
-## Per-project configuration
-
-Create `.claude/unic-format.json` in the consumer repo root to override defaults:
-
-```json
-{
-  "skipPrefixes": ["_bmad/", "_internal/", "vendor/"],
-  "prettierExtensions": [".md", ".json", ".yml", ".yaml"],
-  "eslintExtensions": [".js", ".ts", ".md"]
-}
-```
-
-Each key overrides the full default list for that key. Omit a key to keep the default. Config is loaded once per hook invocation.
-
-## How to disable
-
-To disable the hook for a Claude Code session:
-- In the Claude Code `/hooks` UI: toggle off `unic-claude-code-format`.
-
-To disable permanently for a repo: remove the plugin from `.claude/settings.json`.
-
-## Invariants
-
-- **Always exits 0.** The hook never blocks a Claude edit, even if Prettier/ESLint fails.
-- **Silent on success.** Output only appears on stderr when something goes wrong.
-- **No bundled tools.** The plugin does not bundle Prettier or ESLint — your repo's pinned versions and configs are used, so formatting is always deterministic.
-
-## Contributing
-
-See `docs/plans/README.md` for the ralph-orchestrated implementation roadmap and ground rules.
-```
-
-### Step 2 — Commit
-
-```sh
-git add README.md
-git commit -m "docs(spec-05): write full plugin README with install, config, and invariants"
-```
-
-## Acceptance criteria
-
-- README covers: what, requirements, install, skip list, per-project config, disable, invariants.
-- The install command matches the actual plugin name.
-- The `.claude/unic-format.json` example is valid JSON.
-- The skip list table matches `SKIP_PREFIXES` in `scripts/format-hook.mjs` (from spec 04).
-
-## Verification
-
-```sh
-# 1. Confirm README has install section
-grep -q "plugin install" README.md && echo "OK: install section present"
-
-# 2. Confirm skip list table is present
-grep -q "_bmad/" README.md && echo "OK: _bmad skip mentioned"
-
-# 3. Confirm unic-format.json example is valid JSON
-node -e "JSON.parse(require('fs').readFileSync('README.md','utf8').match(/\`\`\`json\n({[\s\S]*?skipPrefixes[\s\S]*?})\n\`\`\`/)[1])" && echo "OK: config example is valid JSON"
-```
-
-## Out of scope
-
-- No CHANGELOG section in README (CHANGELOG.md covers that).
-- No API documentation (no API exists).
-
-_Ralph: append findings here._
-
-## Deviations
-
-- The `enabledPlugins` settings block uses a plain ` ``` ` fence instead of ` ```json `. The spec shows `json` but the verification regex `/```json\n({[\s\S]*?skipPrefixes[\s\S]*?})\n```/` starts matching from the first ` ```json ` block, spans both code fences, and produces non-parseable content. Using a plain fence for the first block ensures the regex matches only the intended `skipPrefixes` example.
diff --git a/apps/claude-code/auto-format/docs/plans/06-changelog-and-versioning.md b/apps/claude-code/auto-format/docs/plans/06-changelog-and-versioning.md
deleted file mode 100644
index d3e9765d..00000000
--- a/apps/claude-code/auto-format/docs/plans/06-changelog-and-versioning.md
+++ /dev/null
@@ -1,106 +0,0 @@
-# 06. CHANGELOG and Versioning Docs
-**Status: done — 2026-04-27**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-01
-**Touches:** `CHANGELOG.md`, `docs/plans/README.md`
-
-## Context
-
-`CHANGELOG.md` exists as a skeleton (from the initial scaffold). Now that specs 00–05 have landed, we need to backfill the CHANGELOG with entries for all the changes made so far. We also add versioning policy documentation to `docs/plans/README.md` (it already has it, but the `CHANGELOG.md` backfill needs to reflect the real history).
-
-This spec does NOT introduce `pnpm bump` (spec 07) or `pnpm verify:changelog` (spec 08) — those come next. For now, CHANGELOG is maintained manually.
-
-## Current behaviour
-
-After specs 00–05:
-
-`CHANGELOG.md` contains the skeleton:
-```md
-## [Unreleased]
-
-### Breaking
-- (none)
-
-### Added
-- (none)
-
-### Fixed
-- (none)
-```
-
-## Target behaviour
-
-`CHANGELOG.md` has a proper `## [0.1.0] - YYYY-MM-DD` section (use today's date) with bullets for the core features landed in specs 00–05.
-
-## Implementation steps
-
-### Step 1 — Update `CHANGELOG.md`
-
-Replace the `## [Unreleased]` section's body with a dated release section, keeping `[Unreleased]` empty at the top for future changes.
-
-Target content (substitute today's date for YYYY-MM-DD):
-
-```md
-# Changelog
-
-All notable changes to this project will be documented in this file.
-
-The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
-and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
-
-## [Unreleased]
-
-### Breaking
-- (none)
-
-### Added
-- (none)
-
-### Fixed
-- (none)
-
-## [0.1.0] - YYYY-MM-DD
-
-### Added
-- PostToolUse hook for `Write|Edit|MultiEdit|NotebookEdit` Claude Code events.
-- Prettier (`--write --ignore-unknown`) applied to all supported extensions after each edit.
-- ESLint (`--fix`) applied to `.js/.mjs/.cjs/.ts/.mts/.cts/.tsx/.json/.jsonc/.md` after each edit.
-- Defensive `SKIP_PREFIXES` list (incl. `_bmad/`, BMad skills, generated dirs).
-- Path-traversal guard: paths resolving outside consumer project root are skipped.
-- Pre-flight binary existence checks: hook is a no-op if Prettier/ESLint is not installed.
-- Per-project config: `.claude/unic-format.json` in consumer root overrides `skipPrefixes`, `prettierExtensions`, `eslintExtensions`.
-- Plugin manifest: `.claude-plugin/plugin.json` and `.claude-plugin/marketplace.json`.
-```
-
-### Step 2 — Commit
-
-```sh
-git add CHANGELOG.md
-git commit -m "docs(spec-06): backfill CHANGELOG for v0.1.0 release"
-```
-
-## Acceptance criteria
-
-- `CHANGELOG.md` has an `## [Unreleased]` section at top (empty).
-- `CHANGELOG.md` has `## [0.1.0] - YYYY-MM-DD` section with bullets covering all specs 00–05.
-- The file follows Keep a Changelog format.
-
-## Verification
-
-```sh
-# 1. Unreleased section is present
-grep "## \[Unreleased\]" CHANGELOG.md && echo "OK: Unreleased present"
-
-# 2. 0.1.0 release section is present
-grep "## \[0.1.0\]" CHANGELOG.md && echo "OK: 0.1.0 section present"
-```
-
-## Out of scope
-
-- `pnpm bump` tooling (spec 07).
-- `pnpm verify:changelog` (spec 08).
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/07-pnpm-bump-tooling.md b/apps/claude-code/auto-format/docs/plans/07-pnpm-bump-tooling.md
deleted file mode 100644
index ffbadef3..00000000
--- a/apps/claude-code/auto-format/docs/plans/07-pnpm-bump-tooling.md
+++ /dev/null
@@ -1,181 +0,0 @@
-# 07. pnpm bump Tooling
-**Status: done — 2026-04-27**
-
-**Priority:** P1
-**Effort:** M
-**Version impact:** patch
-**Depends on:** spec-06
-**Touches:** `scripts/bump.mjs`, `package.json`
-
-## Context
-
-Manually keeping `.claude-plugin/plugin.json`, `.claude-plugin/marketplace.json`, and `CHANGELOG.md` in sync is error-prone. `pnpm bump` (borrowed from `unic-claude-code-confluence`) atomically:
-
-1. Increments the version in `package.json` and `.claude-plugin/plugin.json` and `.claude-plugin/marketplace.json`.
-2. Promotes the `## [Unreleased]` section in `CHANGELOG.md` to `## [NEW_VERSION] - YYYY-MM-DD`.
-
-## Current behaviour
-
-After spec `06`: no `pnpm bump` script. Version updates are manual.
-
-`package.json` scripts section:
-```json
-"scripts": {
-	"test": "echo 'No tests yet'"
-}
-```
-
-## Target behaviour
-
-- `scripts/bump.mjs` exists and accepts `patch`, `minor`, or `major` as first argument.
-- `pnpm bump patch` / `pnpm bump minor` / `pnpm bump major` runs atomically.
-- After running, `package.json`, `.claude-plugin/plugin.json`, `.claude-plugin/marketplace.json` all have the new version.
-- `CHANGELOG.md`'s `## [Unreleased]` is promoted to `## [NEW_VERSION] - TODAY`.
-- A fresh empty `## [Unreleased]` section is inserted at the top.
-- Script exits 1 with a clear message if the argument is missing or invalid.
-- Script exits 1 if `## [Unreleased]` has no non-(none) entries (prevents empty release notes).
-
-## Implementation steps
-
-### Step 1 — Create `scripts/bump.mjs`
-
-```js
-#!/usr/bin/env node
-/**
- * pnpm bump <patch|minor|major>
- * Atomically bumps version in package.json, plugin.json, marketplace.json
- * and promotes [Unreleased] in CHANGELOG.md.
- */
-
-import { readFileSync, writeFileSync } from 'node:fs'
-import { resolve } from 'node:path'
-
-const ROOT = new URL('..', import.meta.url).pathname.replace(/\/$/, '')
-
-function readJson(rel) {
-	return JSON.parse(readFileSync(resolve(ROOT, rel), 'utf8'))
-}
-
-function writeJson(rel, obj) {
-	writeFileSync(resolve(ROOT, rel), JSON.stringify(obj, null, '\t') + '\n', 'utf8')
-}
-
-function bumpVersion(version, type) {
-	const [major, minor, patch] = version.split('.').map(Number)
-	if (type === 'major') return `${major + 1}.0.0`
-	if (type === 'minor') return `${major}.${minor + 1}.0`
-	if (type === 'patch') return `${major}.${minor}.${patch + 1}`
-	throw new Error(`Invalid bump type: ${type}`)
-}
-
-function today() {
-	return new Date().toISOString().slice(0, 10)
-}
-
-const bumpType = process.argv[2]
-if (!['patch', 'minor', 'major'].includes(bumpType)) {
-	process.stderr.write(`Usage: pnpm bump <patch|minor|major>\n`)
-	process.exit(1)
-}
-
-const pkg = readJson('package.json')
-const newVersion = bumpVersion(pkg.version, bumpType)
-
-// Bump package.json
-pkg.version = newVersion
-writeJson('package.json', pkg)
-
-// Bump .claude-plugin/plugin.json
-const pluginJson = readJson('.claude-plugin/plugin.json')
-pluginJson.version = newVersion
-writeJson('.claude-plugin/plugin.json', pluginJson)
-
-// Bump .claude-plugin/marketplace.json
-const marketJson = readJson('.claude-plugin/marketplace.json')
-marketJson.version = newVersion
-writeJson('.claude-plugin/marketplace.json', marketJson)
-
-// Promote CHANGELOG
-const changelogPath = resolve(ROOT, 'CHANGELOG.md')
-let changelog = readFileSync(changelogPath, 'utf8')
-
-const unreleasedMatch = changelog.match(/## \[Unreleased\]([\s\S]*?)(?=## \[|$)/)
-if (!unreleasedMatch) {
-	process.stderr.write('CHANGELOG.md does not have an [Unreleased] section.\n')
-	process.exit(1)
-}
-
-const unreleasedBody = unreleasedMatch[1]
-const hasEntries = unreleasedBody.split('\n').some((line) => {
-	const trimmed = line.trim()
-	return trimmed.startsWith('-') && !trimmed.includes('(none)')
-})
-
-if (!hasEntries) {
-	process.stderr.write(`[Unreleased] has no entries. Add CHANGELOG entries before bumping.\n`)
-	process.exit(1)
-}
-
-const newUnreleased = `## [Unreleased]\n\n### Breaking\n- (none)\n\n### Added\n- (none)\n\n### Fixed\n- (none)\n\n`
-const newRelease = `## [${newVersion}] - ${today()}${unreleasedBody}`
-changelog = changelog.replace(/## \[Unreleased\][\s\S]*?(?=## \[|$)/, newUnreleased + newRelease)
-
-writeFileSync(changelogPath, changelog, 'utf8')
-
-process.stdout.write(`Bumped to v${newVersion}\n`)
-```
-
-### Step 2 — Add `bump` script to `package.json`
-
-Update `scripts`:
-```json
-"scripts": {
-	"test": "echo 'No tests yet'",
-	"bump": "node scripts/bump.mjs"
-}
-```
-
-### Step 3 — Commit
-
-```sh
-git add scripts/bump.mjs package.json
-git commit -m "feat(spec-07): add pnpm bump tooling for atomic version bumping"
-```
-
-## Test cases
-
-| Command | Expected |
-|---|---|
-| `pnpm bump` (no args) | Exits 1 with usage message |
-| `pnpm bump invalid` | Exits 1 with usage message |
-| `pnpm bump patch` (no [Unreleased] entries) | Exits 1 with "no entries" message |
-| `pnpm bump patch` (with entries) | Exits 0; versions bumped; [Unreleased] promoted |
-
-## Acceptance criteria
-
-- `scripts/bump.mjs` exists and is valid ESM.
-- `pnpm bump patch` exits 0 when `[Unreleased]` has entries, bumps all three version files, and promotes CHANGELOG.
-- `pnpm bump` (no args) exits 1.
-- `pnpm bump` with empty `[Unreleased]` exits 1.
-
-## Verification
-
-```sh
-# 1. Valid ESM
-node --input-type=module <<< "import './scripts/bump.mjs'" 2>&1 | head -3
-
-# 2. No-arg exits 1
-pnpm bump; echo "exit=$?"   # Expect exit=1
-
-# 3. Add an entry and dry-run bump on patch (then git restore)
-# Manual test: add a bullet to [Unreleased] in CHANGELOG, run pnpm bump patch,
-# verify versions all bumped, then git restore to undo.
-```
-
-## Out of scope
-
-- `pnpm tag` (git tagging) — not needed for development phase.
-- `pnpm publish` — not applicable (plugin distributed via git URL).
-- `pnpm verify:changelog` (spec 08).
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/08-pnpm-verify-changelog.md b/apps/claude-code/auto-format/docs/plans/08-pnpm-verify-changelog.md
deleted file mode 100644
index b41de8d5..00000000
--- a/apps/claude-code/auto-format/docs/plans/08-pnpm-verify-changelog.md
+++ /dev/null
@@ -1,131 +0,0 @@
-# 08. pnpm verify:changelog
-**Status: done — 2026-04-27**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-07
-**Touches:** `scripts/verify-changelog.mjs`, `package.json`
-
-## Context
-
-CI needs a way to enforce that `CHANGELOG.md` is properly formatted and that the `[Unreleased]` section is in the expected structure. `pnpm verify:changelog` is a lightweight script that checks these invariants without requiring an external CHANGELOG linting library.
-
-## Current behaviour
-
-After spec `07`: `package.json` has `"bump": "node scripts/bump.mjs"`. No `verify:changelog` script.
-
-## Target behaviour
-
-- `scripts/verify-changelog.mjs` exists.
-- `pnpm verify:changelog` exits 0 if CHANGELOG is well-formed.
-- `pnpm verify:changelog` exits 1 with a descriptive message if:
-  - `## [Unreleased]` section is missing.
-  - `## [Unreleased]` is missing `### Breaking`, `### Added`, or `### Fixed` subsections.
-  - Any `## [X.Y.Z]` section is missing a date suffix (` - YYYY-MM-DD`).
-
-## Implementation steps
-
-### Step 1 — Create `scripts/verify-changelog.mjs`
-
-```js
-#!/usr/bin/env node
-/**
- * pnpm verify:changelog
- * Checks that CHANGELOG.md has the expected structure.
- * Exits 0 on success, 1 on failure.
- */
-
-import { readFileSync } from 'node:fs'
-import { resolve } from 'node:path'
-
-const ROOT = new URL('..', import.meta.url).pathname.replace(/\/$/, '')
-const changelogPath = resolve(ROOT, 'CHANGELOG.md')
-
-let changelog
-try {
-	changelog = readFileSync(changelogPath, 'utf8')
-} catch {
-	process.stderr.write('verify:changelog: CHANGELOG.md not found.\n')
-	process.exit(1)
-}
-
-const errors = []
-
-// 1. [Unreleased] section must exist
-if (!changelog.includes('## [Unreleased]')) {
-	errors.push('Missing ## [Unreleased] section.')
-}
-
-// 2. [Unreleased] must have all three subsections
-const unreleasedIdx = changelog.indexOf('## [Unreleased]')
-if (unreleasedIdx !== -1) {
-	const nextReleaseIdx = changelog.indexOf('\n## [', unreleasedIdx + 1)
-	const unreleasedBlock =
-		nextReleaseIdx === -1
-			? changelog.slice(unreleasedIdx)
-			: changelog.slice(unreleasedIdx, nextReleaseIdx)
-	for (const sub of ['### Breaking', '### Added', '### Fixed']) {
-		if (!unreleasedBlock.includes(sub)) {
-			errors.push(`[Unreleased] is missing subsection: ${sub}`)
-		}
-	}
-}
-
-// 3. Each versioned release must have a date
-const releasePattern = /^## \[(\d+\.\d+\.\d+)\]/gm
-let match
-while ((match = releasePattern.exec(changelog)) !== null) {
-	const end = changelog.indexOf('\n', match.index)
-	const line = changelog.slice(match.index, end === -1 ? undefined : end)
-	if (!/ - \d{4}-\d{2}-\d{2}/.test(line)) {
-		errors.push(`Release section missing date: ${line.trim()}`)
-	}
-}
-
-if (errors.length > 0) {
-	process.stderr.write('verify:changelog failed:\n')
-	errors.forEach((e) => process.stderr.write(`  - ${e}\n`))
-	process.exit(1)
-}
-
-process.stdout.write('verify:changelog: OK\n')
-```
-
-### Step 2 — Add `verify:changelog` to `package.json` scripts
-
-Update scripts section to:
-```json
-"scripts": {
-	"test": "echo 'No tests yet'",
-	"bump": "node scripts/bump.mjs",
-	"verify:changelog": "node scripts/verify-changelog.mjs"
-}
-```
-
-### Step 3 — Commit
-
-```sh
-git add scripts/verify-changelog.mjs package.json
-git commit -m "feat(spec-08): add pnpm verify:changelog for CI enforcement"
-```
-
-## Acceptance criteria
-
-- `pnpm verify:changelog` exits 0 on the current well-formed CHANGELOG.
-- `pnpm verify:changelog` exits 1 if `[Unreleased]` is missing.
-- `pnpm verify:changelog` exits 1 if a versioned release is missing its date.
-
-## Verification
-
-```sh
-# 1. Passes on current CHANGELOG
-pnpm verify:changelog && echo "OK"
-```
-
-## Out of scope
-
-- No CHANGELOG format migration.
-- No auto-fix mode.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/09-smoke-tests.md b/apps/claude-code/auto-format/docs/plans/09-smoke-tests.md
deleted file mode 100644
index e8822489..00000000
--- a/apps/claude-code/auto-format/docs/plans/09-smoke-tests.md
+++ /dev/null
@@ -1,213 +0,0 @@
-# 09. Smoke Tests
-**Status: done — 2026-04-27**
-
-**Priority:** P1
-**Effort:** M
-**Version impact:** patch
-**Depends on:** spec-03, spec-04
-**Touches:** `tests/format-hook.test.mjs`, `package.json`
-
-## Context
-
-The hook script has no automated tests. Spec 09 adds smoke tests using Node's built-in `node:test` runner (no external test framework needed). Tests cover the core routing logic: which files are skipped, which get processed, and error paths.
-
-Since the tests can't run Prettier/ESLint in isolation (that would require a consumer repo), the test harness mocks `spawnSync` by redirecting `node_modules/.bin/prettier` to a no-op. Tests focus on: path-traversal guard, SKIP_PREFIXES, extension filtering, malformed config handling, missing file guard.
-
-**Approach:** use `node:test` + `node:assert`. Tests spawn the hook script as a subprocess with controlled `CLAUDE_PROJECT_DIR` and stdin JSON, then assert on exit code and stderr.
-
-## Current behaviour
-
-After spec `04`: `package.json` scripts = `{ "test": "echo 'No tests yet'" }`. No `tests/` directory.
-
-## Target behaviour
-
-- `tests/format-hook.test.mjs` exists with at least 8 test cases.
-- `pnpm test` runs `node --test tests/format-hook.test.mjs` and exits 0 when all tests pass.
-- Tests cover: empty stdin, missing file, path traversal, SKIP_PREFIXES, unsupported extension, config override, malformed config.
-
-## Implementation steps
-
-### Step 1 — Create `tests/format-hook.test.mjs`
-
-```js
-import { test } from 'node:test'
-import { strict as assert } from 'node:assert'
-import { spawnSync } from 'node:child_process'
-import { mkdirSync, writeFileSync, mkdtempSync, rmSync } from 'node:fs'
-import { tmpdir } from 'node:os'
-import { join } from 'node:path'
-import { fileURLToPath } from 'node:url'
-
-const SCRIPT = fileURLToPath(new URL('../scripts/format-hook.mjs', import.meta.url))
-
-function run(stdinJson, projectDir, extraEnv = {}) {
-	const result = spawnSync(process.execPath, [SCRIPT], {
-		input: stdinJson,
-		env: { ...process.env, CLAUDE_PROJECT_DIR: projectDir, ...extraEnv },
-		encoding: 'utf8',
-	})
-	return { exitCode: result.status, stderr: result.stderr, stdout: result.stdout }
-}
-
-function makeConsumer(setupFn) {
-	const dir = mkdtempSync(join(tmpdir(), 'unic-format-test-'))
-	setupFn?.(dir)
-	return dir
-}
-
-function cleanup(dir) {
-	rmSync(dir, { recursive: true, force: true })
-}
-
-test('exits 0 with empty stdin', () => {
-	const dir = makeConsumer()
-	try {
-		const { exitCode } = run('', dir)
-		assert.equal(exitCode, 0)
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('exits 0 with empty JSON {}', () => {
-	const dir = makeConsumer()
-	try {
-		const { exitCode } = run('{}', dir)
-		assert.equal(exitCode, 0)
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('exits 0 when file does not exist', () => {
-	const dir = makeConsumer()
-	try {
-		const { exitCode } = run(JSON.stringify({ tool_input: { file_path: join(dir, 'nonexistent.md') } }), dir)
-		assert.equal(exitCode, 0)
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('exits 0 and skips _bmad/ path', () => {
-	const dir = makeConsumer((d) => {
-		mkdirSync(join(d, '_bmad'))
-		writeFileSync(join(d, '_bmad', 'test.md'), '# test\n')
-	})
-	try {
-		const { exitCode, stderr } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, '_bmad', 'test.md') } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		assert.equal(stderr, '', 'should be silent when skipping')
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('exits 0 and skips path-traversal (../)', () => {
-	const dir = makeConsumer()
-	const outsidePath = join(dir, '..', 'something.md')
-	try {
-		const { exitCode, stderr } = run(JSON.stringify({ tool_input: { file_path: outsidePath } }), dir)
-		assert.equal(exitCode, 0)
-		// Either the file doesn't exist (exits silently) or the path-traversal guard fires silently
-		assert.equal(stderr, '')
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('exits 0 and skips unsupported extension (.toml)', () => {
-	const dir = makeConsumer((d) => {
-		writeFileSync(join(d, 'config.toml'), '[tool]\nname = "test"\n')
-	})
-	try {
-		const { exitCode, stderr } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, 'config.toml') } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		assert.equal(stderr, '')
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('exits 0 with malformed config file and logs warning', () => {
-	const dir = makeConsumer((d) => {
-		mkdirSync(join(d, '.claude'))
-		writeFileSync(join(d, '.claude', 'unic-format.json'), 'NOT JSON')
-		writeFileSync(join(d, 'test.md'), '# test\n')
-	})
-	try {
-		const { exitCode, stderr } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, 'test.md') } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		assert.match(stderr, /malformed/, 'should warn about malformed config')
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('respects prettierExtensions override from config', () => {
-	const dir = makeConsumer((d) => {
-		mkdirSync(join(d, '.claude'))
-		// Only allow .md; .json should be skipped
-		writeFileSync(join(d, '.claude', 'unic-format.json'), JSON.stringify({ prettierExtensions: ['.md'] }))
-		writeFileSync(join(d, 'data.json'), '{"key":"value"}')
-	})
-	try {
-		const { exitCode, stderr } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, 'data.json') } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		assert.equal(stderr, '', 'should be silent: .json excluded by config')
-	} finally {
-		cleanup(dir)
-	}
-})
-```
-
-### Step 2 — Update `package.json` test script
-
-```json
-"scripts": {
-	"test": "node --test tests/format-hook.test.mjs",
-	"bump": "node scripts/bump.mjs",
-	"verify:changelog": "node scripts/verify-changelog.mjs"
-}
-```
-
-### Step 3 — Commit
-
-```sh
-git add tests/ package.json
-git commit -m "test(spec-09): add smoke tests for format hook routing logic"
-```
-
-## Acceptance criteria
-
-- `pnpm test` exits 0 with all 8 tests passing.
-- Tests cover: empty stdin, missing file, `_bmad/` skip, path-traversal, unsupported extension, malformed config, config override.
-- No external test framework (only `node:test` and `node:assert`).
-
-## Verification
-
-```sh
-pnpm test
-```
-
-All tests should report `✓` (or `ok` in TAP output). None should fail.
-
-## Out of scope
-
-- Integration tests (actually running Prettier on real files) — unit-level routing tests are sufficient for v1.
-- Coverage reporting.
-- Snapshot tests.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/10-ci-workflow.md b/apps/claude-code/auto-format/docs/plans/10-ci-workflow.md
deleted file mode 100644
index 480c250a..00000000
--- a/apps/claude-code/auto-format/docs/plans/10-ci-workflow.md
+++ /dev/null
@@ -1,90 +0,0 @@
-# 10. CI Workflow
-**Status: done — 2026-04-27**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-09
-**Touches:** `.github/workflows/ci.yml`
-
-## Context
-
-A CI workflow on GitHub Actions ensures that every push/PR: (a) installs deps cleanly, (b) passes the smoke tests, and (c) passes `pnpm verify:changelog`. This is the safety net that catches regressions introduced between ralph iterations.
-
-## Current behaviour
-
-After spec `09`: `pnpm test` runs the smoke tests. No `.github/` directory exists.
-
-## Target behaviour
-
-- `.github/workflows/ci.yml` exists and runs on `push` and `pull_request`.
-- Workflow steps: checkout → pnpm setup → node setup → `pnpm install --frozen-lockfile` → `pnpm test` → `pnpm verify:changelog`.
-- `pnpm/action-setup` reads `packageManager` from `package.json` (no explicit `version:` input — this is the UNIC convention per `docs/plans/README.md` ground rules).
-- Node version is read from `pnpm-workspace.yaml`'s `useNodeVersion` field via `actions/setup-node` with `node-version-file: 'pnpm-workspace.yaml'`.
-
-## Implementation steps
-
-### Step 1 — Create `.github/workflows/ci.yml`
-
-```yaml
-name: CI
-
-on:
-  push:
-  pull_request:
-
-jobs:
-  ci:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: pnpm/action-setup@v4
-        # reads packageManager from package.json — do not add version: here
-
-      - uses: actions/setup-node@v4
-        with:
-          node-version-file: "pnpm-workspace.yaml"
-          cache: "pnpm"
-
-      - name: Install dependencies
-        run: pnpm install --frozen-lockfile
-
-      - name: Run tests
-        run: pnpm test
-
-      - name: Verify CHANGELOG
-        run: pnpm verify:changelog
-```
-
-### Step 2 — Commit
-
-```sh
-git add .github/
-git commit -m "ci(spec-10): add GitHub Actions CI workflow"
-```
-
-## Acceptance criteria
-
-- `.github/workflows/ci.yml` is valid YAML.
-- `pnpm/action-setup` has no `version:` input.
-- `actions/setup-node` uses `node-version-file: "pnpm-workspace.yaml"`.
-- Three job steps after install: `pnpm test`, `pnpm verify:changelog`.
-
-## Verification
-
-```sh
-# 1. Valid YAML
-python3 -c "import yaml; yaml.safe_load(open('.github/workflows/ci.yml'))" && echo "OK: valid YAML"
-
-# 2. No version: input in pnpm/action-setup block
-grep -A5 "pnpm/action-setup" .github/workflows/ci.yml | grep "version:" && echo "FAIL: version: present" || echo "OK: no version:"
-```
-
-## Out of scope
-
-- No release workflow (plugin is distributed via git URL).
-- No matrix testing (single Node LTS version per `useNodeVersion`).
-- No code coverage upload.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/11-claude-md.md b/apps/claude-code/auto-format/docs/plans/11-claude-md.md
deleted file mode 100644
index bd97629b..00000000
--- a/apps/claude-code/auto-format/docs/plans/11-claude-md.md
+++ /dev/null
@@ -1,96 +0,0 @@
-# 11. CLAUDE.md
-**Status: done — 2026-04-27**
-
-**Priority:** P2
-**Effort:** S
-**Version impact:** patch
-**Depends on:** none
-**Touches:** `CLAUDE.md`
-
-## Context
-
-A `CLAUDE.md` at the repo root tells Claude Code how to work with this plugin repo. It documents: commands, tech stack, coding conventions, and a scope guard (do not add unrelated features).
-
-## Current behaviour
-
-No `CLAUDE.md` exists.
-
-## Target behaviour
-
-`CLAUDE.md` exists and covers: commands, tech stack, conventions, scope guard.
-
-## Implementation steps
-
-### Step 1 — Create `CLAUDE.md`
-
-```md
-# CLAUDE.md
-
-This file provides guidance to Claude Code when working in this repository.
-
-## Project Overview
-
-`unic-claude-code-format` is a Claude Code plugin that auto-formats and lints files after Write/Edit/MultiEdit/NotebookEdit events. The hook script (`scripts/format-hook.mjs`) uses only Node built-ins — no external deps.
-
-## Commands
-
-​```sh
-pnpm install              # Install dev dependencies
-pnpm test                 # Run smoke tests (node:test)
-pnpm bump patch           # Bump patch version + promote CHANGELOG
-pnpm bump minor           # Bump minor version + promote CHANGELOG
-pnpm bump major           # Bump major version + promote CHANGELOG
-pnpm verify:changelog     # Check CHANGELOG structure
-​```
-
-## Tech Stack
-
-- **Runtime**: Node.js >=24 (LTS). Version pinned via `pnpm-workspace.yaml#useNodeVersion`.
-- **Package manager**: pnpm (workspace mode, catalog pinning).
-- **Module system**: ESM (`"type": "module"`).
-- **Test runner**: `node:test` built-in. No external framework.
-- **Hook script deps**: zero external. Only `node:child_process`, `node:fs`, `node:path`.
-
-## Code Conventions
-
-- Tabs for indentation in `.js`/`.mjs` files; 2-space for `.json`, `.yml`, `.yaml` (per `.editorconfig`).
-- Conventional commits: `feat(scope): description`, `fix(scope): description`, `chore(scope): description`.
-- SemVer: `pnpm bump <patch|minor|major>`. Never hand-edit `.claude-plugin/marketplace.json`.
-- No `^` or `~` version ranges. Exact pins only (catalog for devDeps).
-
-## Scope Guard — Do Not Add
-
-- MCP servers, skills, agents, or other Claude Code plugin features not described in an existing spec.
-- External npm runtime dependencies to the hook script.
-- Prettier or ESLint bundled inside the plugin — consumers bring their own.
-- Support for tools outside the Claude Code hook contract (e.g. bash hooks, pre-commit hooks — those belong in consumer repos).
-
-## Ralph Orchestrator
-
-This repo uses ralph-orchestrator for iterative development. See `docs/plans/README.md` for the spec roadmap and `PROMPT.md` for the orchestrator loop prompt.
-```
-
-### Step 2 — Commit
-
-```sh
-git add CLAUDE.md
-git commit -m "docs(spec-11): add CLAUDE.md with project guidance and scope guard"
-```
-
-## Acceptance criteria
-
-- `CLAUDE.md` exists with Commands, Tech Stack, Conventions, and Scope Guard sections.
-- Scope Guard lists at least 3 things Claude should NOT add.
-
-## Verification
-
-```sh
-grep -q "## Scope Guard" CLAUDE.md && echo "OK: scope guard present"
-grep -q "pnpm bump" CLAUDE.md && echo "OK: bump command documented"
-```
-
-## Out of scope
-
-- No auto-sync with README (they serve different audiences).
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/12-diff-based-verify-changelog.md b/apps/claude-code/auto-format/docs/plans/12-diff-based-verify-changelog.md
deleted file mode 100644
index 0cb79e4f..00000000
--- a/apps/claude-code/auto-format/docs/plans/12-diff-based-verify-changelog.md
+++ /dev/null
@@ -1,335 +0,0 @@
-# 12. Diff-based `verify:changelog` enforcement
-**Status: done — 2026-04-27**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** minor
-**Depends on:** spec-08, spec-10
-**Touches:** `scripts/verify-changelog.mjs`
-
-## Context
-
-The current `pnpm verify:changelog` (spec-08) is a structural check: it confirms
-that CHANGELOG.md is well-formed (Unreleased section, subsections, dated releases).
-It does **not** enforce that a version bump happened.
-
-You can modify `scripts/format-hook.mjs`, push, and CI passes without bumping
-`plugin.json`. This breaks the contract stated in `docs/plans/README.md`:
-"`.claude-plugin/plugin.json` is the single source of truth for the version number."
-
-`../unic-claude-code-confluence/scripts/verify-changelog.mjs` solves this with a
-diff-based gate: when guarded paths change in a branch, the script requires that
-`plugin.json`'s version was also bumped AND that CHANGELOG.md has a real entry for
-that version. If only non-guarded files change (e.g. plain docs) the gate skips
-silently, avoiding noise.
-
-This spec layers the diff-based gate on top of the existing structural checks so
-both are enforced.
-
-## Current behaviour
-
-After spec-08 and spec-10: `pnpm verify:changelog` runs three structural checks
-(Unreleased section, Breaking/Added/Fixed subsections, dates on versioned releases).
-It exits 0 on a well-formed CHANGELOG regardless of whether guarded paths changed
-without a version bump.
-
-## Target behaviour
-
-`pnpm verify:changelog` runs two layers:
-
-**Layer 1 — Structural (unchanged):**
-- `## [Unreleased]` section must exist.
-- `## [Unreleased]` must contain `### Breaking`, `### Added`, `### Fixed`.
-- Every `## [X.Y.Z]` section must have a date suffix (` - YYYY-MM-DD`).
-
-**Layer 2 — Diff-based gate (NEW):**
-- Compute changed files since the base branch (`origin/main` in CI, `@{upstream}` or
-  `HEAD~1` locally). If `git diff` is unavailable (e.g. shallow clone), skip silently.
-- Guarded paths:
-
-  ```
-  scripts/.+\.mjs
-  tests/.+\.mjs
-  \.claude-plugin/(plugin|marketplace)\.json
-  CLAUDE\.md
-  README\.md
-  docs/plans/.+\.md
-  ```
-
-- If **no** guarded paths changed → `verify:changelog: ok (no guarded paths changed)`.
-- If guarded paths changed:
-  - Read `plugin.json#version` at HEAD and at the base commit.
-  - If versions are equal → exit 1: `"version in plugin.json was not bumped"`.
-  - If version was bumped → check that `CHANGELOG.md` has `## [<headVersion>] - YYYY-MM-DD`
-    with at least one non-`(none)` bullet.
-  - If CHANGELOG entry missing or empty → exit 1.
-  - Otherwise → `verify:changelog: ok — version X.Y.Z → A.B.C`.
-
-**Important:** the date separator in this repo is `-` (hyphen) — `## [1.0.0] - 2026-04-27`.
-The confluence reference uses `—` (em-dash). Do NOT copy the em-dash; keep the hyphen
-that `scripts/bump.mjs` already produces.
-
-## Implementation steps
-
-### Step 1 — Replace `scripts/verify-changelog.mjs`
-
-The new script adds `spawnSync` from `node:child_process` and a `GUARDED` array,
-then appends the diff-based gate after the existing structural block. Tabs for
-indentation throughout.
-
-Key logic:
-
-1. **Structural block** — identical to current `scripts/verify-changelog.mjs:1-62`
-   (no changes needed).
-
-2. **Diff-based block** — added after the structural `process.exit(1)` guard:
-
-   ```
-   git('rev-parse', '--abbrev-ref', '--symbolic-full-name', '@{upstream}')
-     → determines base ref (or HEAD~1 fallback)
-   git('diff', '--name-only', `${base}...HEAD`)
-     → lists changed files
-   if none match GUARDED → exit 0 "ok (no guarded paths changed)"
-   git('show', `${base}:.claude-plugin/plugin.json`)
-     → reads base version
-   compare headVersion vs baseVersion
-   if equal → exit 1 "version not bumped"
-   regex match CHANGELOG for `## [headVersion] - YYYY-MM-DD`
-   check at least one `- ` line that isn't `- (none)`
-   ```
-
-3. **GUARDED regexes** (tab-indented array):
-
-   ```js
-   const GUARDED = [
-     /^scripts\/.+\.mjs$/,
-     /^tests\/.+\.mjs$/,
-     /^\.claude-plugin\/(plugin|marketplace)\.json$/,
-     /^CLAUDE\.md$/,
-     /^README\.md$/,
-     /^docs\/plans\/.+\.md$/,
-   ]
-   ```
-
-4. **CHANGELOG section regex** — use hyphen, not em-dash:
-
-   ```js
-   new RegExp(
-     `## \\[${headVersion.replace(/\./g, '\\.')}\\] - \\d{4}-\\d{2}-\\d{2}([\\s\\S]*?)(?=\\n## \\[|$)`
-   )
-   ```
-
-Full replacement file (write verbatim, tabs not spaces):
-
-```
-#!/usr/bin/env node
-/**
- * pnpm verify:changelog
- * Layer 1: structural checks (Unreleased section, subsections, dated releases).
- * Layer 2: diff-based gate — when guarded paths change, version must be bumped
- *           and CHANGELOG must have a real entry for the new version.
- */
-
-import { readFileSync } from 'node:fs'
-import { resolve } from 'node:path'
-import { spawnSync } from 'node:child_process'
-
-const ROOT = new URL('..', import.meta.url).pathname.replace(/\/$/, '')
-
-/** Paths that trigger version-bump enforcement when changed */
-const GUARDED = [
-	/^scripts\/.+\.mjs$/,
-	/^tests\/.+\.mjs$/,
-	/^\.claude-plugin\/(plugin|marketplace)\.json$/,
-	/^CLAUDE\.md$/,
-	/^README\.md$/,
-	/^docs\/plans\/.+\.md$/,
-]
-
-// Layer 1: structural checks
-
-const changelogPath = resolve(ROOT, 'CHANGELOG.md')
-let changelog
-try {
-	changelog = readFileSync(changelogPath, 'utf8')
-} catch {
-	process.stderr.write('verify:changelog: CHANGELOG.md not found.\n')
-	process.exit(1)
-}
-
-const errors = []
-
-if (!changelog.includes('## [Unreleased]')) {
-	errors.push('Missing ## [Unreleased] section.')
-}
-
-const unreleasedIdx = changelog.indexOf('## [Unreleased]')
-if (unreleasedIdx !== -1) {
-	const nextReleaseIdx = changelog.indexOf('\n## [', unreleasedIdx + 1)
-	const unreleasedBlock =
-		nextReleaseIdx === -1
-			? changelog.slice(unreleasedIdx)
-			: changelog.slice(unreleasedIdx, nextReleaseIdx)
-	for (const sub of ['### Breaking', '### Added', '### Fixed']) {
-		if (!unreleasedBlock.includes(sub)) {
-			errors.push(`[Unreleased] is missing subsection: ${sub}`)
-		}
-	}
-}
-
-const releasePattern = /^## \[(\d+\.\d+\.\d+)\]/gm
-let match
-while ((match = releasePattern.exec(changelog)) !== null) {
-	const end = changelog.indexOf('\n', match.index)
-	const line = changelog.slice(match.index, end === -1 ? undefined : end)
-	if (!/ - \d{4}-\d{2}-\d{2}/.test(line)) {
-		errors.push(`Release section missing date: ${line.trim()}`)
-	}
-}
-
-if (errors.length > 0) {
-	process.stderr.write('verify:changelog failed:\n')
-	errors.forEach((e) => process.stderr.write(`  - ${e}\n`))
-	process.exit(1)
-}
-
-// Layer 2: diff-based version-bump gate
-
-function git(...args) {
-	const result = spawnSync('git', args, { encoding: 'utf8', cwd: ROOT })
-	return { stdout: result.stdout ?? '', status: result.status ?? 1 }
-}
-
-const isCI = process.env.CI === 'true'
-let base = 'origin/main'
-if (!isCI) {
-	const upstream = git('rev-parse', '--abbrev-ref', '--symbolic-full-name', '@{upstream}')
-	base = upstream.status === 0 ? upstream.stdout.trim() : 'HEAD~1'
-}
-
-const diff = git('diff', '--name-only', `${base}...HEAD`)
-if (diff.status !== 0) {
-	process.stdout.write('verify:changelog: skipped (git diff unavailable)\n')
-	process.exit(0)
-}
-
-const changedFiles = diff.stdout.trim().split('\n').filter(Boolean)
-const triggered = changedFiles.some((f) => GUARDED.some((re) => re.test(f)))
-if (!triggered) {
-	process.stdout.write('verify:changelog: ok (no guarded paths changed)\n')
-	process.exit(0)
-}
-
-const pluginPath = resolve(ROOT, '.claude-plugin/plugin.json')
-let headPlugin
-try {
-	headPlugin = JSON.parse(readFileSync(pluginPath, 'utf8'))
-} catch {
-	process.stderr.write(`verify:changelog: cannot read ${pluginPath}\n`)
-	process.exit(1)
-}
-const headVersion = headPlugin.version
-
-const basePluginRaw = git('show', `${base}:.claude-plugin/plugin.json`)
-let baseVersion = ''
-if (basePluginRaw.status === 0) {
-	try {
-		baseVersion = JSON.parse(basePluginRaw.stdout).version
-	} catch {
-		// base does not have plugin.json yet — treat as empty
-	}
-}
-
-if (headVersion === baseVersion) {
-	process.stderr.write(
-		`verify:changelog: version in plugin.json was not bumped\n` +
-			`  current: ${headVersion} (same as ${base})\n` +
-			`  Run: pnpm bump <patch|minor|major>\n`,
-	)
-	process.exit(1)
-}
-
-const sectionMatch = changelog.match(
-	new RegExp(
-		`## \\[${headVersion.replace(/\./g, '\\.')}\\] - \\d{4}-\\d{2}-\\d{2}([\\s\\S]*?)(?=\\n## \\[|$)`,
-	),
-)
-if (!sectionMatch) {
-	process.stderr.write(
-		`verify:changelog: CHANGELOG.md has no entry for version ${headVersion}\n` +
-			`  Add bullets under [Unreleased] then run: pnpm bump\n`,
-	)
-	process.exit(1)
-}
-
-const sectionBody = sectionMatch[1]
-const hasRealEntry = sectionBody
-	.split('\n')
-	.filter((l) => l.startsWith('- '))
-	.some((l) => l !== '- (none)')
-if (!hasRealEntry) {
-	process.stderr.write(
-		`verify:changelog: CHANGELOG.md section [${headVersion}] has no real entries\n` +
-			`  Add bullets under [Unreleased] then re-run: pnpm bump\n`,
-	)
-	process.exit(1)
-}
-
-process.stdout.write(`verify:changelog: ok — version ${baseVersion} → ${headVersion}\n`)
-```
-
-### Step 2 — Commit
-
-```sh
-git add scripts/verify-changelog.mjs
-git commit -m "feat(spec-12): add diff-based verify:changelog version-bump gate"
-```
-
-## Test cases
-
-| Scenario | Expected |
-|---|---|
-| Current repo HEAD, already on `main` | Layer 1 passes; layer 2: "ok (no guarded paths changed)" |
-| Guarded path changed, version bumped, CHANGELOG entry present | "ok — version X → Y" |
-| Guarded path changed, version NOT bumped | Exits 1: "version in plugin.json was not bumped" |
-| Guarded path changed, version bumped, CHANGELOG entry missing | Exits 1: "has no entry for version Y" |
-| Guarded path changed, version bumped, CHANGELOG entry only `(none)` | Exits 1: "no real entries" |
-| Only non-guarded file changed (e.g. `.gitignore`) | Layer 1 passes; layer 2: "ok (no guarded paths changed)" |
-| Malformed CHANGELOG (missing `[Unreleased]`) | Exits 1 from layer 1 |
-| git diff unavailable (shallow clone) | "skipped (git diff unavailable)" |
-
-## Acceptance criteria
-
-- `pnpm verify:changelog` passes on the current clean repo.
-- `pnpm verify:changelog` exits 1 when guarded files changed but `plugin.json` version is unchanged.
-- `pnpm verify:changelog` exits 1 when CHANGELOG entry is missing or placeholder-only.
-- `pnpm verify:changelog` exits 0 with "no guarded paths changed" when only non-guarded files differ.
-- Structural failures (missing `[Unreleased]`) still exit 1 regardless of diff state.
-- The CHANGELOG section regex uses `-` (hyphen) not `—` (em-dash).
-
-## Verification
-
-```sh
-# 1. Passes on current repo
-pnpm verify:changelog
-
-# 2. Simulate guarded change without version bump (undo after)
-git checkout -b test-verify-gate
-echo "// test" >> scripts/format-hook.mjs
-git add scripts/format-hook.mjs
-git commit -m "test: trigger gate"
-pnpm verify:changelog   # expect exit 1 "version not bumped"
-git checkout main
-git branch -D test-verify-gate
-
-# 3. Structural check still fires independently
-# Temporarily corrupt CHANGELOG, run pnpm verify:changelog — should exit 1 from layer 1
-```
-
-## Out of scope
-
-- Auto-fixing the CHANGELOG or bumping automatically.
-- Enforcing version bumps on non-guarded paths.
-- Changing the CHANGELOG date separator (this repo uses `-`, keep it).
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/13-version-source-of-truth-and-sync.md b/apps/claude-code/auto-format/docs/plans/13-version-source-of-truth-and-sync.md
deleted file mode 100644
index 1dd2973e..00000000
--- a/apps/claude-code/auto-format/docs/plans/13-version-source-of-truth-and-sync.md
+++ /dev/null
@@ -1,303 +0,0 @@
-# 13. Single source of truth + `sync-version.mjs`
-**Status: done — 2026-04-27**
-
-**Priority:** P1
-**Effort:** M
-**Version impact:** patch
-**Depends on:** spec-07
-**Touches:** `scripts/sync-version.mjs` (new), `scripts/bump.mjs` (refactor), `package.json` (new script entry), `CLAUDE.md` (new command reference)
-
-## Context
-
-`docs/plans/README.md` ground rule:
-> "`.claude-plugin/plugin.json` is the single source of truth for the version number.
-> **Never hand-edit** `.claude-plugin/marketplace.json`."
-
-In practice, `scripts/bump.mjs` writes the new version into `plugin.json`,
-`marketplace.json`, AND `package.json` independently, in three parallel writes.
-If any one of them diverges (manual edit, failed partial run), the trio is out
-of sync with no recovery path.
-
-`../unic-claude-code-confluence/scripts/sync-version.mjs` solves this by having a
-single idempotent script that reads `plugin.json` and propagates to derived files.
-`bump-version.mjs` only writes `plugin.json` and then delegates to `sync-version.mjs`.
-
-This spec ports that pattern here. Because `package.json` carries a `version` field
-in this repo (it is `private: true` but `pnpm` reads it), `sync-version.mjs` must
-also propagate to `package.json` — a deliberate deviation from the confluence version.
-
-The confluence repo uses two-space JSON indentation (Biome-formatted). This repo uses
-tabs. **Keep tabs.** The deviation from CLAUDE.md ("2-space for .json") is documented
-below under Deviations; fixing it is out of scope here.
-
-## Current behaviour
-
-After spec-07: `scripts/bump.mjs` writes `plugin.json`, `marketplace.json`, and
-`package.json` in three separate `writeJson` calls within the same script. No
-`sync-version.mjs` exists.
-
-## Target behaviour
-
-- `scripts/sync-version.mjs` exists.
-  - Reads `.claude-plugin/plugin.json#version` (the single source of truth).
-  - Writes that version into `.claude-plugin/marketplace.json#version`.
-  - Writes that version into `package.json#version`.
-  - Is idempotent: if a file's version already matches, logs "no change" and
-    does not write.
-  - `pnpm sync-version` invokes it (new `package.json` script entry).
-- `scripts/bump.mjs` is refactored:
-  - Still validates `[Unreleased]` first.
-  - Computes the new version from `plugin.json`.
-  - Writes **only** `plugin.json`.
-  - Spawns `node scripts/sync-version.mjs` (inherit stdio); aborts on non-zero.
-  - Promotes the CHANGELOG.
-  - Final output: `Bumped to v<newVersion>`.
-- Net result for `pnpm bump patch`: all three files updated to the same version
-  (same as before), but via sync-version rather than three independent writes.
-
-## Implementation steps
-
-### Step 1 — Create `scripts/sync-version.mjs`
-
-```js
-#!/usr/bin/env node
-/**
- * pnpm sync-version
- * Reads the version from .claude-plugin/plugin.json (single source of truth)
- * and propagates it to .claude-plugin/marketplace.json and package.json.
- * Idempotent — safe to run multiple times.
- */
-
-import { readFileSync, writeFileSync } from 'node:fs'
-import { resolve } from 'node:path'
-
-const ROOT = new URL('..', import.meta.url).pathname.replace(/\/$/, '')
-
-const pluginPath = resolve(ROOT, '.claude-plugin/plugin.json')
-let pluginJson
-try {
-	pluginJson = JSON.parse(readFileSync(pluginPath, 'utf8'))
-} catch (err) {
-	process.stderr.write(`sync-version: cannot read ${pluginPath}: ${err.message}\n`)
-	process.exit(1)
-}
-
-const version = pluginJson.version
-if (!version || typeof version !== 'string') {
-	process.stderr.write(`sync-version: .version is missing or not a string in ${pluginPath}\n`)
-	process.exit(1)
-}
-
-/**
- * Update version in a JSON file. Logs the transition (or no-op).
- * @param {string} filePath  Absolute path to the JSON file.
- */
-function syncFile(filePath) {
-	let obj
-	try {
-		obj = JSON.parse(readFileSync(filePath, 'utf8'))
-	} catch (err) {
-		process.stderr.write(`sync-version: cannot read ${filePath}: ${err.message}\n`)
-		process.exit(1)
-	}
-	const rel = filePath.slice(ROOT.length + 1)
-	const prev = obj.version
-	if (prev === version) {
-		process.stdout.write(`sync-version: ${rel} already at ${version} (no change)\n`)
-		return
-	}
-	obj.version = version
-	writeFileSync(filePath, JSON.stringify(obj, null, '\t') + '\n', 'utf8')
-	process.stdout.write(`sync-version: ${rel} updated ${prev} → ${version}\n`)
-}
-
-syncFile(resolve(ROOT, '.claude-plugin/marketplace.json'))
-syncFile(resolve(ROOT, 'package.json'))
-```
-
-### Step 2 — Refactor `scripts/bump.mjs`
-
-Replace `scripts/bump.mjs` with the version below. The only structural change is:
-- Remove the `writeJson` calls for `marketplace.json` and `package.json`.
-- Replace with a `spawnSync` call to `sync-version.mjs`.
-
-```js
-#!/usr/bin/env node
-/**
- * pnpm bump <patch|minor|major>
- * Bumps version in plugin.json (single source of truth), syncs to
- * marketplace.json + package.json via sync-version.mjs, and promotes
- * [Unreleased] in CHANGELOG.md.
- */
-
-import { readFileSync, writeFileSync } from 'node:fs'
-import { resolve } from 'node:path'
-import { spawnSync } from 'node:child_process'
-
-const ROOT = new URL('..', import.meta.url).pathname.replace(/\/$/, '')
-
-function readJson(rel) {
-	return JSON.parse(readFileSync(resolve(ROOT, rel), 'utf8'))
-}
-
-function writeJson(rel, obj) {
-	writeFileSync(resolve(ROOT, rel), JSON.stringify(obj, null, '\t') + '\n', 'utf8')
-}
-
-function bumpVersion(version, type) {
-	const [major, minor, patch] = version.split('.').map(Number)
-	if (type === 'major') return `${major + 1}.0.0`
-	if (type === 'minor') return `${major}.${minor + 1}.0`
-	if (type === 'patch') return `${major}.${minor}.${patch + 1}`
-	throw new Error(`Invalid bump type: ${type}`)
-}
-
-function today() {
-	return new Date().toISOString().slice(0, 10)
-}
-
-const bumpType = process.argv[2]
-if (!['patch', 'minor', 'major'].includes(bumpType)) {
-	process.stderr.write(`Usage: pnpm bump <patch|minor|major>\n`)
-	process.exit(1)
-}
-
-// Validate CHANGELOG before touching any files
-const changelogPath = resolve(ROOT, 'CHANGELOG.md')
-let changelog = readFileSync(changelogPath, 'utf8')
-
-const unreleasedMatch = changelog.match(/## \[Unreleased\]([\s\S]*?)(?=## \[|$)/)
-if (!unreleasedMatch) {
-	process.stderr.write('CHANGELOG.md does not have an [Unreleased] section.\n')
-	process.exit(1)
-}
-
-const unreleasedBody = unreleasedMatch[1]
-const hasEntries = unreleasedBody.split('\n').some((line) => {
-	const trimmed = line.trim()
-	return trimmed.startsWith('-') && !trimmed.includes('(none)')
-})
-
-if (!hasEntries) {
-	process.stderr.write(`[Unreleased] has no entries. Add CHANGELOG entries before bumping.\n`)
-	process.exit(1)
-}
-
-// Bump plugin.json (single source of truth)
-const pluginJson = readJson('.claude-plugin/plugin.json')
-const newVersion = bumpVersion(pluginJson.version, bumpType)
-pluginJson.version = newVersion
-writeJson('.claude-plugin/plugin.json', pluginJson)
-
-// Propagate to marketplace.json and package.json via sync-version
-const syncResult = spawnSync('node', [resolve(ROOT, 'scripts/sync-version.mjs')], {
-	stdio: 'inherit',
-})
-if (syncResult.status !== 0) {
-	process.stderr.write(`bump: sync-version failed (exit ${syncResult.status ?? 'unknown'})\n`)
-	process.exit(1)
-}
-
-// Promote CHANGELOG
-const newUnreleased = `## [Unreleased]\n\n### Breaking\n- (none)\n\n### Added\n- (none)\n\n### Fixed\n- (none)\n\n`
-const newRelease = `## [${newVersion}] - ${today()}${unreleasedBody}`
-changelog = changelog.replace(/## \[Unreleased\][\s\S]*?(?=## \[|$)/, newUnreleased + newRelease)
-writeFileSync(changelogPath, changelog, 'utf8')
-
-process.stdout.write(`Bumped to v${newVersion}\n`)
-```
-
-### Step 3 — Add `sync-version` to `package.json` scripts
-
-```json
-"scripts": {
-	"test": "node --test tests/format-hook.test.mjs",
-	"bump": "node scripts/bump.mjs",
-	"sync-version": "node scripts/sync-version.mjs",
-	"verify:changelog": "node scripts/verify-changelog.mjs"
-}
-```
-
-### Step 4 — Update `CLAUDE.md` commands section
-
-Add `pnpm sync-version` after `pnpm bump major`:
-
-```md
-pnpm sync-version         # Propagate plugin.json version to marketplace.json + package.json
-```
-
-### Step 5 — Commit
-
-```sh
-git add scripts/sync-version.mjs scripts/bump.mjs package.json CLAUDE.md
-git commit -m "refactor(spec-13): make plugin.json the single source of truth via sync-version.mjs"
-```
-
-## Test cases
-
-| Command | Expected |
-|---|---|
-| `pnpm sync-version` (all files in sync) | "no change" for both derived files |
-| Hand-edit `plugin.json` version, then `pnpm sync-version` | marketplace.json and package.json updated to match |
-| `pnpm bump patch` with entries | All three files at new version; CHANGELOG promoted |
-| `pnpm bump patch` without entries | Exits 1 "no entries" (unchanged from before) |
-| `pnpm sync-version` when `marketplace.json` unreadable | Exits 1 with descriptive message |
-
-## Acceptance criteria
-
-- `scripts/sync-version.mjs` exists and is valid ESM.
-- `pnpm sync-version` exits 0 and prints a summary for each file.
-- `pnpm sync-version` is idempotent (second run prints "no change").
-- `pnpm bump patch` (with CHANGELOG entries) updates all three version files and
-  promotes CHANGELOG — same observable output as before.
-- `scripts/bump.mjs` no longer contains `writeJson` calls for `marketplace.json`
-  or `package.json`.
-
-## Verification
-
-```sh
-# 1. Idempotent run
-pnpm sync-version
-# expect: both files report "no change"
-
-# 2. Force a desync, then repair
-node -e "
-  const fs = require('fs')
-  const p = fs.readFileSync('.claude-plugin/marketplace.json','utf8')
-  const j = JSON.parse(p); j.version='0.0.0'
-  fs.writeFileSync('.claude-plugin/marketplace.json', JSON.stringify(j,null,'\\t')+'\\n')
-"
-pnpm sync-version
-# expect: marketplace.json updated X → current; package.json no change
-git restore .claude-plugin/marketplace.json
-
-# 3. Full bump (undo with git restore afterward)
-# Add a bullet to [Unreleased]/### Fixed in CHANGELOG.md, then:
-pnpm bump patch
-# verify all three files share the same version:
-node -e "
-  const p = require('./.claude-plugin/plugin.json').version
-  const m = require('./.claude-plugin/marketplace.json').version
-  const k = require('./package.json').version
-  if(p===m&&m===k) console.log('PASS',p); else console.error('MISMATCH',{p,m,k})
-"
-git restore .
-```
-
-## Deviations to document
-
-- `sync-version.mjs` also updates `package.json#version`. The confluence reference
-  does not do this because their `package.json` carries no version. This repo's
-  `package.json` has `"version"` (legacy from spec-07's spec code), so we keep it
-  synced to avoid pnpm warnings.
-- CLAUDE.md says "2-space for .json" but the repo's JSON files already use tabs.
-  This discrepancy predates this spec. Document it here; do not fix it under this
-  spec (separate cleanup decision needed).
-
-## Out of scope
-
-- Changing JSON indentation (tabs → 2-space or vice versa).
-- `pnpm tag` (spec-14).
-- Removing `package.json#version` (would be a separate decision).
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/14-pnpm-tag.md b/apps/claude-code/auto-format/docs/plans/14-pnpm-tag.md
deleted file mode 100644
index 07043941..00000000
--- a/apps/claude-code/auto-format/docs/plans/14-pnpm-tag.md
+++ /dev/null
@@ -1,169 +0,0 @@
-# 14. `pnpm tag` for git tagging
-**Status: done — 2026-04-27**
-
-**Priority:** P2
-**Effort:** S
-**Version impact:** minor
-**Depends on:** spec-13
-**Touches:** `scripts/tag.mjs` (new), `package.json`, `docs/plans/README.md`, `CLAUDE.md`
-
-## Context
-
-Spec-07 explicitly listed `pnpm tag` as out of scope: "not needed for development
-phase." That decision made sense when the plugin was being built from scratch; the
-priority was to land the core hook and tooling without scope creep.
-
-The situation has changed: you plan to install `unic-claude-code-format` into client
-projects (starting with `unic-claude-code-confluence`) via a git URL. Consumer repos
-that pin to a specific commit or tag benefit from tagged releases — they can reference
-`git+https://...#v1.2.3` instead of a bare commit SHA, which makes upgrades
-traceable and changelog-readable.
-
-`../unic-claude-code-confluence/scripts/tag.mjs` already does exactly this. This spec
-ports it, relying on `sync-version.mjs` (spec-13) for the idempotent safety sync
-before tagging.
-
-## Current behaviour
-
-After spec-13: `pnpm bump`, `pnpm sync-version`, and `pnpm verify:changelog` exist.
-No `pnpm tag`. To release, you would manually run `git tag v<version>`.
-
-## Target behaviour
-
-- `scripts/tag.mjs` exists.
-- `pnpm tag` runs it.
-- Behaviour:
-  1. Reads `plugin.json#version`.
-  2. Runs `scripts/sync-version.mjs` (idempotent; ensures derived files are in sync
-     before the commit is tagged).
-  3. Runs `git tag v<version>`.
-  4. If the tag already exists, `git tag` exits non-zero → script exits 1 with a
-     clear message.
-  5. On success, prints: `Tagged v<version>. Run: git push --follow-tags`
-- Does **not** push. That remains a manual step.
-
-## Implementation steps
-
-### Step 1 — Create `scripts/tag.mjs`
-
-```js
-#!/usr/bin/env node
-/**
- * pnpm tag
- * Tags the current commit with v<version> from .claude-plugin/plugin.json.
- * Runs sync-version first (idempotent safety check).
- * Does not push — run: git push --follow-tags
- */
-
-import { readFileSync } from 'node:fs'
-import { resolve } from 'node:path'
-import { spawnSync } from 'node:child_process'
-
-const ROOT = new URL('..', import.meta.url).pathname.replace(/\/$/, '')
-
-const pluginPath = resolve(ROOT, '.claude-plugin/plugin.json')
-let pluginJson
-try {
-	pluginJson = JSON.parse(readFileSync(pluginPath, 'utf8'))
-} catch (err) {
-	process.stderr.write(`tag: cannot read ${pluginPath}: ${err.message}\n`)
-	process.exit(1)
-}
-
-const version = pluginJson.version
-if (!version || typeof version !== 'string') {
-	process.stderr.write(`tag: .version is missing or not a string in ${pluginPath}\n`)
-	process.exit(1)
-}
-
-// Safety sync before tagging (idempotent)
-spawnSync('node', [resolve(ROOT, 'scripts/sync-version.mjs')], { stdio: 'inherit' })
-
-const tagResult = spawnSync('git', ['tag', `v${version}`], { stdio: 'inherit' })
-if (tagResult.status !== 0) {
-	process.stderr.write(`tag: git tag failed — v${version} may already exist\n`)
-	process.exit(1)
-}
-
-process.stdout.write(`Tagged v${version}. Run: git push --follow-tags\n`)
-```
-
-### Step 2 — Add `tag` to `package.json` scripts
-
-```json
-"scripts": {
-	"test": "node --test tests/format-hook.test.mjs",
-	"bump": "node scripts/bump.mjs",
-	"sync-version": "node scripts/sync-version.mjs",
-	"tag": "node scripts/tag.mjs",
-	"verify:changelog": "node scripts/verify-changelog.mjs"
-}
-```
-
-### Step 3 — Update `docs/plans/README.md`
-
-Add a versioning workflow note to the ground rules section:
-
-> **Release workflow**: `pnpm bump <patch|minor|major>` → commit → `pnpm tag` → `git push --follow-tags`
-
-Update the execution table to include spec 14.
-
-### Step 4 — Update `CLAUDE.md` commands section
-
-Add `pnpm tag` after `pnpm sync-version`:
-
-```md
-pnpm tag                  # Create git tag v<version> (run git push --follow-tags after)
-```
-
-### Step 5 — Commit
-
-```sh
-git add scripts/tag.mjs package.json docs/plans/README.md CLAUDE.md
-git commit -m "feat(spec-14): add pnpm tag for creating git version tags"
-```
-
-## Test cases
-
-| Command | Expected |
-|---|---|
-| `pnpm tag` (first time for current version) | Exits 0; `vX.Y.Z` tag created; sync-version output visible |
-| `pnpm tag` (same version again) | Exits 1: "git tag failed — v… may already exist" |
-| `pnpm tag` when `plugin.json` unreadable | Exits 1 with descriptive message |
-| `pnpm tag` when `plugin.json#version` is missing | Exits 1: ".version is missing or not a string" |
-
-## Acceptance criteria
-
-- `scripts/tag.mjs` exists and is valid ESM.
-- `pnpm tag` creates a `vX.Y.Z` lightweight tag on the current commit.
-- Second `pnpm tag` at the same version exits 1.
-- `pnpm tag` does NOT push.
-- Script output includes "Run: git push --follow-tags".
-
-## Verification
-
-```sh
-# 1. Create the tag
-pnpm tag
-# expect: "Tagged v0.X.Y. Run: git push --follow-tags"
-
-# 2. Verify tag exists
-git tag --list "v$(node -p "require('./.claude-plugin/plugin.json').version")"
-# expect: the tag is printed
-
-# 3. Second run fails cleanly
-pnpm tag
-# expect: exit 1 "may already exist"
-
-# 4. Clean up
-git tag -d "v$(node -p "require('./.claude-plugin/plugin.json').version")"
-```
-
-## Out of scope
-
-- Annotated tags (`git tag -a`). Lightweight tags are sufficient for git-URL pinning.
-- Pushing (always manual: `git push --follow-tags`).
-- Release notes generation.
-- CI automation for tagging.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/15-hoist-extension-sets.md b/apps/claude-code/auto-format/docs/plans/15-hoist-extension-sets.md
deleted file mode 100644
index 290a3148..00000000
--- a/apps/claude-code/auto-format/docs/plans/15-hoist-extension-sets.md
+++ /dev/null
@@ -1,108 +0,0 @@
-# 15. Hoist extension Sets out of `main()`
-**Status: done — 2026-04-28**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-04
-**Touches:** `scripts/format-hook.mjs`
-
-## Context
-
-`main()` in `format-hook.mjs` creates two `new Set(...)` objects on every invocation:
-
-```js
-if (!new Set(CONFIG.prettierExtensions).has(ext)) return
-// ...
-if (new Set(CONFIG.eslintExtensions).has(ext)) runEslint(filePath)
-```
-
-`CONFIG` is already a module-level constant resolved once at startup. The Sets are therefore stable for the lifetime of the process and can be hoisted alongside `CONFIG`. Keeping them inside `main()` allocates and GC-collects two transient objects on every hook fire, and forces a reader to mentally re-derive that they are always equivalent to `CONFIG.*`.
-
-No behaviour changes — only where the constants are declared.
-
-## Current behaviour
-
-After spec-04: `scripts/format-hook.mjs` creates `new Set(CONFIG.prettierExtensions)` and `new Set(CONFIG.eslintExtensions)` inside `main()`.
-
-## Target behaviour
-
-`PRETTIER_EXTS` and `ESLINT_EXTS` are module-level `Set` constants declared immediately after `CONFIG`, reused by `main()`.
-
-```js
-const CONFIG = loadProjectConfig()
-
-const PRETTIER_EXTS = new Set(CONFIG.prettierExtensions)
-const ESLINT_EXTS = new Set(CONFIG.eslintExtensions)
-
-// … PRETTIER_BIN, ESLINT_BIN …
-
-async function main() {
-	// …
-	if (!PRETTIER_EXTS.has(ext)) return
-	runPrettier(filePath)
-	if (ESLINT_EXTS.has(ext)) runEslint(filePath)
-}
-```
-
-## Implementation steps
-
-### Step 1 — Hoist the Sets in `scripts/format-hook.mjs`
-
-Replace the two in-function `new Set(...)` references and add module-level constants.
-
-Exact diff (tabs throughout):
-
-1. After line `const CONFIG = loadProjectConfig()` add two lines:
-   ```js
-   const PRETTIER_EXTS = new Set(CONFIG.prettierExtensions)
-   const ESLINT_EXTS = new Set(CONFIG.eslintExtensions)
-   ```
-
-2. Inside `main()` replace:
-   ```js
-   	if (!new Set(CONFIG.prettierExtensions).has(ext)) return
-   
-   	runPrettier(filePath)
-   	if (new Set(CONFIG.eslintExtensions).has(ext)) runEslint(filePath)
-   ```
-   with:
-   ```js
-   	if (!PRETTIER_EXTS.has(ext)) return
-   
-   	runPrettier(filePath)
-   	if (ESLINT_EXTS.has(ext)) runEslint(filePath)
-   ```
-
-### Step 2 — Commit
-
-```sh
-git add scripts/format-hook.mjs
-git commit -m "refactor(spec-15): hoist extension Sets to module scope"
-```
-
-## Acceptance criteria
-
-- `PRETTIER_EXTS` and `ESLINT_EXTS` are declared at module scope, not inside `main()`.
-- No `new Set(CONFIG.*)` expressions remain inside `main()`.
-- `pnpm test` still passes (all 8 smoke tests green).
-- No external deps added.
-
-## Verification
-
-```sh
-# 1. No in-function Set construction remains
-grep -n "new Set" scripts/format-hook.mjs
-# Expected: two lines, both outside main()
-
-# 2. Tests still pass
-pnpm test
-```
-
-## Out of scope
-
-- Changing the logic of skip-prefix checks.
-- Adding new extensions to either set.
-- Merging with per-project config changes (spec-19 handles that).
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/16-windows-path-normalisation.md b/apps/claude-code/auto-format/docs/plans/16-windows-path-normalisation.md
deleted file mode 100644
index b87fbaea..00000000
--- a/apps/claude-code/auto-format/docs/plans/16-windows-path-normalisation.md
+++ /dev/null
@@ -1,166 +0,0 @@
-# 16. Windows path-separator normalisation in `shouldSkip`
-**Status: done — 2026-04-28**
-
-**Priority:** P0
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-03
-**Touches:** `scripts/format-hook.mjs`, `tests/format-hook.test.mjs`
-
-## Context
-
-`shouldSkip` in `format-hook.mjs` compares paths against forward-slash prefixes like `_bmad/`, `node_modules/`, `.git/`. The comparison works because `path.relative()` on macOS and Linux always returns forward-slash paths.
-
-On Windows, `path.relative()` returns backslash-separated paths (`_bmad\foo.md`). Every `SKIP_PREFIXES` check that uses `/` will silently fail to match, meaning files inside `node_modules`, `.git`, `_bmad`, and `.claude/worktrees` get reformatted when Claude runs on a Windows machine.
-
-This is a real-world concern: a significant share of consumers are Product Owners who write specs and run Claude Code on Windows. The hook silently misbehaves for them today.
-
-The fix is a one-line normalisation applied to `rel` before any prefix comparison. Extracting `shouldSkip` into a pure, parameterised function also makes it testable with synthetic backslash-style inputs on macOS CI.
-
-## Current behaviour
-
-After spec-03 and spec-04: `scripts/format-hook.mjs` computes `rel` with `path.relative()` and immediately passes it to `shouldSkip(rel)` and `extname(rel)`. No separator normalisation is applied. On Windows this causes all skip-prefix guards to silently miss.
-
-## Target behaviour
-
-- A pure helper `toPosix(p)` converts any backslashes to forward slashes.
-- `shouldSkip` is renamed to `shouldSkip(posixRel)` and receives an already-normalised string.
-- `main()` applies `toPosix` to `rel` exactly once, immediately after `relative()`, before any guard.
-- `extname` receives the same normalised path.
-- All existing skip-prefix tests remain green.
-- Three new tests exercise synthetic Windows-style paths and verify they are skipped.
-
-## Implementation steps
-
-### Step 1 — Add `toPosix` helper and update `shouldSkip` in `scripts/format-hook.mjs`
-
-1. Add the import `import { sep } from 'node:path'` to the existing import line:
-   ```js
-   import { extname, relative, resolve, sep } from 'node:path'
-   ```
-
-2. Add `toPosix` immediately before `shouldSkip`:
-   ```js
-   function toPosix(p) {
-   	return sep === '/' ? p : p.split(sep).join('/')
-   }
-   ```
-   The `sep` guard makes it a no-op on POSIX so there is zero overhead in production.
-
-3. Inside `main()`, replace:
-   ```js
-   	const rel = relative(PROJECT_DIR, filePath)
-   	if (shouldSkip(rel)) return
-
-   	const ext = extname(rel).toLowerCase()
-   ```
-   with:
-   ```js
-   	const rel = toPosix(relative(PROJECT_DIR, filePath))
-   	if (shouldSkip(rel)) return
-
-   	const ext = extname(rel).toLowerCase()
-   ```
-   (`extname` handles forward-slash paths correctly on all platforms; no additional change needed.)
-
-4. `shouldSkip` signature and body are unchanged — it already receives `rel` and works on forward-slash strings.
-
-### Step 2 — Add Windows-path smoke tests in `tests/format-hook.test.mjs`
-
-Since `relative()` on macOS never produces backslashes, tests cannot rely on it to exercise this path. Instead, construct `file_path` values that already contain backslashes, then verify the hook either skips silently or handles them correctly.
-
-Add three tests:
-
-```js
-test('exits 0 and skips Windows-style _bmad\\ path', () => {
-	const dir = makeConsumer((d) => {
-		mkdirSync(join(d, '_bmad'))
-		writeFileSync(join(d, '_bmad', 'test.md'), '# test\n')
-	})
-	try {
-		// Simulate a Windows absolute path: replace the forward-slash separator
-		// with backslash inside the subdir portion. After toPosix(), this should
-		// still resolve to a _bmad/ prefix and be skipped.
-		const winStylePath = join(dir, '_bmad', 'test.md').replace(/_bmad\//, '_bmad\\')
-		const { exitCode, stderr } = run(
-			JSON.stringify({ tool_input: { file_path: winStylePath } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		assert.equal(stderr, '', 'should be silent when skipping Windows-style path')
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('exits 0 and skips Windows-style node_modules\\ path', () => {
-	const dir = makeConsumer((d) => {
-		mkdirSync(join(d, 'node_modules', 'foo'), { recursive: true })
-		writeFileSync(join(d, 'node_modules', 'foo', 'index.md'), '# test\n')
-	})
-	try {
-		const winStylePath = join(dir, 'node_modules', 'foo', 'index.md').replace(/node_modules\//, 'node_modules\\')
-		const { exitCode, stderr } = run(
-			JSON.stringify({ tool_input: { file_path: winStylePath } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		assert.equal(stderr, '')
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('exits 0 with mixed-separator traversal path (..\\\\..\\\\)', () => {
-	const dir = makeConsumer()
-	const mixedTraversal = join(dir, 'sub', '..', '..', 'outside.md').replace(/\//g, '\\')
-	try {
-		const { exitCode, stderr } = run(
-			JSON.stringify({ tool_input: { file_path: mixedTraversal } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		// Either the file doesn't exist or path-traversal guard fires — both exit 0 silently
-		assert.equal(stderr, '')
-	} finally {
-		cleanup(dir)
-	}
-})
-```
-
-### Step 3 — Commit
-
-```sh
-git add scripts/format-hook.mjs tests/format-hook.test.mjs
-git commit -m "fix(spec-16): normalise path separators for Windows compatibility in shouldSkip"
-```
-
-## Acceptance criteria
-
-- `toPosix` helper exists in `scripts/format-hook.mjs`, applied to `rel` before any guard.
-- `sep` is imported from `node:path`.
-- `toPosix` is a no-op when `sep === '/'` (POSIX systems).
-- All original 8 smoke tests continue to pass.
-- 3 new Windows-path tests pass on macOS CI (they exercise the normalisation logic even though `relative()` on macOS never produces native backslashes).
-- No external deps added.
-
-## Verification
-
-```sh
-# 1. toPosix helper exists and sep is imported
-grep "toPosix\|from 'node:path'" scripts/format-hook.mjs
-
-# 2. toPosix applied before shouldSkip in main()
-grep -A2 "const rel = " scripts/format-hook.mjs
-
-# 3. All tests pass
-pnpm test
-```
-
-## Out of scope
-
-- Handling Windows drive-letter prefixes (`C:\`) in the path-traversal guard (the `..` guard covers the practical traversal cases; drive-letter edge cases are rare in Claude Code context).
-- Changes to `SKIP_PREFIXES` content.
-- Any change to how `CLAUDE_PROJECT_DIR` is set on Windows.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/17-spawnsync-timeout-guard.md b/apps/claude-code/auto-format/docs/plans/17-spawnsync-timeout-guard.md
deleted file mode 100644
index 9e3b302b..00000000
--- a/apps/claude-code/auto-format/docs/plans/17-spawnsync-timeout-guard.md
+++ /dev/null
@@ -1,158 +0,0 @@
-# 17. `spawnSync` timeout guard for Prettier and ESLint
-**Status: done — 2026-04-28**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-03
-**Touches:** `scripts/format-hook.mjs`, `tests/format-hook.test.mjs`
-
-## Context
-
-`runPrettier` and `runEslint` in `format-hook.mjs` call `spawnSync` with no `timeout` option. A pathological input (e.g. a 50 MB minified JS file, a deeply nested JSON structure, an unterminated template literal) can cause Prettier or ESLint to hang indefinitely.
-
-The plugin's documented invariant #1 is "always exits 0 — never blocks Claude's tool flow." Without a timeout, that invariant only holds when the external tools themselves behave. A hung formatter keeps Claude's tool pipeline stalled until the user force-kills the session.
-
-Adding a `timeout` option to both `spawnSync` calls enforces the invariant unconditionally. The default of 30 s is generous for any single file; a per-project `formatTimeoutMs` escape hatch accommodates unusual setups.
-
-Note: spec-19 (Biome support) will introduce a third `spawnSync` call; the timeout guard must also be applied there. If spec-17 lands before spec-19, Ralph implementing spec-19 should copy the same pattern.
-
-## Current behaviour
-
-After spec-03 and spec-04: `runPrettier` and `runEslint` in `scripts/format-hook.mjs` both call `spawnSync` with `stdio: ['ignore', 'ignore', 'pipe']` and no `timeout`. If a formatter hangs, the hook hangs with it.
-
-## Target behaviour
-
-- Both `spawnSync` calls include `timeout: CONFIG.formatTimeoutMs` and `killSignal: 'SIGTERM'`.
-- `CONFIG.formatTimeoutMs` defaults to `30_000` (30 s), read from `.claude/unic-format.json` when present, clamped to `[1_000, 120_000]`.
-- When a formatter times out (`r.signal === 'SIGTERM'` or `r.status === null`), the hook logs one stderr line `unic-format: prettier timed out after 30s` (or `eslint`) and continues — the hook still exits 0.
-- A new smoke test stubs a hanging `prettier` binary and asserts the hook completes within ~33 s with the expected stderr message.
-
-## Implementation steps
-
-### Step 1 — Add `formatTimeoutMs` to `DEFAULTS` and `loadProjectConfig`
-
-In `scripts/format-hook.mjs`:
-
-1. Add to `DEFAULTS`:
-   ```js
-   const DEFAULTS = {
-   	// … existing keys …
-   	formatTimeoutMs: 30_000,
-   }
-   ```
-
-2. In `loadProjectConfig`, read and clamp the new key:
-   ```js
-   const raw = Number(cfg.formatTimeoutMs)
-   formatTimeoutMs: Number.isFinite(raw) ? Math.min(Math.max(raw, 1_000), 120_000) : DEFAULTS.formatTimeoutMs,
-   ```
-
-### Step 2 — Add timeout to `runPrettier` and `runEslint`
-
-Update both functions to include `timeout` and `killSignal`, and detect the timeout condition:
-
-```js
-function runPrettier(filePath) {
-	if (!existsSync(PRETTIER_BIN)) return
-	const r = spawnSync('node', [PRETTIER_BIN, '--write', '--ignore-unknown', '--log-level', 'warn', filePath], {
-		cwd: PROJECT_DIR,
-		stdio: ['ignore', 'ignore', 'pipe'],
-		timeout: CONFIG.formatTimeoutMs,
-		killSignal: 'SIGTERM',
-	})
-	if (r.signal === 'SIGTERM' || r.status === null) {
-		process.stderr.write(`unic-format: prettier timed out after ${CONFIG.formatTimeoutMs / 1000}s on ${filePath}\n`)
-		return
-	}
-	if (r.status !== 0) {
-		process.stderr.write(`unic-format: prettier failed: ${r.stderr?.toString().trim() || 'unknown error'}\n`)
-	}
-}
-
-function runEslint(filePath) {
-	if (!existsSync(ESLINT_BIN)) return
-	const r = spawnSync('node', [ESLINT_BIN, '--fix', '--no-error-on-unmatched-pattern', filePath], {
-		cwd: PROJECT_DIR,
-		stdio: ['ignore', 'ignore', 'pipe'],
-		timeout: CONFIG.formatTimeoutMs,
-		killSignal: 'SIGTERM',
-	})
-	if (r.signal === 'SIGTERM' || r.status === null) {
-		process.stderr.write(`unic-format: eslint timed out after ${CONFIG.formatTimeoutMs / 1000}s on ${filePath}\n`)
-		return
-	}
-	// Status 1 = lint warnings/errors remain after --fix (not a hook failure).
-	if (r.status !== 0 && r.status !== 1) {
-		process.stderr.write(`unic-format: eslint failed (exit ${r.status}): ${r.stderr?.toString().trim() || 'unknown error'}\n`)
-	}
-}
-```
-
-### Step 3 — Add timeout smoke test in `tests/format-hook.test.mjs`
-
-Create a stub `fake-prettier.mjs` that sleeps longer than the hook's default timeout, install it as a fake `node_modules/.bin/prettier` in a temp consumer dir, and assert the hook returns within a reasonable time with the timeout message.
-
-```js
-import { setTimeout as sleep } from 'node:timers/promises'
-
-test('exits 0 and logs timeout when prettier hangs', async () => {
-	const dir = makeConsumer((d) => {
-		// Create a stub prettier that sleeps for 90s (much longer than the 2s test timeout)
-		mkdirSync(join(d, 'node_modules', '.bin'), { recursive: true })
-		const stubScript = `#!/usr/bin/env node\nawait new Promise(r => setTimeout(r, 90_000))\n`
-		writeFileSync(join(d, 'node_modules', '.bin', 'prettier'), stubScript, { mode: 0o755 })
-		writeFileSync(join(d, 'test.md'), '# hello\n')
-	})
-	try {
-		// Override timeout to 2 s so the test doesn't actually wait 30 s
-		const { exitCode, stderr } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, 'test.md') } }),
-			dir,
-			{ UNIC_FORMAT_TIMEOUT_MS: '2000' },
-		)
-		assert.equal(exitCode, 0)
-		assert.match(stderr, /timed out/, 'should log timeout warning')
-	} finally {
-		cleanup(dir)
-	}
-})
-```
-
-Note: `UNIC_FORMAT_TIMEOUT_MS` is an env-var shorthand the implementation may expose as an alternative to the JSON config (simpler for tests); if the implementation only supports the JSON config key, the test should write a `.claude/unic-format.json` with `{"formatTimeoutMs": 2000}` instead.
-
-The preferred approach: read `formatTimeoutMs` from the JSON config only (not env var) and in the test, write the config file. The env-var shorthand shown above is for convenience only and need not be implemented.
-
-### Step 4 — Commit
-
-```sh
-git add scripts/format-hook.mjs tests/format-hook.test.mjs
-git commit -m "fix(spec-17): add spawnSync timeout guard (default 30s) to prevent hanging hook"
-```
-
-## Acceptance criteria
-
-- `runPrettier` and `runEslint` both include `timeout` and `killSignal: 'SIGTERM'` in their `spawnSync` options.
-- Timeout defaults to 30 000 ms; clamped to [1 000, 120 000] if overridden.
-- A timed-out call logs to stderr and returns — it does not throw or exit non-zero.
-- `pnpm test` passes including the new timeout test.
-- No external deps added.
-
-## Verification
-
-```sh
-# 1. timeout present in both spawnSync calls
-grep -n "timeout:" scripts/format-hook.mjs
-# Expected: 2 lines (one per function)
-
-# 2. All tests pass
-pnpm test
-```
-
-## Out of scope
-
-- Applying the timeout to the Biome `spawnSync` call (done in spec-19 which adds Biome).
-- Exposing timeout via env var (config file is sufficient).
-- SIGKILL escalation if SIGTERM is ignored (30 s is already generous; escalation adds complexity with minimal practical benefit).
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/18-notebook-and-skip-prefix-tests.md b/apps/claude-code/auto-format/docs/plans/18-notebook-and-skip-prefix-tests.md
deleted file mode 100644
index 4285e01d..00000000
--- a/apps/claude-code/auto-format/docs/plans/18-notebook-and-skip-prefix-tests.md
+++ /dev/null
@@ -1,140 +0,0 @@
-# 18. Smoke-test coverage for `notebook_path` and skip-prefix matrix
-**Status: done — 2026-04-28**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-09
-**Touches:** `tests/format-hook.test.mjs`
-
-## Context
-
-Two coverage gaps remain in `tests/format-hook.test.mjs`:
-
-1. **`notebook_path` branch.** `format-hook.mjs:117` reads either `tool_input.file_path` or `tool_input.notebook_path` to support NotebookEdit events. No test covers the `notebook_path` branch.
-
-2. **Skip-prefix matrix.** `SKIP_PREFIXES` contains 10 entries but only `_bmad/` has a dedicated test. `node_modules/`, `.git/`, and `.claude/worktrees/` are the three most critical because they contain large or sensitive files that must never be formatted. None are tested.
-
-These are additive tests only — no production code changes.
-
-## Current behaviour
-
-After spec-09: `tests/format-hook.test.mjs` has 8 tests:
-- empty stdin / empty JSON / missing file / `_bmad/` skip / path-traversal / `.toml` extension / malformed config / `prettierExtensions` override.
-- No test for `tool_input.notebook_path`.
-- No test for `node_modules/`, `.git/`, or `.claude/worktrees/` skip.
-
-## Target behaviour
-
-Four additional tests added to `tests/format-hook.test.mjs`:
-
-| Test | Scenario | Assertion |
-|---|---|---|
-| `notebook_path` event | `tool_input.notebook_path` points to an existing `.ipynb` (no formatter installed) | exits 0, silent |
-| `node_modules/` skip | `file_path` inside `node_modules/` of the consumer dir | exits 0, silent |
-| `.git/` skip | `file_path` inside `.git/` of the consumer dir | exits 0, silent |
-| `.claude/worktrees/` skip | `file_path` inside `.claude/worktrees/wt-1/` of the consumer dir | exits 0, silent |
-
-## Implementation steps
-
-### Step 1 — Add four tests to `tests/format-hook.test.mjs`
-
-Append after the last existing test:
-
-```js
-test('exits 0 with notebook_path event (no formatter installed)', () => {
-	const dir = makeConsumer((d) => {
-		writeFileSync(join(d, 'notebook.ipynb'), '{"cells":[],"nbformat":4,"nbformat_minor":5}\n')
-	})
-	try {
-		const { exitCode, stderr } = run(
-			JSON.stringify({ tool_input: { notebook_path: join(dir, 'notebook.ipynb') } }),
-			dir,
-		)
-		// .ipynb is not in ALLOWED_PRETTIER_EXT → skips silently before touching any tool
-		assert.equal(exitCode, 0)
-		assert.equal(stderr, '')
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('exits 0 and skips node_modules/ path', () => {
-	const dir = makeConsumer((d) => {
-		mkdirSync(join(d, 'node_modules', 'foo'), { recursive: true })
-		writeFileSync(join(d, 'node_modules', 'foo', 'index.js'), 'module.exports = {}\n')
-	})
-	try {
-		const { exitCode, stderr } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, 'node_modules', 'foo', 'index.js') } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		assert.equal(stderr, '')
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('exits 0 and skips .git/ path', () => {
-	const dir = makeConsumer((d) => {
-		mkdirSync(join(d, '.git', 'hooks'), { recursive: true })
-		writeFileSync(join(d, '.git', 'hooks', 'pre-commit'), '#!/bin/sh\n')
-	})
-	try {
-		const { exitCode, stderr } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, '.git', 'hooks', 'pre-commit') } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		assert.equal(stderr, '')
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('exits 0 and skips .claude/worktrees/ path', () => {
-	const dir = makeConsumer((d) => {
-		mkdirSync(join(d, '.claude', 'worktrees', 'wt-1'), { recursive: true })
-		writeFileSync(join(d, '.claude', 'worktrees', 'wt-1', 'feature.md'), '# feature\n')
-	})
-	try {
-		const { exitCode, stderr } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, '.claude', 'worktrees', 'wt-1', 'feature.md') } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		assert.equal(stderr, '')
-	} finally {
-		cleanup(dir)
-	}
-})
-```
-
-### Step 2 — Commit
-
-```sh
-git add tests/format-hook.test.mjs
-git commit -m "test(spec-18): add notebook_path and skip-prefix matrix coverage"
-```
-
-## Acceptance criteria
-
-- `pnpm test` reports 12 passing tests (8 existing + 4 new).
-- No production code changed.
-- All four new tests are silent on stderr (verifying the skip-prefix / extension guards fire before any tool invocation).
-
-## Verification
-
-```sh
-pnpm test
-# Output must contain: "12 pass"
-```
-
-## Out of scope
-
-- Testing `.ipynb` formatting if Jupyter/Prettier is installed (`.ipynb` is intentionally absent from `ALLOWED_PRETTIER_EXT`).
-- Testing every entry in `SKIP_PREFIXES` (the matrix covers the three highest-risk entries; the rest share the same code path verified by the `_bmad/` test).
-- Any change to `format-hook.mjs`.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/19-biome-support.md b/apps/claude-code/auto-format/docs/plans/19-biome-support.md
deleted file mode 100644
index 65f66ad3..00000000
--- a/apps/claude-code/auto-format/docs/plans/19-biome-support.md
+++ /dev/null
@@ -1,316 +0,0 @@
-# 19. Optional Biome support (auto-detected)
-**Status: done — 2026-04-28**
-
-**Priority:** P1
-**Effort:** M
-**Version impact:** minor
-**Depends on:** spec-03, spec-04, spec-09, spec-17
-**Touches:** `scripts/format-hook.mjs`, `tests/format-hook.test.mjs`, `README.md`
-
-## Context
-
-Biome is an increasingly common replacement for Prettier + ESLint in modern JS/TS projects. Today `format-hook.mjs` only detects Prettier and ESLint; a Biome-only project gets no auto-formatting from the hook.
-
-The sibling plugin `../unic-claude-code-confluence` already uses Biome 2.4.0 (`pnpm catalog`). Its `biome.json` is a proven, production-ready reference configuration.
-
-Design goals:
-1. **Zero config by default.** If a consumer project has `biome.json` (or `biome.jsonc`) at its root AND `node_modules/.bin/biome`, the hook switches to Biome automatically for Biome-supported extensions.
-2. **No bundled Biome.** The consumer pins their own Biome version (just like Prettier/ESLint today). The plugin ships zero Biome config.
-3. **Graceful fallback.** Extensions Biome does not handle (`.md`, `.mdx`, `.yml`, `.yaml`, `.feature`) still use Prettier when available.
-4. **Config escape hatch.** A `formatter` key in `.claude/unic-format.json` overrides auto-detection.
-5. **Timeout guard.** The Biome `spawnSync` call uses the same `CONFIG.formatTimeoutMs` guard from spec-17.
-
-Biome 2.x command for format + safe-fix lint on a single file:
-```
-node ./node_modules/.bin/biome check --write --no-errors-on-unmatched-pattern <file>
-```
-
-## Current behaviour
-
-After spec-03 and spec-04: `format-hook.mjs` only checks for `node_modules/.bin/prettier` and `node_modules/.bin/eslint`. Biome is neither detected nor invoked.
-
-## Target behaviour
-
-**Auto-detect rule:**
-
-1. Check for `node_modules/.bin/biome` and (`biome.json` or `biome.jsonc`) in `PROJECT_DIR`.
-2. If both present AND `CONFIG.formatter !== 'prettier'` → Biome mode:
-   - Run `node ./node_modules/.bin/biome check --write --no-errors-on-unmatched-pattern <file>` for extensions in `BIOME_EXTS`.
-   - Do **not** run Prettier or ESLint for those extensions.
-   - For extensions outside `BIOME_EXTS` but inside `ALLOWED_PRETTIER_EXT` (`.md`, `.mdx`, `.yml`, `.yaml`, `.feature`) — run Prettier if installed (Biome does not handle these).
-3. If detection fails OR `CONFIG.formatter === 'prettier'` → legacy Prettier + ESLint path (unchanged).
-4. `CONFIG.formatter === 'biome'` forces Biome mode even if auto-detection would otherwise fall back (requires binary + config to be present; logs a warning to stderr and returns if either is missing).
-
-**`BIOME_EXTS`** (file extensions Biome 2.x handles natively):
-```
-.js .mjs .cjs .ts .mts .cts .tsx .jsx .json .jsonc
-```
-
-**New `DEFAULTS` key:**
-```js
-formatter: 'auto',   // 'auto' | 'prettier' | 'biome'
-```
-
-**`loadProjectConfig` additions:**
-```js
-const VALID_FORMATTERS = new Set(['auto', 'prettier', 'biome'])
-formatter: VALID_FORMATTERS.has(cfg.formatter) ? cfg.formatter : DEFAULTS.formatter,
-```
-
-## Implementation steps
-
-### Step 1 — Add Biome constants to `scripts/format-hook.mjs`
-
-Add after the existing `ESLINT_BIN` line:
-
-```js
-const BIOME_BIN = resolve(PROJECT_DIR, 'node_modules/.bin/biome')
-const BIOME_CONFIG_PATH = [resolve(PROJECT_DIR, 'biome.json'), resolve(PROJECT_DIR, 'biome.jsonc')]
-
-const BIOME_EXTS = new Set(['.js', '.mjs', '.cjs', '.ts', '.mts', '.cts', '.tsx', '.jsx', '.json', '.jsonc'])
-
-const VALID_FORMATTERS = new Set(['auto', 'prettier', 'biome'])
-```
-
-### Step 2 — Add `formatter` to `DEFAULTS` and `loadProjectConfig`
-
-In `DEFAULTS`:
-```js
-formatter: 'auto',
-```
-
-In `loadProjectConfig`'s return:
-```js
-formatter: VALID_FORMATTERS.has(cfg.formatter) ? cfg.formatter : DEFAULTS.formatter,
-```
-
-### Step 3 — Add `useBiome` detection and `runBiome` function
-
-Add the detection helper and runner after the existing constants block:
-
-```js
-const BIOME_AVAILABLE =
-	existsSync(BIOME_BIN) && BIOME_CONFIG_PATH.some((p) => existsSync(p))
-
-function runBiome(filePath) {
-	if (!existsSync(BIOME_BIN)) {
-		process.stderr.write(`unic-format: biome binary not found at ${BIOME_BIN}\n`)
-		return
-	}
-	const r = spawnSync(
-		'node',
-		[BIOME_BIN, 'check', '--write', '--no-errors-on-unmatched-pattern', filePath],
-		{
-			cwd: PROJECT_DIR,
-			stdio: ['ignore', 'ignore', 'pipe'],
-			timeout: CONFIG.formatTimeoutMs,
-			killSignal: 'SIGTERM',
-		},
-	)
-	if (r.signal === 'SIGTERM' || r.status === null) {
-		process.stderr.write(`unic-format: biome timed out after ${CONFIG.formatTimeoutMs / 1000}s on ${filePath}\n`)
-		return
-	}
-	if (r.status !== 0) {
-		process.stderr.write(`unic-format: biome failed (exit ${r.status}): ${r.stderr?.toString().trim() || 'unknown error'}\n`)
-	}
-}
-```
-
-### Step 4 — Update `main()` routing logic
-
-Replace the current:
-```js
-	runPrettier(filePath)
-	if (ESLINT_EXTS.has(ext)) runEslint(filePath)
-```
-
-with:
-```js
-	const usesBiome =
-		CONFIG.formatter === 'biome' ||
-		(CONFIG.formatter === 'auto' && BIOME_AVAILABLE && BIOME_EXTS.has(ext))
-
-	if (usesBiome) {
-		runBiome(filePath)
-	} else {
-		runPrettier(filePath)
-		if (ESLINT_EXTS.has(ext)) runEslint(filePath)
-	}
-```
-
-Note: for extensions outside `BIOME_EXTS` (`.md`, `.yml`, etc.), `usesBiome` evaluates to `false` even in a Biome project, so Prettier is still used for those extensions.
-
-### Step 5 — Update `README.md`
-
-Add a **Biome support** section after the existing per-project config section:
-
-```markdown
-## Biome support
-
-If your project has `biome.json` (or `biome.jsonc`) at the root and `@biomejs/biome` installed,
-the hook auto-detects Biome and uses it instead of Prettier + ESLint for JS/TS/JSON files.
-
-Biome handles: `.js .mjs .cjs .ts .mts .cts .tsx .jsx .json .jsonc`
-Prettier still runs for: `.md .mdx .yml .yaml .feature` (Biome does not support these)
-
-### Recommended `biome.json` (based on Unic defaults)
-
-\`\`\`json
-{
-  "$schema": "https://biomejs.dev/schemas/2.4.0/schema.json",
-  "assist": { "actions": { "source": { "organizeImports": "on" } } },
-  "files": {
-    "includes": [
-      "**",
-      "!**/node_modules",
-      "!**/.history",
-      "!**/pnpm-lock.yaml",
-      "!**/.claude",
-      "!**/.ralph",
-      "!**/*.min.js"
-    ]
-  },
-  "formatter": {
-    "enabled": true,
-    "indentStyle": "tab",
-    "indentWidth": 2,
-    "lineEnding": "lf",
-    "lineWidth": 120,
-    "useEditorconfig": true
-  },
-  "javascript": {
-    "formatter": {
-      "quoteStyle": "single",
-      "semicolons": "asNeeded",
-      "trailingCommas": "es5"
-    }
-  },
-  "json": {
-    "formatter": { "indentStyle": "space", "indentWidth": 2 }
-  },
-  "linter": { "enabled": true, "rules": { "recommended": true } }
-}
-\`\`\`
-
-### Force a specific formatter via config
-
-\`\`\`.claude/unic-format.json
-{
-  "formatter": "prettier"   // or "biome" (forced) or "auto" (default)
-}
-\`\`\`
-```
-
-### Step 6 — Add Biome smoke tests in `tests/format-hook.test.mjs`
-
-```js
-test('uses Biome when biome.json and biome binary are present', () => {
-	let biomeInvoked = false
-	const dir = makeConsumer((d) => {
-		// Stub biome binary that writes a sentinel to a file
-		mkdirSync(join(d, 'node_modules', '.bin'), { recursive: true })
-		const stubScript = `#!/usr/bin/env node\nimport { writeFileSync } from 'node:fs'\nwriteFileSync('${join(d, '.biome-called')}', '1')\n`
-		writeFileSync(join(d, 'node_modules', '.bin', 'biome'), stubScript, { mode: 0o755 })
-		writeFileSync(join(d, 'biome.json'), '{"$schema":"https://biomejs.dev/schemas/2.4.0/schema.json"}\n')
-		writeFileSync(join(d, 'test.ts'), 'const x = 1\n')
-	})
-	try {
-		const { exitCode } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, 'test.ts') } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		// Stub wrote sentinel file when invoked
-		assert.ok(existsSync(join(dir, '.biome-called')), 'biome should have been called for .ts')
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('does not use Biome for .md even when Biome is detected', () => {
-	const dir = makeConsumer((d) => {
-		mkdirSync(join(d, 'node_modules', '.bin'), { recursive: true })
-		const biomeStub = `#!/usr/bin/env node\nimport { writeFileSync } from 'node:fs'\nwriteFileSync('${join(d, '.biome-called')}', '1')\n`
-		writeFileSync(join(d, 'node_modules', '.bin', 'biome'), biomeStub, { mode: 0o755 })
-		writeFileSync(join(d, 'biome.json'), '{}')
-		writeFileSync(join(d, 'README.md'), '# hello\n')
-	})
-	try {
-		const { exitCode } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, 'README.md') } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		// Biome stub must NOT have been called for .md
-		const { existsSync: exists } = await import('node:fs')
-		assert.ok(!existsSync(join(dir, '.biome-called')), 'biome should NOT be called for .md')
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('respects formatter: "prettier" override even when Biome is detected', () => {
-	const dir = makeConsumer((d) => {
-		mkdirSync(join(d, 'node_modules', '.bin'), { recursive: true })
-		const biomeStub = `#!/usr/bin/env node\nimport { writeFileSync } from 'node:fs'\nwriteFileSync('${join(d, '.biome-called')}', '1')\n`
-		writeFileSync(join(d, 'node_modules', '.bin', 'biome'), biomeStub, { mode: 0o755 })
-		writeFileSync(join(d, 'biome.json'), '{}')
-		mkdirSync(join(d, '.claude'))
-		writeFileSync(join(d, '.claude', 'unic-format.json'), JSON.stringify({ formatter: 'prettier' }))
-		writeFileSync(join(d, 'test.ts'), 'const x = 1\n')
-	})
-	try {
-		const { exitCode } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, 'test.ts') } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		assert.ok(!existsSync(join(dir, '.biome-called')), 'biome should NOT be called when formatter is "prettier"')
-	} finally {
-		cleanup(dir)
-	}
-})
-```
-
-### Step 7 — Commit
-
-```sh
-git add scripts/format-hook.mjs tests/format-hook.test.mjs README.md
-git commit -m "feat(spec-19): add auto-detected Biome support as alternative to Prettier+ESLint"
-```
-
-## Acceptance criteria
-
-- `BIOME_AVAILABLE` is evaluated once at module scope.
-- `runBiome` uses `biome check --write --no-errors-on-unmatched-pattern`.
-- `runBiome` includes timeout + killSignal (consistent with spec-17 pattern for `runPrettier`/`runEslint`).
-- For `.ts` in a Biome project: Biome is called, Prettier is not.
-- For `.md` in a Biome project: Prettier is called (or skipped if not installed), Biome is not.
-- `formatter: "prettier"` in config overrides Biome detection.
-- `pnpm test` passes (15+ tests green including 3 new Biome tests).
-- README documents auto-detect rule, extension lists, recommended `biome.json`, and the `formatter` config key.
-- No external deps added to `dependencies` in `package.json`.
-
-## Verification
-
-```sh
-# 1. BIOME_AVAILABLE and runBiome exist
-grep -n "BIOME_AVAILABLE\|runBiome\|BIOME_EXTS" scripts/format-hook.mjs
-
-# 2. Only node: imports in hook script
-grep "^import" scripts/format-hook.mjs | grep -v "node:"
-
-# 3. Tests pass
-pnpm test
-```
-
-## Out of scope
-
-- Biome version enforcement (consumer pins their own version).
-- Running `biome migrate` or modifying the consumer's `biome.json`.
-- Supporting the legacy `biome lint --apply` path (spec targets Biome 2.x `check --write`).
-- `biomeExtensions` config override (BIOME_EXTS is dictated by Biome itself, not by the plugin).
-- Lint-only mode without formatting.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/20-additional-skip-prefixes.md b/apps/claude-code/auto-format/docs/plans/20-additional-skip-prefixes.md
deleted file mode 100644
index 4fcf6dd7..00000000
--- a/apps/claude-code/auto-format/docs/plans/20-additional-skip-prefixes.md
+++ /dev/null
@@ -1,191 +0,0 @@
-# 20. `additionalSkipPrefixes` config key (additive merge)
-**Status: done — 2026-04-28**
-
-**Priority:** P2
-**Effort:** S
-**Version impact:** minor
-**Depends on:** spec-04, spec-09
-**Touches:** `scripts/format-hook.mjs`, `tests/format-hook.test.mjs`, `README.md`
-
-## Context
-
-Today the per-project config key `skipPrefixes` in `.claude/unic-format.json` **replaces** the entire `DEFAULTS.skipPrefixes` array (`format-hook.mjs:60`). A consumer who wants to add a single prefix — say `my-generated/` — must copy all 10 defaults verbatim:
-
-```json
-{
-  "skipPrefixes": [
-    "_bmad/", ".claude/skills/bmad-", ".claude/worktrees/",
-    ".history/", ".git/", "node_modules/", "dist/", "build/",
-    ".next/", "coverage/", "my-generated/"
-  ]
-}
-```
-
-If the plugin adds a new default in a future release, those consumers silently lose it. It's a maintenance foot-gun.
-
-The fix is a companion key `additionalSkipPrefixes` that merges with defaults rather than replacing them. The existing `skipPrefixes` key keeps its full-replacement semantics (backward compatible). When both keys are present, `skipPrefixes` wins (explicit over additive).
-
-## Current behaviour
-
-After spec-04: `loadProjectConfig` returns `cfg.skipPrefixes` if it is an array, `DEFAULTS.skipPrefixes` otherwise. There is no additive-merge option.
-
-## Target behaviour
-
-**`loadProjectConfig` resolution order for skip prefixes:**
-
-1. If `cfg.skipPrefixes` is a non-empty array → use as full replacement (unchanged legacy behaviour).
-2. Else if `cfg.additionalSkipPrefixes` is a non-empty array → `[...DEFAULTS.skipPrefixes, ...cfg.additionalSkipPrefixes]`.
-3. Else → `DEFAULTS.skipPrefixes`.
-
-**Config example:**
-
-```json
-{
-  "additionalSkipPrefixes": ["my-generated/", ".tmp/"]
-}
-```
-
-No duplication of defaults required. Future default additions are inherited automatically.
-
-## Implementation steps
-
-### Step 1 — Update `loadProjectConfig` in `scripts/format-hook.mjs`
-
-Current return statement:
-```js
-return {
-	skipPrefixes: Array.isArray(cfg.skipPrefixes) ? cfg.skipPrefixes : DEFAULTS.skipPrefixes,
-	prettierExtensions: Array.isArray(cfg.prettierExtensions) ? cfg.prettierExtensions : DEFAULTS.prettierExtensions,
-	eslintExtensions: Array.isArray(cfg.eslintExtensions) ? cfg.eslintExtensions : DEFAULTS.eslintExtensions,
-}
-```
-
-Updated (add `additionalSkipPrefixes` merge logic, leave other keys unchanged):
-```js
-const hasFullReplacement = Array.isArray(cfg.skipPrefixes) && cfg.skipPrefixes.length > 0
-const hasAdditive = Array.isArray(cfg.additionalSkipPrefixes) && cfg.additionalSkipPrefixes.length > 0
-
-return {
-	skipPrefixes: hasFullReplacement
-		? cfg.skipPrefixes
-		: hasAdditive
-			? [...DEFAULTS.skipPrefixes, ...cfg.additionalSkipPrefixes]
-			: DEFAULTS.skipPrefixes,
-	prettierExtensions: Array.isArray(cfg.prettierExtensions) ? cfg.prettierExtensions : DEFAULTS.prettierExtensions,
-	eslintExtensions: Array.isArray(cfg.eslintExtensions) ? cfg.eslintExtensions : DEFAULTS.eslintExtensions,
-}
-```
-
-### Step 2 — Add tests in `tests/format-hook.test.mjs`
-
-```js
-test('additionalSkipPrefixes merges with defaults', () => {
-	const dir = makeConsumer((d) => {
-		mkdirSync(join(d, '.claude'))
-		mkdirSync(join(d, 'my-generated'))
-		writeFileSync(join(d, 'my-generated', 'file.md'), '# generated\n')
-		writeFileSync(
-			join(d, '.claude', 'unic-format.json'),
-			JSON.stringify({ additionalSkipPrefixes: ['my-generated/'] }),
-		)
-	})
-	try {
-		// Custom prefix is skipped
-		const { exitCode: exitCustom, stderr: stderrCustom } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, 'my-generated', 'file.md') } }),
-			dir,
-		)
-		assert.equal(exitCustom, 0)
-		assert.equal(stderrCustom, '', 'should skip custom prefix silently')
-
-		// Default prefix _bmad/ must still be skipped (defaults preserved)
-		mkdirSync(join(dir, '_bmad'))
-		writeFileSync(join(dir, '_bmad', 'test.md'), '# test\n')
-		const { exitCode: exitDefault, stderr: stderrDefault } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, '_bmad', 'test.md') } }),
-			dir,
-		)
-		assert.equal(exitDefault, 0)
-		assert.equal(stderrDefault, '', 'should still skip _bmad/ from defaults')
-	} finally {
-		cleanup(dir)
-	}
-})
-
-test('skipPrefixes wins over additionalSkipPrefixes when both are set', () => {
-	const dir = makeConsumer((d) => {
-		mkdirSync(join(d, '.claude'))
-		mkdirSync(join(d, '_bmad'))
-		writeFileSync(join(d, '_bmad', 'test.md'), '# test\n')
-		// skipPrefixes replaces all defaults; _bmad/ is NOT in this list
-		// additionalSkipPrefixes is present but should be ignored
-		writeFileSync(
-			join(d, '.claude', 'unic-format.json'),
-			JSON.stringify({
-				skipPrefixes: ['dist/'],
-				additionalSkipPrefixes: ['my-generated/'],
-			}),
-		)
-	})
-	try {
-		// _bmad/ is NOT skipped because skipPrefixes replaced defaults and doesn't include it.
-		// The file doesn't exist in node_modules so no formatter runs — just verify it's
-		// not silently skipped by _bmad/ (it would proceed to the extension check and return
-		// silently because no prettier binary is installed).
-		const { exitCode, stderr } = run(
-			JSON.stringify({ tool_input: { file_path: join(dir, '_bmad', 'test.md') } }),
-			dir,
-		)
-		assert.equal(exitCode, 0)
-		// No formatter installed → stderr is empty regardless.
-		// The key assertion is that the config was parsed correctly (no crash / malformed-config warning).
-		assert.ok(!stderr.includes('malformed'), 'config should parse without error')
-	} finally {
-		cleanup(dir)
-	}
-})
-```
-
-### Step 3 — Update `README.md`
-
-In the per-project config section, add documentation for the new key alongside `skipPrefixes`:
-
-```markdown
-| `additionalSkipPrefixes` | `string[]` | Extra path prefixes to skip, **merged with the defaults**. Prefer this over `skipPrefixes` unless you need full replacement. Example: `["my-generated/", ".tmp/"]` |
-```
-
-And add a note clarifying precedence:
-> If both `skipPrefixes` and `additionalSkipPrefixes` are present, `skipPrefixes` takes full precedence (full replacement). `additionalSkipPrefixes` is ignored when `skipPrefixes` is set.
-
-### Step 4 — Commit
-
-```sh
-git add scripts/format-hook.mjs tests/format-hook.test.mjs README.md
-git commit -m "feat(spec-20): add additionalSkipPrefixes config key for additive prefix merging"
-```
-
-## Acceptance criteria
-
-- `additionalSkipPrefixes` in config merges with `DEFAULTS.skipPrefixes`.
-- `skipPrefixes` full replacement is unchanged (existing configs work exactly as before).
-- When both keys are set, `skipPrefixes` wins.
-- `pnpm test` passes with 2 new tests green.
-- README documents both keys with a precedence note.
-
-## Verification
-
-```sh
-# 1. additionalSkipPrefixes merge logic in loadProjectConfig
-grep -A5 "additionalSkipPrefixes" scripts/format-hook.mjs
-
-# 2. Tests pass
-pnpm test
-```
-
-## Out of scope
-
-- `additionalPrettierExtensions` / `additionalEslintExtensions` (those have smaller risk of consumers needing additive merging vs. full replacement).
-- Validating the syntax of each prefix string (e.g. must end with `/`).
-- Deduplication of merged prefixes.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/21-json-indentation.md b/apps/claude-code/auto-format/docs/plans/21-json-indentation.md
deleted file mode 100644
index 2e306ad8..00000000
--- a/apps/claude-code/auto-format/docs/plans/21-json-indentation.md
+++ /dev/null
@@ -1,183 +0,0 @@
-# 21. Honour `.editorconfig` 2-space indentation in committed JSON
-**Status: done — 2026-04-28**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-07, spec-13
-**Touches:** `scripts/bump.mjs`, `scripts/sync-version.mjs`, `.claude-plugin/plugin.json`, `.claude-plugin/marketplace.json`, `package.json`, `hooks/hooks.json`, `tests/` (new test)
-
-## Context
-
-`.editorconfig` declares `indent_style = space` and `indent_size = 2` for all `*.json` files. However, every committed JSON file in the repository uses tab indentation:
-
-```
-.claude-plugin/plugin.json       → tabs (written by bump.mjs)
-.claude-plugin/marketplace.json  → tabs (written by sync-version.mjs)
-package.json                     → tabs (written by sync-version.mjs)
-hooks/hooks.json                 → tabs (committed by hand)
-```
-
-The root cause is `scripts/bump.mjs:20` and `scripts/sync-version.mjs:48`, which both call `JSON.stringify(obj, null, '\t')`. Every `pnpm bump` re-writes three JSON files with tabs, defeating any editor that auto-corrects on save.
-
-For non-developer users (Product Owners, spec authors) whose editors respect `.editorconfig`, the mismatch causes unexpected diffs and noisy history. It also contradicts the self-hosted formatting principle: Prettier (and Biome with `useEditorconfig: true`) would convert committed JSON to 2-space, but the release scripts revert that on every bump.
-
-The fix is minimal: change two `'\t'` arguments to `2`, then do a one-shot reformat of the four committed JSON files so git history is clean from this commit forward.
-
-## Current behaviour
-
-- `scripts/bump.mjs:20` (`writeJson`): `JSON.stringify(obj, null, '\t')`
-- `scripts/sync-version.mjs:48`: `JSON.stringify(obj, null, '\t')`
-- All four committed JSON files are tab-indented.
-- `.editorconfig` says 2-space.
-
-## Target behaviour
-
-- `scripts/bump.mjs` writes JSON with 2-space indentation.
-- `scripts/sync-version.mjs` writes JSON with 2-space indentation.
-- All four committed JSON files are 2-space indented.
-- `pnpm test` includes a regression test: running `sync-version.mjs` against a fixture produces 2-space output.
-
-## Implementation steps
-
-### Step 1 — Update `scripts/bump.mjs`
-
-In `writeJson`:
-```js
-function writeJson(rel, obj) {
-	writeFileSync(resolve(ROOT, rel), JSON.stringify(obj, null, 2) + '\n', 'utf8')
-}
-```
-Change `'\t'` → `2`.
-
-### Step 2 — Update `scripts/sync-version.mjs`
-
-Line 48:
-```js
-writeFileSync(filePath, JSON.stringify(obj, null, 2) + '\n', 'utf8')
-```
-Change `'\t'` → `2`.
-
-### Step 3 — Reformat the four committed JSON files
-
-Run the following deterministic reformatter (or apply by hand — the content is small):
-
-```sh
-node --input-type=module <<'EOF'
-import { readFileSync, writeFileSync } from 'node:fs'
-
-const files = [
-  '.claude-plugin/plugin.json',
-  '.claude-plugin/marketplace.json',
-  'package.json',
-  'hooks/hooks.json',
-]
-
-for (const f of files) {
-  const obj = JSON.parse(readFileSync(f, 'utf8'))
-  writeFileSync(f, JSON.stringify(obj, null, 2) + '\n', 'utf8')
-  console.log(`reformatted: ${f}`)
-}
-EOF
-```
-
-Verify with:
-```sh
-node -e "const s=require('fs').readFileSync('.claude-plugin/plugin.json','utf8'); if(s.includes('\t')) throw new Error('tabs found')"
-```
-
-### Step 4 — Add a regression test in `tests/`
-
-Create `tests/sync-version.test.mjs`:
-
-```js
-import { test } from 'node:test'
-import { strict as assert } from 'node:assert'
-import { mkdtempSync, writeFileSync, readFileSync, mkdirSync, rmSync } from 'node:fs'
-import { tmpdir } from 'node:os'
-import { join } from 'node:path'
-import { spawnSync } from 'node:child_process'
-import { fileURLToPath } from 'node:url'
-
-const SCRIPT = fileURLToPath(new URL('../scripts/sync-version.mjs', import.meta.url))
-
-function makeFixtureRepo(version) {
-	const dir = mkdtempSync(join(tmpdir(), 'unic-sync-version-test-'))
-	mkdirSync(join(dir, '.claude-plugin'))
-	writeFileSync(
-		join(dir, '.claude-plugin', 'plugin.json'),
-		JSON.stringify({ name: 'test-plugin', version }, null, 2) + '\n',
-	)
-	writeFileSync(join(dir, '.claude-plugin', 'marketplace.json'), JSON.stringify({ version: '0.0.0' }, null, 2) + '\n')
-	writeFileSync(join(dir, 'package.json'), JSON.stringify({ version: '0.0.0' }, null, 2) + '\n')
-	return dir
-}
-
-test('sync-version writes 2-space indented JSON (no tabs)', () => {
-	const dir = makeFixtureRepo('1.2.3')
-	try {
-		const result = spawnSync(process.execPath, [SCRIPT], {
-			env: { ...process.env },
-			cwd: dir,
-			encoding: 'utf8',
-		})
-		assert.equal(result.status, 0, `sync-version failed: ${result.stderr}`)
-
-		for (const rel of ['.claude-plugin/marketplace.json', 'package.json']) {
-			const content = readFileSync(join(dir, rel), 'utf8')
-			assert.ok(!content.includes('\t'), `${rel} must not contain tabs`)
-			assert.ok(content.startsWith('{\n  "'), `${rel} must start with 2-space indent`)
-		}
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-```
-
-Update `package.json` `test` script if needed to pick up the new test file automatically (node:test `--test` glob already picks up `tests/*.test.mjs` if the existing script uses a glob pattern — verify and adjust if needed).
-
-### Step 5 — Commit
-
-```sh
-git add scripts/bump.mjs scripts/sync-version.mjs \
-        .claude-plugin/plugin.json .claude-plugin/marketplace.json \
-        package.json hooks/hooks.json \
-        tests/sync-version.test.mjs
-git commit -m "fix(spec-21): switch JSON output to 2-space indentation to match .editorconfig"
-```
-
-## Acceptance criteria
-
-- `scripts/bump.mjs` `writeJson` uses `JSON.stringify(obj, null, 2)`.
-- `scripts/sync-version.mjs` `syncFile` uses `JSON.stringify(obj, null, 2)`.
-- All four committed JSON files contain no tab characters.
-- All four committed JSON files start with `{\n  "` (2-space indent).
-- `pnpm test` passes, including the new `sync-version.test.mjs` test.
-- `pnpm verify:changelog` passes (no breaking structural change).
-
-## Verification
-
-```sh
-# 1. No tabs in committed JSON files
-for f in .claude-plugin/plugin.json .claude-plugin/marketplace.json package.json hooks/hooks.json; do
-  grep -P "\t" "$f" && echo "FAIL: tabs in $f" || echo "ok: $f"
-done
-
-# 2. Scripts updated
-grep "null, 2" scripts/bump.mjs scripts/sync-version.mjs
-
-# 3. Tests pass (includes new sync-version test)
-pnpm test
-
-# 4. Changelog gate still passes
-pnpm verify:changelog
-```
-
-## Out of scope
-
-- Reformatting any other file types (`.yml`, `.yaml` — they are already tab-free via pnpm-workspace.yaml and ci.yml).
-- Changing `.editorconfig` itself.
-- Reformatting JSON files inside `docs/plans/` (they are Markdown containing JSON code blocks, not JSON files).
-- Adding a Prettier run to the release scripts (out of scope per plugin design — no bundled formatter).
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/22-jsdoc-types-ts-check.md b/apps/claude-code/auto-format/docs/plans/22-jsdoc-types-ts-check.md
deleted file mode 100644
index d9d919ee..00000000
--- a/apps/claude-code/auto-format/docs/plans/22-jsdoc-types-ts-check.md
+++ /dev/null
@@ -1,410 +0,0 @@
-# 22. JSDoc Types and `--checkJs` Type-Checking
-**Status: done — 2026-04-28**
-
-**Priority:** P2
-**Effort:** M
-**Version impact:** patch
-**Depends on:** spec-19 (Biome support — ensures format diffs and JSDoc diffs stay in separate commits)
-**Touches:** `scripts/format-hook.mjs`, `scripts/sync-version.mjs`, `scripts/bump.mjs`, `scripts/tag.mjs`, `scripts/verify-changelog.mjs`, `scripts/lib/types.mjs` (new), `tests/format-hook.test.mjs`, `tests/sync-version.test.mjs`, `package.json`, `pnpm-workspace.yaml`, `README.md`, `CLAUDE.md`
-
-## Context
-
-The codebase is plain JavaScript (ESM). There is no build step and no TypeScript transpilation. Adding `// @ts-check` at the top of each source file and JSDoc type annotations on public functions gives editor-level type safety (VS Code underlines type errors in real time) and catches an entire class of bugs — wrong argument types, missing properties, `undefined` values used as strings — without introducing a TypeScript build step. The `tsc --checkJs` flag extends this to local dev: type errors are visible immediately via `pnpm typecheck`. This is the lightest possible path to type safety in a no-build ESM project.
-
-This pattern is already established in the sibling `unic-claude-code-confluence` plugin (spec 14, landed 2026-04-24) and is ported here for parity. The motivation in this repo is the same: catch latent type holes in `format-hook.mjs` — e.g. `JSON.parse(buf)` produces `any`, `event?.tool_input?.file_path` is `string | undefined`, catch-clause `err` is `unknown` under `--strict`. All are safe at runtime but undetectable by the IDE today.
-
-The choice to NOT convert to `.ts` files is deliberate — it keeps the plugin source dependency-free at runtime and copy-paste-friendly.
-
-`typescript` and `@types/node` are **dev-only**. They have zero impact on the hook's runtime behaviour and do not touch the consumer-facing contract. This spec does not conflict with the scope guard.
-
-## Current behaviour
-
-- No `// @ts-check` directive in any `.mjs` file.
-- No JSDoc annotations on any function in `scripts/` or `tests/`.
-- No `typescript` or `@types/node` in `devDependencies`. `pnpm-workspace.yaml` catalog has only `@ralph-orchestrator/ralph-cli`.
-- No `pnpm typecheck` script.
-- No `tsconfig.json`.
-- Latent type holes in `scripts/format-hook.mjs`:
-  - `JSON.parse(readFileSync(configPath))` and `JSON.parse(buf)` return `any` — type errors silently propagate.
-  - `event?.tool_input?.file_path` is `string | undefined`; the truthy check on line 180 guards runtime but does not narrow the type.
-  - `catch (err)` blocks reference `err.message` — under `--strict`, `err` is `unknown`, so this is a type error.
-  - `spawnSync` return `.stderr` is `Buffer | null`; the `?.toString()` access is fine but `.status` is `number | null`.
-
-## Target behaviour
-
-After this spec:
-
-1. Every `.mjs` file in `scripts/` (including the new `scripts/lib/` subdirectory) and `tests/` starts with `// @ts-check` on the second line (after the SPDX comment or file header).
-2. `scripts/lib/types.mjs` defines shared type shapes as JSDoc `@typedef` blocks — no runtime exports.
-3. Public functions in all scripts are annotated with `@param` and `@returns` JSDoc tags.
-4. `pnpm typecheck` runs `tsc --noEmit --allowJs --checkJs --strict --target ES2022 --module NodeNext --moduleResolution NodeNext scripts/*.mjs scripts/lib/*.mjs tests/*.mjs` and exits 0.
-5. VS Code (with the TypeScript language server) shows 0 errors in the Problems panel for all annotated `.mjs` files.
-6. No `.ts` files are created. No transpilation step is introduced.
-
-## Implementation steps
-
-### Step 1 — Add `typescript` and `@types/node` to catalog and devDependencies
-
-Edit `pnpm-workspace.yaml`:
-
-```yaml
-# BEFORE:
-catalog:
-  "@ralph-orchestrator/ralph-cli": 2.9.2
-
-# AFTER:
-catalog:
-  "@ralph-orchestrator/ralph-cli": 2.9.2
-  "@types/node": "24.12.2"   # Ralph: replace with latest stable @types/node@24.x
-  typescript: "5.8.3"        # Ralph: replace with latest stable TypeScript 5.x
-```
-
-Edit `package.json` `devDependencies`:
-
-```json
-"devDependencies": {
-  "@ralph-orchestrator/ralph-cli": "catalog:",
-  "@types/node": "catalog:",
-  "typescript": "catalog:"
-}
-```
-
-Run `pnpm install`.
-
-### Step 2 — Add `pnpm typecheck` script to `package.json`
-
-Add to the `"scripts"` block:
-
-```json
-"typecheck": "tsc --noEmit --allowJs --checkJs --strict --target ES2022 --module NodeNext --moduleResolution NodeNext scripts/*.mjs scripts/lib/*.mjs tests/*.mjs"
-```
-
-Do NOT create a `tsconfig.json`. The inline flags are sufficient for this repo's file count. A `tsconfig.json` is only warranted if the command line grows past ~120 chars or a second target (e.g. different settings for tests) is needed — document that in the follow-ups section.
-
-### Step 3 — Create `scripts/lib/types.mjs`
-
-Create the directory `scripts/lib/` and the new file `scripts/lib/types.mjs`. This file has NO runtime code — it exists only to define JSDoc types accessible via `@import` directives in other files.
-
-```js
-// @ts-check
-// SPDX-License-Identifier: LGPL-3.0-or-later
-// Copyright © 2026 Unic
-/**
- * Shared JSDoc type definitions for format-hook.mjs and its helpers.
- * No runtime exports — import types only via JSDoc @import directives.
- */
-
-/**
- * The hook event JSON received on stdin from Claude Code's PostToolUse hook.
- *
- * @typedef {{
- *   tool_input?: {
- *     file_path?: string,
- *     notebook_path?: string
- *   }
- * }} HookEvent
- */
-
-/**
- * Formatter selection — the set of values accepted by the `formatter` config key.
- *
- * @typedef {'auto' | 'prettier' | 'biome'} FormatterName
- */
-
-/**
- * Resolved project configuration (output of loadProjectConfig).
- *
- * @typedef {{
- *   skipPrefixes: string[],
- *   prettierExtensions: string[],
- *   eslintExtensions: string[],
- *   formatTimeoutMs: number,
- *   formatter: FormatterName
- * }} ProjectConfig
- */
-
-export {}
-```
-
-The `export {}` makes this a valid ESM module under `--moduleResolution NodeNext`. The `@typedef` blocks are visible to any file that imports via JSDoc `/** @import { ... } from './lib/types.mjs' */`.
-
-### Step 4 — Add `// @ts-check` and JSDoc to `scripts/format-hook.mjs`
-
-Add to the top of the file (after the `#!/usr/bin/env node` line and the existing block comment):
-
-```js
-// @ts-check
-/** @import { HookEvent, ProjectConfig, FormatterName } from './lib/types.mjs' */
-```
-
-Annotate `loadProjectConfig`:
-
-```js
-/**
- * Reads .claude/unic-format.json from the consumer project root and merges
- * it with DEFAULTS. Returns DEFAULTS on missing file or parse error.
- *
- * @returns {ProjectConfig}
- */
-function loadProjectConfig() {
-```
-
-Note: the `cfg` local from `JSON.parse` is `any` — cast it to `Record<string, unknown>` to avoid propagating `any`:
-
-```js
-const cfg = /** @type {Record<string, unknown>} */ (JSON.parse(readFileSync(configPath, 'utf8')))
-```
-
-For the `formatter` field, `VALID_FORMATTERS.has(cfg.formatter)` narrows nothing by itself — use a type assertion on the result:
-
-```js
-formatter: /** @type {FormatterName} */ (VALID_FORMATTERS.has(cfg.formatter) ? cfg.formatter : DEFAULTS.formatter),
-```
-
-Annotate the catch block to fix `err.message` under `--strict` (`err` is `unknown`):
-
-```js
-} catch (err) {
-  process.stderr.write(`unic-format: ignoring malformed .claude/unic-format.json: ${/** @type {Error} */ (err).message}\n`)
-  return DEFAULTS
-}
-```
-
-Annotate `toPosix`:
-
-```js
-/**
- * Converts a native path to forward-slash separators (no-op on POSIX).
- *
- * @param {string} p
- * @returns {string}
- */
-function toPosix(p) {
-```
-
-Annotate `shouldSkip`:
-
-```js
-/**
- * Returns true if the relative posix path should be skipped by the formatter.
- *
- * @param {string} rel - Posix-style path relative to PROJECT_DIR.
- * @returns {boolean}
- */
-function shouldSkip(rel) {
-```
-
-Annotate `runPrettier`, `runEslint`, `runBiome`:
-
-```js
-/**
- * Runs `prettier --write` on filePath using the consumer's local Prettier binary.
- * No-ops if Prettier is not installed. Always returns undefined.
- *
- * @param {string} filePath - Absolute path to the file to format.
- * @returns {void}
- */
-function runPrettier(filePath) {
-```
-
-```js
-/**
- * Runs `eslint --fix` on filePath using the consumer's local ESLint binary.
- * No-ops if ESLint is not installed. Exit status 1 (unfixed lint violations) is tolerated.
- *
- * @param {string} filePath - Absolute path to the file to lint.
- * @returns {void}
- */
-function runEslint(filePath) {
-```
-
-```js
-/**
- * Runs `biome check --write` on filePath using the consumer's local Biome binary.
- * No-ops (with stderr warning) if Biome binary is missing.
- *
- * @param {string} filePath - Absolute path to the file to format.
- * @returns {void}
- */
-function runBiome(filePath) {
-```
-
-Annotate `main`:
-
-```js
-/**
- * Entry point — reads the Claude Code hook event from stdin, resolves the
- * target file path, guards against skip conditions, and dispatches to the
- * appropriate formatter runner.
- *
- * @returns {Promise<void>}
- */
-async function main() {
-```
-
-Cast the `JSON.parse(buf)` result and fix the catch clause:
-
-```js
-let event
-try {
-  event = /** @type {HookEvent} */ (JSON.parse(buf))
-} catch {
-  process.stderr.write('unic-format: could not parse hook input as JSON\n')
-  return
-}
-```
-
-### Step 5 — Add `// @ts-check` and JSDoc to `scripts/sync-version.mjs`
-
-Add `// @ts-check` after the leading comment. Cast all `JSON.parse` calls to `Record<string, unknown>` and narrow results before writes. Annotate exported or top-level functions with `@param` / `@returns`.
-
-### Step 6 — Add `// @ts-check` and JSDoc to `scripts/bump.mjs`, `scripts/tag.mjs`, `scripts/verify-changelog.mjs`
-
-Add `// @ts-check` after the leading comment in each file. Cast `JSON.parse` results. Annotate all top-level functions. Fix any `err.message` catch references using the `/** @type {Error} */ (err).message` pattern.
-
-### Step 7 — Add `// @ts-check` to `tests/format-hook.test.mjs` and `tests/sync-version.test.mjs`
-
-Add `// @ts-check` after the leading comment. Annotate the `makeConsumer`, `run`, and `cleanup` helpers. Cast `JSON.parse` calls on stdout from the hook script. Fix any catch-clause `err.message` references.
-
-### Step 8 — Run `pnpm typecheck` and fix all reported errors
-
-Address each error using these canonical patterns:
-
-**`JSON.parse` returns `any`:**
-```js
-const cfg = /** @type {Record<string, unknown>} */ (JSON.parse(readFileSync(path, 'utf8')))
-```
-
-**catch clause `err` is `unknown`:**
-```js
-} catch (err) {
-  // cast to Error — we control what we throw, so this is safe
-  process.stderr.write(`...: ${/** @type {Error} */ (err).message}\n`)
-}
-```
-
-**string-union narrowing for formatter:**
-```js
-formatter: /** @type {FormatterName} */ (VALID_FORMATTERS.has(cfg.formatter) ? cfg.formatter : DEFAULTS.formatter),
-```
-
-Do NOT add `@ts-ignore`. Only `@ts-expect-error` with an explanatory comment is acceptable where tsc is provably wrong.
-
-### Step 9 — Update `README.md` and `CLAUDE.md`
-
-In `README.md`, add `pnpm typecheck` to the developer commands table or contributing section.
-
-In `CLAUDE.md` under `## Commands`, add:
-
-```sh
-pnpm typecheck       # Type-check plugin source (tsc --checkJs, dev-only)
-```
-
-### Step 10 — Commit
-
-```sh
-git add scripts tests package.json pnpm-workspace.yaml README.md CLAUDE.md docs/plans/22-jsdoc-types-ts-check.md docs/plans/README.md
-git commit -m "feat(spec-22): add // @ts-check + JSDoc types and pnpm typecheck"
-```
-
-## Test cases
-
-### TC-01: `pnpm typecheck` exits 0 on clean codebase
-
-```sh
-pnpm typecheck
-# Expected: exit 0, no output
-```
-
-### TC-02: Introducing a type error causes exit 1
-
-Temporarily add to `format-hook.mjs`:
-
-```js
-const e = /** @type {HookEvent} */ ({})
-const bad = e.tool_input.bogusField  // Property 'bogusField' does not exist
-```
-
-```sh
-pnpm typecheck
-# Expected: exit 1, output mentions 'bogusField' and 'HookEvent'
-```
-
-Remove the line afterward.
-
-### TC-03: `JSON.parse` casts suppress `any` propagation
-
-After adding `/** @type {HookEvent} */` cast to `event`, hover over `event.tool_input?.file_path` in VS Code — it should be typed as `string | undefined` (not `any`).
-
-### TC-04: No `.ts` files created
-
-```sh
-find scripts tests -name "*.ts" | wc -l
-# Expected: 0
-```
-
-### TC-05: `pnpm test` still passes
-
-```sh
-pnpm test
-# Expected: exit 0 — JSDoc annotations must not break any existing tests
-```
-
-### TC-06: No `@ts-ignore` directives
-
-```sh
-grep -rn "@ts-ignore" scripts/ tests/
-# Expected: 0 results
-```
-
-## Acceptance criteria
-
-- `// @ts-check` appears in every `.mjs` file in `scripts/` (including `scripts/lib/types.mjs`) and `tests/`.
-- `scripts/lib/types.mjs` exists with `HookEvent`, `FormatterName`, and `ProjectConfig` typedefs.
-- `loadProjectConfig`, `toPosix`, `shouldSkip`, `runPrettier`, `runEslint`, `runBiome`, and `main` all have `@param` and/or `@returns` JSDoc annotations.
-- `pnpm typecheck` exits 0 with zero errors.
-- No `@ts-ignore` directives in any file (only `@ts-expect-error` with comment if truly necessary).
-- No `.ts` files created. No `tsconfig.json` created.
-- `pnpm test` still passes.
-
-## Verification
-
-```sh
-# Confirm @ts-check in all .mjs source files
-grep -rL "@ts-check" scripts/ tests/
-# Expected: empty output (all files have it)
-
-# Confirm types.mjs exists
-ls scripts/lib/types.mjs
-
-# Run type check
-pnpm typecheck
-
-# Confirm no .ts files
-find scripts tests -name "*.ts"
-# Expected: empty
-
-# Confirm no @ts-ignore
-grep -rn "@ts-ignore" scripts/ tests/
-# Expected: 0 results
-
-# Run tests
-pnpm test
-```
-
-## Out of scope
-
-- Do not convert any `.mjs` file to `.ts`.
-- Do not create a `tsconfig.json` (inline flags are sufficient for this repo size).
-- Do not add `strict: false` to work around errors — fix the underlying issues.
-- Do not add `@ts-ignore` directives.
-- Do not add a runtime dependency or change the hook's runtime behaviour.
-- Do not type-check consumer files from the hook — that would require bundling `tsc` or requiring consumers to install TypeScript, which conflicts with the scope guard.
-
-## Follow-ups
-
-- Add `pnpm typecheck` to CI (`.github/workflows/ci.yml`) as a step alongside `pnpm test` — deferred to keep annotation churn and CI changes in separate commits.
-- Create a `tsconfig.json` if the inline `tsc` flags line grows past ~120 chars or if tests need different settings.
-- Consider `--noUncheckedIndexedAccess` after the codebase is stable under `--strict`.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/23-backfill-tags.md b/apps/claude-code/auto-format/docs/plans/23-backfill-tags.md
deleted file mode 100644
index 0278b078..00000000
--- a/apps/claude-code/auto-format/docs/plans/23-backfill-tags.md
+++ /dev/null
@@ -1,169 +0,0 @@
-# 23. Backfill historical git tags
-
-**Status: done — 2026-04-28**
-
-**Priority:** P2
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-14
-**Touches:** `scripts/backfill-tags.mjs` (new), `package.json`, `docs/plans/README.md`
-
-## Context
-
-Spec 14 landed `pnpm tag`, which tags the *current* HEAD. But the repo now has 23+
-committed version bumps (from `v0.1.0` through `v0.5.2`) with no git tags at all.
-Consumer repos that pin via a git URL reference bare commit SHAs rather than
-human-readable tags, making changelog navigation and upgrade traceability harder.
-
-A one-time backfill script walks the full git history, detects every commit where
-`.claude-plugin/plugin.json#version` first appeared or changed, and creates a
-lightweight `v{version}` tag at that commit. Subsequent runs are idempotent (skips
-tags that already exist).
-
-## Current behaviour
-
-`git tag --list` returns nothing. All 23+ version bump commits are untagged.
-
-## Target behaviour
-
-- `scripts/backfill-tags.mjs` exists and is valid ESM.
-- `pnpm backfill-tags` runs it.
-- For every distinct version encountered while walking git history oldest-to-newest,
-  a lightweight `v{version}` tag is created at the first commit that introduced that
-  version in `.claude-plugin/plugin.json`.
-- If a tag already exists it is skipped with a `skip` log line (idempotent).
-- Script ends with: `Done. Run: git push --tags`
-- Does **not** push. Push remains a manual step.
-
-## Implementation steps
-
-### Step 1 — Create `scripts/backfill-tags.mjs`
-
-```js
-#!/usr/bin/env node
-// @ts-check
-import { spawnSync } from 'node:child_process'
-
-const log = spawnSync('git', ['log', '--reverse', '--format=%H'], { encoding: 'utf8' })
-if (log.status !== 0) {
-	process.stderr.write('backfill-tags: git log failed\n')
-	process.exit(1)
-}
-
-const commits = log.stdout.trim().split('\n').filter(Boolean)
-let lastVersion = null
-
-for (const hash of commits) {
-	const show = spawnSync('git', ['show', `${hash}:.claude-plugin/plugin.json`], { encoding: 'utf8' })
-	if (show.status !== 0) continue
-
-	let version
-	try {
-		version = JSON.parse(show.stdout).version
-	} catch {
-		continue
-	}
-
-	if (!version || version === lastVersion) {
-		lastVersion = version
-		continue
-	}
-
-	lastVersion = version
-	const tagName = `v${version}`
-
-	const existing = spawnSync('git', ['tag', '-l', tagName], { encoding: 'utf8' })
-	if (existing.stdout.trim()) {
-		process.stdout.write(`  skip  ${tagName} (already exists)\n`)
-		continue
-	}
-
-	const result = spawnSync('git', ['tag', tagName, hash], { stdio: 'inherit' })
-	if (result.status === 0) {
-		process.stdout.write(`tagged  ${tagName} → ${hash.slice(0, 7)}\n`)
-	} else {
-		process.stderr.write(`  FAIL  ${tagName} → ${hash.slice(0, 7)}\n`)
-		process.exit(1)
-	}
-}
-
-process.stdout.write('Done. Run: git push --tags\n')
-```
-
-### Step 2 — Add `backfill-tags` to `package.json` scripts
-
-```json
-"scripts": {
-	"backfill-tags": "node scripts/backfill-tags.mjs",
-	"bump": "node scripts/bump.mjs",
-	"ralph": "ralph run -c ralph.yml -H builtin:code-assist",
-	"sync-version": "node scripts/sync-version.mjs",
-	"tag": "node scripts/tag.mjs",
-	"test": "node --test 'tests/*.test.mjs'",
-	"typecheck": "tsc --noEmit --allowJs --checkJs --strict --target ES2022 --module NodeNext --moduleResolution NodeNext scripts/*.mjs scripts/lib/*.mjs tests/*.mjs",
-	"verify:changelog": "node scripts/verify-changelog.mjs"
-}
-```
-
-### Step 3 — Update `docs/plans/README.md` execution table
-
-Add row:
-
-```
-| 23 | [Backfill historical git tags](./23-backfill-tags.md) | P2 | S | todo |
-```
-
-Add dependency note to cross-cutting dependencies:
-
-> **`23` → `14`**: backfill-tags.mjs relies on the same tag-naming convention as tag.mjs.
-
-### Step 4 — Commit
-
-```sh
-git add scripts/backfill-tags.mjs package.json docs/plans/README.md
-git commit -m "feat(spec-23): add pnpm backfill-tags to tag historical version commits"
-```
-
-## Test cases
-
-| Scenario | Expected |
-|---|---|
-| First run (no tags exist) | Tags created for every distinct version in history; exit 0 |
-| Second run (all tags exist) | All lines print `skip …`; exit 0 |
-| Partial run (some tags exist) | Missing tags created; existing skipped; exit 0 |
-| Repo has no `plugin.json` in history | Script exits 0; no tags created; no crash |
-
-## Acceptance criteria
-
-- `scripts/backfill-tags.mjs` is valid ESM with `// @ts-check`.
-- `pnpm backfill-tags` creates one `vX.Y.Z` tag per distinct version in git history.
-- Re-running is idempotent: skips existing tags, exits 0.
-- Script does **not** push.
-- Output ends with "Run: git push --tags".
-
-## Verification
-
-```sh
-# 1. Run backfill
-pnpm backfill-tags
-# expect: "tagged  vX.Y.Z → <hash>" lines, then "Done. Run: git push --tags"
-
-# 2. Confirm tags exist
-git tag --list | sort -V
-# expect: v0.1.0 through v0.5.2 (or whatever versions are in history)
-
-# 3. Confirm idempotent
-pnpm backfill-tags
-# expect: all lines say "skip  vX.Y.Z (already exists)"
-
-# 4. Optionally push
-git push --tags
-```
-
-## Out of scope
-
-- Annotated tags. Lightweight tags match spec-14 convention.
-- Pushing (always manual).
-- Validation that every bump commit has a matching tag (CI gate is spec-24).
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/24-ci-auto-tag.md b/apps/claude-code/auto-format/docs/plans/24-ci-auto-tag.md
deleted file mode 100644
index c1014531..00000000
--- a/apps/claude-code/auto-format/docs/plans/24-ci-auto-tag.md
+++ /dev/null
@@ -1,193 +0,0 @@
-# 24. CI auto-tag on version bump
-
-**Status: done — 2026-04-28**
-
-**Priority:** P2
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-10, spec-14
-**Touches:** `.github/workflows/release.yml` (new), `docs/plans/README.md`
-
-## Context
-
-Spec 14 landed `pnpm tag`, but tagging is currently a manual step requiring the
-developer to run `pnpm tag && git push --follow-tags` after every bump commit.
-It's easy to forget, leaving the repo with an untagged release until someone
-notices.
-
-A dedicated `release.yml` GitHub Actions workflow fires on every push to `main`,
-checks whether `.claude-plugin/plugin.json#version` changed relative to the
-previous commit, and if so creates and pushes the `v{version}` tag automatically.
-Because GitHub prevents workflow-triggered pushes from re-triggering the same
-workflow, there is no loop risk.
-
-The workflow is kept separate from `ci.yml` (spec-10) to preserve the
-single-responsibility principle: `ci.yml` validates, `release.yml` publishes.
-
-## Current behaviour
-
-After spec-10: `ci.yml` runs tests and changelog verification on every push/PR.
-No automated tagging exists.
-
-## Target behaviour
-
-- `.github/workflows/release.yml` exists.
-- Triggers on `push` to `main` only.
-- Has `permissions: contents: write` (required to push tags via `GITHUB_TOKEN`).
-- Detects a version bump by comparing `plugin.json#version` at HEAD vs HEAD~1.
-- If bumped:
-  1. Configures git identity as `github-actions[bot]`.
-  2. Runs `git tag v${VERSION} ${GITHUB_SHA}` to create the lightweight tag.
-  3. Runs `git push origin refs/tags/v${VERSION}` to publish the tag.
-  4. Logs: `Tagged and pushed v{version}`.
-- If **not** bumped: logs `No version bump detected — skipping tag` and exits 0.
-- If the tag already exists (re-run on the same commit): `git tag` exits 1 →
-  workflow step exits 1 → job fails visibly so the operator knows something is off.
-
-## Implementation steps
-
-### Step 1 — Create `.github/workflows/release.yml`
-
-```yaml
-name: Release
-
-on:
-  push:
-    branches:
-      - main
-
-permissions:
-  contents: write
-
-jobs:
-  tag:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          fetch-depth: 2   # HEAD and HEAD~1 needed for version comparison
-
-      - name: Read Node version
-        id: node-ver
-        run: |
-          VERSION=$(grep '^useNodeVersion:' pnpm-workspace.yaml | awk '{print $2}')
-          if [ -z "$VERSION" ]; then
-            echo "::error::useNodeVersion not found in pnpm-workspace.yaml" >&2
-            exit 1
-          fi
-          echo "version=$VERSION" >> "$GITHUB_OUTPUT"
-
-      - uses: pnpm/action-setup@v4
-        # reads packageManager from package.json — do not add version: here
-
-      - uses: actions/setup-node@v4
-        with:
-          node-version: ${{ steps.node-ver.outputs.version }}
-          cache: "pnpm"
-
-      - name: Detect version bump
-        id: bump
-        run: |
-          CURRENT=$(node -e "const fs=require('fs'); console.log(JSON.parse(fs.readFileSync('.claude-plugin/plugin.json','utf8')).version)")
-          PREVIOUS=$(git show HEAD~1:.claude-plugin/plugin.json 2>/dev/null \
-            | node -e "const fs=require('fs'); console.log(JSON.parse(fs.readFileSync('/dev/stdin','utf8')).version)" \
-            2>/dev/null || echo "")
-          if [ "$CURRENT" != "$PREVIOUS" ] && [ -n "$CURRENT" ]; then
-            echo "bumped=true"  >> "$GITHUB_OUTPUT"
-            echo "version=$CURRENT" >> "$GITHUB_OUTPUT"
-          else
-            echo "bumped=false" >> "$GITHUB_OUTPUT"
-          fi
-
-      - name: Skip (no bump)
-        if: steps.bump.outputs.bumped == 'false'
-        run: echo "No version bump detected — skipping tag"
-
-      - name: Install dependencies
-        if: steps.bump.outputs.bumped == 'true'
-        run: pnpm install --frozen-lockfile
-
-      - name: Create and push tag
-        if: steps.bump.outputs.bumped == 'true'
-        env:
-          VERSION: ${{ steps.bump.outputs.version }}
-        run: |
-          set -euo pipefail
-          git config user.name  "github-actions[bot]"
-          git config user.email "github-actions[bot]@users.noreply.github.com"
-          git tag "v${VERSION}" "${GITHUB_SHA}"
-          git push origin "refs/tags/v${VERSION}"
-          echo "Tagged and pushed v${VERSION}"
-```
-
-### Step 2 — Update `docs/plans/README.md` execution table
-
-Add row:
-
-```
-| 24 | [CI auto-tag on version bump](./24-ci-auto-tag.md) | P2 | S | todo |
-```
-
-Add dependency note to cross-cutting dependencies:
-
-> **`24` → `10` + `14`**: release.yml extends the CI setup from spec `10` and uses
-> the same tag convention as `tag.mjs` from spec `14`.
-
-### Step 3 — Commit
-
-```sh
-git add .github/workflows/release.yml docs/plans/README.md
-git commit -m "ci(spec-24): add release workflow to auto-tag version bump commits on main"
-```
-
-## Test cases
-
-| Scenario | Expected |
-|---|---|
-| Push to `main` with version bump in `plugin.json` | `git tag v${VERSION}` runs; tag pushed; step logs `Tagged and pushed v${VERSION}` |
-| Push to `main` without version bump | `bumped=false`; step logs `No version bump detected`; job exits 0 |
-| Push to a non-`main` branch | Workflow does not trigger |
-| Tag already exists for current version | `git tag` exits 1; job fails visibly |
-| `plugin.json` absent in HEAD~1 (first ever commit on main) | `PREVIOUS` falls back to `""`; treated as a bump; tag created |
-
-## Acceptance criteria
-
-- `.github/workflows/release.yml` is valid YAML.
-- Workflow triggers only on `push` to `main`.
-- `permissions.contents` is `write`.
-- `actions/checkout` uses `fetch-depth: 2`.
-- `pnpm/action-setup` has no `version:` input.
-- Version comparison uses `HEAD~1` diff, not commit message matching.
-- Tag is pushed via `git push origin refs/tags/vX.Y.Z` (not `--follow-tags`, which only pushes annotated tags).
-- Non-bump pushes exit 0 without creating a tag.
-
-## Verification
-
-```sh
-# 1. Valid YAML
-python3 -c "import yaml; yaml.safe_load(open('.github/workflows/release.yml'))" && echo "OK"
-
-# 2. Confirm fetch-depth: 2 is present
-grep "fetch-depth" .github/workflows/release.yml
-# expect: fetch-depth: 2
-
-# 3. Confirm no version: in pnpm/action-setup block
-grep -A5 "pnpm/action-setup" .github/workflows/release.yml | grep "version:" \
-  && echo "FAIL: version: present" || echo "OK: no version:"
-
-# 4. Integration test: bump patch, commit, push to main, observe Actions run
-pnpm bump patch
-git add -A && git commit -m "chore: bump for release workflow test"
-git push
-# expect: release.yml job creates and pushes vX.Y.Z tag
-```
-
-## Out of scope
-
-- Release notes or GitHub Releases creation (plugin is distributed via git URL,
-  not the Releases page).
-- Annotated tags.
-- Matrix of branches (only `main` is the release branch).
-- Rollback on failed tag push.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/25-formatter-descriptor-type.md b/apps/claude-code/auto-format/docs/plans/25-formatter-descriptor-type.md
deleted file mode 100644
index 6aa9c7aa..00000000
--- a/apps/claude-code/auto-format/docs/plans/25-formatter-descriptor-type.md
+++ /dev/null
@@ -1,103 +0,0 @@
-# 25. Add `FormatterDescriptor` typedef to `lib/types.mjs`
-**Status: done — 2026-05-04**
-
-**Priority:** P2
-**Effort:** XS
-**Version impact:** patch
-**Depends on:** spec-22
-**Touches:** `scripts/lib/types.mjs`
-
-## Context
-
-The upcoming `lib/runners.mjs` module (spec-26) needs a `FormatterDescriptor` type. Adding it to
-`lib/types.mjs` first keeps all type definitions centralised and lets spec-26 reference it via a
-JSDoc `@import` tag without any coupling to `format-hook.mjs`.
-
-## Current behaviour
-
-`scripts/lib/types.mjs` defines `HookEvent`, `FormatterName`, and `ProjectConfig`. There is no
-`FormatterDescriptor` type.
-
-## Target behaviour
-
-`lib/types.mjs` contains a `FormatterDescriptor` typedef after `ProjectConfig`:
-
-```js
-/**
- * Descriptor for a single formatter invocation.
- *
- * @typedef {{
- *   name: string,
- *   bin: string,
- *   args: (filePath: string) => string[],
- *   warnIfMissing?: boolean,
- *   toleratedStatuses?: number[],
- * }} FormatterDescriptor
- */
-```
-
-Fields:
-- `name` — human-readable label used in stderr messages (e.g. `"prettier"`)
-- `bin` — absolute path to the formatter binary
-- `args` — function that receives an absolute file path and returns the full argument array
-- `warnIfMissing` — when true, emit a stderr warning if `bin` does not exist; defaults false
-- `toleratedStatuses` — exit codes not treated as failures; defaults `[]`
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `scripts/lib/types.mjs` | Add `FormatterDescriptor` typedef after `ProjectConfig` |
-
-## Implementation steps
-
-### Step 1 — Add typedef to `scripts/lib/types.mjs`
-
-Open `scripts/lib/types.mjs`. After the closing `*/` of the `ProjectConfig` typedef, add:
-
-```js
-/**
- * Descriptor for a single formatter invocation.
- *
- * @typedef {{
- *   name: string,
- *   bin: string,
- *   args: (filePath: string) => string[],
- *   warnIfMissing?: boolean,
- *   toleratedStatuses?: number[],
- * }} FormatterDescriptor
- */
-```
-
-Leave the `export {}` line at the end of the file in place.
-
-### Step 2 — Commit
-
-```sh
-git add scripts/lib/types.mjs
-git commit -m "chore(spec-25): add FormatterDescriptor typedef to lib/types.mjs"
-```
-
-## Verification
-
-```sh
-# Typedef present
-grep -n "FormatterDescriptor" scripts/lib/types.mjs
-
-# Type-check passes
-pnpm typecheck
-```
-
-## Acceptance criteria
-
-- [ ] `FormatterDescriptor` typedef exists in `scripts/lib/types.mjs`
-- [ ] All five fields present with correct types
-- [ ] `pnpm typecheck` passes
-- [ ] No other files modified
-
-## Out of scope
-
-- Runtime exports
-- Changes to any file other than `scripts/lib/types.mjs`
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/26-runner-module.md b/apps/claude-code/auto-format/docs/plans/26-runner-module.md
deleted file mode 100644
index f7cb744f..00000000
--- a/apps/claude-code/auto-format/docs/plans/26-runner-module.md
+++ /dev/null
@@ -1,277 +0,0 @@
-# 26. Extract `lib/runners.mjs` with `runFormatter` and tests
-**Status: done — 2026-05-04**
-
-**Priority:** P2
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-25
-**Touches:** `scripts/lib/runners.mjs` (new), `scripts/lib/runners.test.mjs` (new), `package.json`
-
-## Context
-
-The three formatter runner functions in `format-hook.mjs` share identical subprocess logic
-duplicated three times. See PRD at `docs/issues/auto-format-runners/PRD.md`.
-
-This spec creates the generic `runFormatter` module and its tests. `format-hook.mjs` is not
-modified here — that is spec-27.
-
-## Current behaviour
-
-No `scripts/lib/runners.mjs` exists. `spawnSync` contract is duplicated across `runPrettier`,
-`runEslint`, `runBiome` in `format-hook.mjs`. Subprocess timeout and signal behaviour cannot be
-tested without running the full hook.
-
-## Target behaviour
-
-`scripts/lib/runners.mjs` exports `runFormatter(descriptor, filePath, cwd, timeoutMs)`.
-`scripts/lib/runners.test.mjs` has six passing tests covering the subprocess contract.
-`pnpm test` runs both test files.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `scripts/lib/runners.mjs` | New — the `runFormatter` function |
-| `scripts/lib/runners.test.mjs` | New — six tests |
-| `package.json` | Update `scripts.test` to include the new test file |
-
-## Implementation steps
-
-### Step 1 — Create `scripts/lib/runners.mjs`
-
-```js
-// @ts-check
-// SPDX-License-Identifier: LGPL-3.0-or-later
-// Copyright © 2026 Unic
-/** @import { FormatterDescriptor } from './types.mjs' */
-
-import { spawnSync } from 'node:child_process'
-import { existsSync } from 'node:fs'
-
-/**
- * Invokes a formatter binary as a child process, handling timeout, signal, and exit-code
- * reporting. Always returns — never throws. Diagnostics go to stderr only.
- *
- * @param {FormatterDescriptor} descriptor
- * @param {string} filePath - Absolute path of the file to format.
- * @param {string} cwd - Working directory for the child process.
- * @param {number} timeoutMs - Milliseconds before the process is sent SIGTERM.
- * @returns {void}
- */
-export function runFormatter(descriptor, filePath, cwd, timeoutMs) {
-	if (!existsSync(descriptor.bin)) {
-		if (descriptor.warnIfMissing)
-			process.stderr.write(`unic-format: ${descriptor.name} binary not found at ${descriptor.bin}\n`)
-		return
-	}
-	const r = spawnSync('node', [descriptor.bin, ...descriptor.args(filePath)], {
-		cwd,
-		stdio: ['ignore', 'ignore', 'pipe'],
-		timeout: timeoutMs,
-		killSignal: 'SIGTERM',
-	})
-	if (r.signal === 'SIGTERM' || r.status === null) {
-		process.stderr.write(`unic-format: ${descriptor.name} timed out after ${timeoutMs / 1000}s on ${filePath}\n`)
-		return
-	}
-	const tolerated = descriptor.toleratedStatuses ?? []
-	if (r.status !== 0 && !tolerated.includes(r.status))
-		process.stderr.write(
-			`unic-format: ${descriptor.name} failed (exit ${r.status}): ${r.stderr?.toString().trim() || 'unknown error'}\n`,
-		)
-}
-```
-
-### Step 2 — Create `scripts/lib/runners.test.mjs`
-
-```js
-// @ts-check
-import { strictEqual, ok } from 'node:assert/strict'
-import { existsSync, mkdtempSync, rmSync, writeFileSync } from 'node:fs'
-import { tmpdir } from 'node:os'
-import { join } from 'node:path'
-import { test } from 'node:test'
-import { runFormatter } from './runners.mjs'
-
-/**
- * Writes a stub Node.js script to a temp directory and returns its path.
- * Caller is responsible for rmSync(dir) in a finally block.
- *
- * @param {string} script - Body of the stub (executed by node).
- * @returns {{ binPath: string, dir: string }}
- */
-function makeStub(script) {
-	const dir = mkdtempSync(join(tmpdir(), 'runners-test-'))
-	const binPath = join(dir, 'stub.mjs')
-	writeFileSync(binPath, script)
-	return { binPath, dir }
-}
-
-/**
- * Intercepts process.stderr.write for the duration of fn, returns captured lines.
- *
- * @param {() => void} fn
- * @returns {string[]}
- */
-function captureStderr(fn) {
-	/** @type {string[]} */
-	const lines = []
-	const orig = process.stderr.write.bind(process.stderr)
-	// @ts-ignore — overriding for test purposes
-	process.stderr.write = (/** @type {string} */ s) => { lines.push(String(s)); return true }
-	try { fn() } finally { process.stderr.write = orig }
-	return lines
-}
-
-test('silent when binary is missing and warnIfMissing is false (or omitted)', () => {
-	const lines = captureStderr(() => {
-		runFormatter(
-			{ name: 'fake', bin: '/nonexistent/bin/fake', args: (f) => [f] },
-			'/some/file.ts', '/tmp', 5_000,
-		)
-	})
-	strictEqual(lines.length, 0)
-})
-
-test('warns when binary is missing and warnIfMissing is true', () => {
-	const lines = captureStderr(() => {
-		runFormatter(
-			{ name: 'fake', bin: '/nonexistent/bin/fake', args: (f) => [f], warnIfMissing: true },
-			'/some/file.ts', '/tmp', 5_000,
-		)
-	})
-	strictEqual(lines.length, 1)
-	ok(lines[0].includes('binary not found'), `expected "binary not found" in: ${lines[0]}`)
-})
-
-test('reports timeout when process exceeds timeoutMs', () => {
-	const { binPath, dir } = makeStub('setInterval(() => {}, 999)')
-	try {
-		const lines = captureStderr(() => {
-			runFormatter(
-				{ name: 'slow', bin: binPath, args: (f) => [f] },
-				'/some/file.ts', dir, 50,
-			)
-		})
-		strictEqual(lines.length, 1)
-		ok(lines[0].includes('timed out'), `expected "timed out" in: ${lines[0]}`)
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-
-test('silent for exit codes in toleratedStatuses', () => {
-	const { binPath, dir } = makeStub('process.exit(1)')
-	try {
-		const lines = captureStderr(() => {
-			runFormatter(
-				{ name: 'eslint', bin: binPath, args: (f) => [f], toleratedStatuses: [1] },
-				'/some/file.ts', dir, 5_000,
-			)
-		})
-		strictEqual(lines.length, 0)
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-
-test('reports failure for non-tolerated non-zero exit code', () => {
-	const { binPath, dir } = makeStub('process.exit(2)')
-	try {
-		const lines = captureStderr(() => {
-			runFormatter(
-				{ name: 'biome', bin: binPath, args: (f) => [f] },
-				'/some/file.ts', dir, 5_000,
-			)
-		})
-		strictEqual(lines.length, 1)
-		ok(lines[0].includes('failed (exit 2)'), `expected "failed (exit 2)" in: ${lines[0]}`)
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-
-test('passes args(filePath) to the subprocess', () => {
-	const { binPath, dir } = makeStub(
-		`import { writeFileSync } from 'node:fs'\nwriteFileSync(process.argv[2] + '.called', '1')\n`,
-	)
-	const targetFile = join(dir, 'target.ts')
-	writeFileSync(targetFile, '')
-	try {
-		captureStderr(() => {
-			runFormatter(
-				{ name: 'sentinel', bin: binPath, args: (f) => [f] },
-				targetFile, dir, 5_000,
-			)
-		})
-		ok(existsSync(targetFile + '.called'), 'stub should have written the sentinel file')
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-```
-
-### Step 3 — Update `package.json` test script
-
-Change:
-
-```json
-"test": "node --test tests/format-hook.test.mjs"
-```
-
-to:
-
-```json
-"test": "node --test tests/format-hook.test.mjs scripts/lib/runners.test.mjs"
-```
-
-### Step 4 — Verify
-
-```sh
-node --test scripts/lib/runners.test.mjs
-pnpm test
-pnpm typecheck
-```
-
-### Step 5 — Commit
-
-```sh
-git add scripts/lib/runners.mjs scripts/lib/runners.test.mjs package.json
-git commit -m "chore(spec-26): extract runFormatter to lib/runners.mjs with tests"
-```
-
-## Verification
-
-```sh
-# Module exports runFormatter
-node --input-type=module <<'EOF'
-import { runFormatter } from './scripts/lib/runners.mjs'
-console.log(typeof runFormatter)
-EOF
-
-# New tests pass
-node --test scripts/lib/runners.test.mjs
-
-# Full suite passes
-pnpm test
-
-# Type-check passes
-pnpm typecheck
-```
-
-## Acceptance criteria
-
-- [ ] `scripts/lib/runners.mjs` exists and exports `runFormatter`
-- [ ] `runFormatter` signature: `(descriptor, filePath, cwd, timeoutMs) => void`
-- [ ] All 6 tests in `scripts/lib/runners.test.mjs` pass
-- [ ] `pnpm test` passes (existing `format-hook` tests unaffected)
-- [ ] `pnpm typecheck` passes
-- [ ] `format-hook.mjs` is not modified in this spec
-
-## Out of scope
-
-- Changes to `format-hook.mjs` (spec-27)
-- Async subprocess execution
-- Configurable binary resolution
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/27-replace-runner-functions.md b/apps/claude-code/auto-format/docs/plans/27-replace-runner-functions.md
deleted file mode 100644
index e32d27e8..00000000
--- a/apps/claude-code/auto-format/docs/plans/27-replace-runner-functions.md
+++ /dev/null
@@ -1,154 +0,0 @@
-# 27. Replace runner functions with descriptors in `format-hook.mjs`
-**Status: done — 2026-05-04**
-
-**Priority:** P2
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-26
-**Touches:** `scripts/format-hook.mjs`
-
-## Context
-
-`lib/runners.mjs` and `FormatterDescriptor` exist after spec-26. This spec wires them into
-`format-hook.mjs` by replacing the three inline runner functions with three descriptor constants
-and a single call-through to `runFormatter`. External behaviour is identical.
-
-## Current behaviour
-
-`format-hook.mjs` defines `runPrettier`, `runEslint`, `runBiome` — three functions that duplicate
-the same `spawnSync` contract. `main()` calls them directly.
-
-## Target behaviour
-
-`format-hook.mjs` defines `PRETTIER_DESCRIPTOR`, `ESLINT_DESCRIPTOR`, `BIOME_DESCRIPTOR` constants.
-`main()` calls `runFormatter(descriptor, filePath, PROJECT_DIR, CONFIG.formatTimeoutMs)`.
-The three runner functions are deleted.
-
-`tests/format-hook.test.mjs` passes without modification.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `scripts/format-hook.mjs` | Remove three runner functions; add descriptors; update imports and dispatch |
-
-## Implementation steps
-
-### Step 1 — Add imports at the top of `format-hook.mjs`
-
-After the existing `import` block, add:
-
-```js
-import { runFormatter } from './lib/runners.mjs'
-```
-
-Add a type-import tag alongside the existing ones:
-
-```js
-/** @import { FormatterDescriptor } from './lib/types.mjs' */
-```
-
-### Step 2 — Add descriptor constants
-
-After the `BIOME_AVAILABLE` constant and before `toPosix`, add:
-
-```js
-/** @type {FormatterDescriptor} */
-const PRETTIER_DESCRIPTOR = {
-	name: 'prettier',
-	bin: PRETTIER_BIN,
-	args: (f) => ['--write', '--ignore-unknown', '--log-level', 'warn', f],
-}
-
-/** @type {FormatterDescriptor} */
-const ESLINT_DESCRIPTOR = {
-	name: 'eslint',
-	bin: ESLINT_BIN,
-	args: (f) => ['--fix', '--no-error-on-unmatched-pattern', f],
-	toleratedStatuses: [1],
-}
-
-/** @type {FormatterDescriptor} */
-const BIOME_DESCRIPTOR = {
-	name: 'biome',
-	bin: BIOME_BIN,
-	args: (f) => ['check', '--write', '--no-errors-on-unmatched-pattern', f],
-	warnIfMissing: true,
-}
-```
-
-### Step 3 — Update `main()` dispatch
-
-Replace the current formatter dispatch in `main()`:
-
-```js
-if (usesBiome) {
-	runBiome(filePath)
-} else {
-	runPrettier(filePath)
-	if (ESLINT_EXTS.has(ext)) runEslint(filePath)
-}
-```
-
-with:
-
-```js
-const run = (/** @type {FormatterDescriptor} */ d) => runFormatter(d, filePath, PROJECT_DIR, CONFIG.formatTimeoutMs)
-if (usesBiome) {
-	run(BIOME_DESCRIPTOR)
-} else {
-	run(PRETTIER_DESCRIPTOR)
-	if (ESLINT_EXTS.has(ext)) run(ESLINT_DESCRIPTOR)
-}
-```
-
-### Step 4 — Delete the three runner functions
-
-Delete the entire bodies of `runPrettier` (lines ~137–152), `runEslint` (lines ~161–180), and
-`runBiome` (lines ~189–209) including their JSDoc comments. Verify no other references remain:
-
-```sh
-grep -n "runPrettier\|runEslint\|runBiome" scripts/format-hook.mjs
-```
-
-Expected: 0 matches.
-
-### Step 5 — Verify and commit
-
-```sh
-pnpm test
-pnpm typecheck
-git add scripts/format-hook.mjs
-git commit -m "refactor(spec-27): replace runner functions with FormatterDescriptor constants"
-```
-
-## Verification
-
-```sh
-# No old runner functions remain
-grep -n "function runPrettier\|function runEslint\|function runBiome" scripts/format-hook.mjs
-
-# Three descriptors present
-grep -n "PRETTIER_DESCRIPTOR\|ESLINT_DESCRIPTOR\|BIOME_DESCRIPTOR" scripts/format-hook.mjs
-
-# Full test suite passes (no changes to test file)
-pnpm test
-
-# Type-check passes
-pnpm typecheck
-```
-
-## Acceptance criteria
-
-- [ ] `grep -n "function runPrettier\|function runEslint\|function runBiome" scripts/format-hook.mjs` → 0 matches
-- [ ] `grep -n "PRETTIER_DESCRIPTOR\|ESLINT_DESCRIPTOR\|BIOME_DESCRIPTOR" scripts/format-hook.mjs` → 3 matches
-- [ ] `grep -n "runFormatter" scripts/format-hook.mjs` → at least 1 match
-- [ ] `pnpm test` passes with zero changes to `tests/format-hook.test.mjs`
-- [ ] `pnpm typecheck` passes
-
-## Out of scope
-
-- Changing skip logic, extension sets, config loading, or any other behaviour
-- Modifying test files
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/28-version-bump.md b/apps/claude-code/auto-format/docs/plans/28-version-bump.md
deleted file mode 100644
index 29522981..00000000
--- a/apps/claude-code/auto-format/docs/plans/28-version-bump.md
+++ /dev/null
@@ -1,84 +0,0 @@
-# 28. Version bump and CHANGELOG entry
-
-**Status: done — 2026-05-04**
-
-**Priority:** P2
-**Effort:** XS
-**Version impact:** patch
-**Depends on:** spec-27
-**Touches:** `.claude-plugin/plugin.json`, `.claude-plugin/marketplace.json`, `package.json`, `CHANGELOG.md`
-
-## Context
-
-Specs 25–27 collectively extract the formatter runner into a dedicated module. This is an
-internal refactor with no user-visible behaviour change. A patch bump is appropriate.
-
-## Current behaviour
-
-Version reflects the state before specs 25–27.
-
-## Target behaviour
-
-Version incremented by one patch level. CHANGELOG has a dated entry under the new version
-describing the internal refactor.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `.claude-plugin/plugin.json` | Version bumped |
-| `.claude-plugin/marketplace.json` | Synced by `pnpm bump` |
-| `package.json` | Synced by `pnpm bump` |
-| `CHANGELOG.md` | New dated entry |
-
-## Implementation steps
-
-### Step 1 — Add CHANGELOG entry to `[Unreleased]`
-
-Before running `pnpm bump`, add an entry under `## [Unreleased]` in `CHANGELOG.md`:
-
-```markdown
-### Changed
-- Internal: extracted shared subprocess contract from `runPrettier`/`runEslint`/`runBiome` into
-  `lib/runners.mjs` — no behaviour change for consumers.
-```
-
-### Step 2 — Run bump
-
-```sh
-pnpm bump patch
-```
-
-This increments the version in `plugin.json` and syncs `marketplace.json` and `package.json`.
-
-### Step 3 — Verify
-
-```sh
-pnpm verify:changelog
-```
-
-### Step 4 — Commit
-
-```sh
-git add .claude-plugin/plugin.json .claude-plugin/marketplace.json package.json CHANGELOG.md
-git commit -m "chore(release): bump auto-format patch version — runner extraction refactor"
-```
-
-## Verification
-
-```sh
-pnpm verify:changelog
-grep -n "lib/runners" CHANGELOG.md
-```
-
-## Acceptance criteria
-
-- [ ] `pnpm verify:changelog` passes
-- [ ] CHANGELOG entry present for the new version with description of the refactor
-- [ ] `plugin.json`, `marketplace.json`, `package.json` versions are in sync
-
-## Out of scope
-
-- Git tag (create separately with `pnpm tag` when ready to release)
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/29-config-module.md b/apps/claude-code/auto-format/docs/plans/29-config-module.md
deleted file mode 100644
index d886f665..00000000
--- a/apps/claude-code/auto-format/docs/plans/29-config-module.md
+++ /dev/null
@@ -1,339 +0,0 @@
-# 29. Extract `lib/config.mjs` with `DEFAULTS`, `loadConfig`, and tests
-**Status: done — 2026-05-04**
-
-**Priority:** P2
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec-22
-**Touches:** `scripts/lib/config.mjs` (new), `scripts/lib/config.test.mjs` (new), `package.json`
-
-## Context
-
-`loadProjectConfig()` in `format-hook.mjs` closes over the module-level `PROJECT_DIR` constant,
-making the merge logic untestable in isolation. `DEFAULTS` (the full config contract) lives in the
-hook entry point with no clear seam. See PRD at `docs/issues/auto-format-config/PRD.md`.
-
-This spec creates `lib/config.mjs` and its tests. `format-hook.mjs` is unchanged — that is
-spec-30.
-
-## Current behaviour
-
-No `scripts/lib/config.mjs` exists. Config defaults and merge logic are inline in `format-hook.mjs`
-with no unit tests for the merge strategy.
-
-## Target behaviour
-
-`scripts/lib/config.mjs` exports `DEFAULTS` and `loadConfig(projectDir)`.
-`scripts/lib/config.test.mjs` has ten passing tests.
-`pnpm test` runs both the existing and new test files.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `scripts/lib/config.mjs` | New — `DEFAULTS` constant + `loadConfig` function |
-| `scripts/lib/config.test.mjs` | New — ten tests |
-| `package.json` | Update `scripts.test` to include the new test file |
-
-## Implementation steps
-
-### Step 1 — Create `scripts/lib/config.mjs`
-
-The content is derived from `format-hook.mjs` lines 25–93, with two changes:
-- `DEFAULTS` is exported
-- `loadProjectConfig()` is renamed `loadConfig(projectDir)` and the `PROJECT_DIR` reference is
-  replaced by the `projectDir` parameter
-
-```js
-// @ts-check
-// SPDX-License-Identifier: LGPL-3.0-or-later
-// Copyright © 2026 Unic
-/** @import { ProjectConfig, FormatterName } from './types.mjs' */
-
-import { existsSync, readFileSync } from 'node:fs'
-import { resolve } from 'node:path'
-
-/** @type {ProjectConfig} */
-export const DEFAULTS = {
-	skipPrefixes: [
-		'_bmad/',
-		'.claude/skills/bmad-',
-		'.claude/worktrees/',
-		'.history/',
-		'.git/',
-		'node_modules/',
-		'dist/',
-		'build/',
-		'.next/',
-		'coverage/',
-	],
-	prettierExtensions: [
-		'.md',
-		'.mdx',
-		'.json',
-		'.jsonc',
-		'.yml',
-		'.yaml',
-		'.js',
-		'.mjs',
-		'.cjs',
-		'.ts',
-		'.mts',
-		'.cts',
-		'.tsx',
-		'.feature',
-	],
-	eslintExtensions: ['.js', '.mjs', '.cjs', '.ts', '.mts', '.cts', '.tsx', '.json', '.jsonc', '.md'],
-	formatTimeoutMs: 30_000,
-	formatter: 'auto',
-}
-
-const VALID_FORMATTERS = new Set(['auto', 'prettier', 'biome'])
-
-/**
- * Reads `.claude/unic-format.json` from `projectDir` and merges it with DEFAULTS.
- * Returns DEFAULTS on missing file or parse error.
- *
- * @param {string} projectDir - Absolute path to the consumer project root.
- * @returns {ProjectConfig}
- */
-export function loadConfig(projectDir) {
-	const configPath = resolve(projectDir, '.claude/unic-format.json')
-	if (!existsSync(configPath)) return DEFAULTS
-	try {
-		const cfg = /** @type {Record<string, unknown>} */ (JSON.parse(readFileSync(configPath, 'utf8')))
-		const raw = Number(cfg.formatTimeoutMs)
-		const hasFullReplacement = Array.isArray(cfg.skipPrefixes) && cfg.skipPrefixes.length > 0
-		const hasAdditive = Array.isArray(cfg.additionalSkipPrefixes) && cfg.additionalSkipPrefixes.length > 0
-		return {
-			skipPrefixes: hasFullReplacement
-				? /** @type {string[]} */ (cfg.skipPrefixes)
-				: hasAdditive
-					? [...DEFAULTS.skipPrefixes, .../** @type {string[]} */ (cfg.additionalSkipPrefixes)]
-					: DEFAULTS.skipPrefixes,
-			prettierExtensions: Array.isArray(cfg.prettierExtensions) ? cfg.prettierExtensions : DEFAULTS.prettierExtensions,
-			eslintExtensions: Array.isArray(cfg.eslintExtensions) ? cfg.eslintExtensions : DEFAULTS.eslintExtensions,
-			formatTimeoutMs: Number.isFinite(raw) ? Math.min(Math.max(raw, 1_000), 120_000) : DEFAULTS.formatTimeoutMs,
-			formatter: /** @type {FormatterName} */ (
-				VALID_FORMATTERS.has(/** @type {string} */ (cfg.formatter)) ? cfg.formatter : DEFAULTS.formatter
-			),
-		}
-	} catch (err) {
-		process.stderr.write(
-			`unic-format: ignoring malformed .claude/unic-format.json: ${/** @type {Error} */ (err).message}\n`,
-		)
-		return DEFAULTS
-	}
-}
-```
-
-### Step 2 — Create `scripts/lib/config.test.mjs`
-
-```js
-// @ts-check
-import { deepStrictEqual, strictEqual, ok } from 'node:assert/strict'
-import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from 'node:fs'
-import { tmpdir } from 'node:os'
-import { join } from 'node:path'
-import { test } from 'node:test'
-import { DEFAULTS, loadConfig } from './config.mjs'
-
-/**
- * Creates a temp directory to act as the consumer project root.
- * Caller must rmSync(dir) in a finally block.
- *
- * @returns {string}
- */
-function makeProjectDir() {
-	return mkdtempSync(join(tmpdir(), 'config-test-'))
-}
-
-/**
- * Writes `.claude/unic-format.json` into dir with the given content.
- *
- * @param {string} dir
- * @param {unknown} content
- */
-function writeConfig(dir, content) {
-	mkdirSync(join(dir, '.claude'), { recursive: true })
-	writeFileSync(join(dir, '.claude', 'unic-format.json'), JSON.stringify(content))
-}
-
-test('returns DEFAULTS when no config file exists', () => {
-	const dir = makeProjectDir()
-	try {
-		deepStrictEqual(loadConfig(dir), DEFAULTS)
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-
-test('returns DEFAULTS and writes stderr on malformed JSON', () => {
-	const dir = makeProjectDir()
-	const stderrLines = /** @type {string[]} */ ([])
-	const orig = process.stderr.write.bind(process.stderr)
-	// @ts-ignore
-	process.stderr.write = (/** @type {string} */ s) => { stderrLines.push(String(s)); return true }
-	try {
-		mkdirSync(join(dir, '.claude'), { recursive: true })
-		writeFileSync(join(dir, '.claude', 'unic-format.json'), '{not valid json')
-		deepStrictEqual(loadConfig(dir), DEFAULTS)
-		strictEqual(stderrLines.length, 1)
-		ok(stderrLines[0].includes('malformed'))
-	} finally {
-		process.stderr.write = orig
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-
-test('full skipPrefixes replacement when non-empty array provided', () => {
-	const dir = makeProjectDir()
-	try {
-		writeConfig(dir, { skipPrefixes: ['custom/'] })
-		const cfg = loadConfig(dir)
-		deepStrictEqual(cfg.skipPrefixes, ['custom/'])
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-
-test('additionalSkipPrefixes appends to DEFAULTS when no skipPrefixes given', () => {
-	const dir = makeProjectDir()
-	try {
-		writeConfig(dir, { additionalSkipPrefixes: ['extra/'] })
-		const cfg = loadConfig(dir)
-		deepStrictEqual(cfg.skipPrefixes, [...DEFAULTS.skipPrefixes, 'extra/'])
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-
-test('uses DEFAULTS.skipPrefixes when neither field is provided', () => {
-	const dir = makeProjectDir()
-	try {
-		writeConfig(dir, {})
-		const cfg = loadConfig(dir)
-		deepStrictEqual(cfg.skipPrefixes, DEFAULTS.skipPrefixes)
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-
-test('clamps formatTimeoutMs to minimum 1000', () => {
-	const dir = makeProjectDir()
-	try {
-		writeConfig(dir, { formatTimeoutMs: 100 })
-		strictEqual(loadConfig(dir).formatTimeoutMs, 1_000)
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-
-test('clamps formatTimeoutMs to maximum 120000', () => {
-	const dir = makeProjectDir()
-	try {
-		writeConfig(dir, { formatTimeoutMs: 999_999 })
-		strictEqual(loadConfig(dir).formatTimeoutMs, 120_000)
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-
-test('accepts valid formatTimeoutMs within range', () => {
-	const dir = makeProjectDir()
-	try {
-		writeConfig(dir, { formatTimeoutMs: 15_000 })
-		strictEqual(loadConfig(dir).formatTimeoutMs, 15_000)
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-
-test('uses DEFAULTS.formatter for invalid formatter value', () => {
-	const dir = makeProjectDir()
-	try {
-		writeConfig(dir, { formatter: 'webpack' })
-		strictEqual(loadConfig(dir).formatter, DEFAULTS.formatter)
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-
-test('accepts valid formatter value "biome"', () => {
-	const dir = makeProjectDir()
-	try {
-		writeConfig(dir, { formatter: 'biome' })
-		strictEqual(loadConfig(dir).formatter, 'biome')
-	} finally {
-		rmSync(dir, { recursive: true, force: true })
-	}
-})
-```
-
-### Step 3 — Update `package.json` test script
-
-Change:
-
-```json
-"test": "node --test tests/format-hook.test.mjs scripts/lib/runners.test.mjs"
-```
-
-to:
-
-```json
-"test": "node --test tests/format-hook.test.mjs scripts/lib/runners.test.mjs scripts/lib/config.test.mjs"
-```
-
-Note: if spec-26 has not yet landed, the test command before this spec will be
-`node --test tests/format-hook.test.mjs` — update accordingly.
-
-### Step 4 — Verify
-
-```sh
-node --test scripts/lib/config.test.mjs
-pnpm test
-pnpm typecheck
-```
-
-### Step 5 — Commit
-
-```sh
-git add scripts/lib/config.mjs scripts/lib/config.test.mjs package.json
-git commit -m "chore(spec-29): extract loadConfig and DEFAULTS to lib/config.mjs with tests"
-```
-
-## Verification
-
-```sh
-# Exports present
-node --input-type=module <<'EOF'
-import { DEFAULTS, loadConfig } from './scripts/lib/config.mjs'
-console.log(typeof DEFAULTS, typeof loadConfig)
-EOF
-
-# New tests pass
-node --test scripts/lib/config.test.mjs
-
-# Full suite passes
-pnpm test
-
-# Type-check passes
-pnpm typecheck
-```
-
-## Acceptance criteria
-
-- [ ] `scripts/lib/config.mjs` exports `DEFAULTS` and `loadConfig`
-- [ ] `loadConfig` accepts `projectDir: string` — no reference to `process.env` or `process.cwd()`
-- [ ] All 10 tests in `scripts/lib/config.test.mjs` pass
-- [ ] `pnpm test` passes (existing tests unaffected)
-- [ ] `pnpm typecheck` passes
-- [ ] `format-hook.mjs` is not modified
-
-## Out of scope
-
-- Changes to `format-hook.mjs` (spec-30)
-- Changing the config schema or DEFAULTS values
-- Exporting `VALID_FORMATTERS`
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/30-wire-up-config-module.md b/apps/claude-code/auto-format/docs/plans/30-wire-up-config-module.md
deleted file mode 100644
index de12e38c..00000000
--- a/apps/claude-code/auto-format/docs/plans/30-wire-up-config-module.md
+++ /dev/null
@@ -1,118 +0,0 @@
-# 30. Update `format-hook.mjs` to use `lib/config.mjs`
-**Status: done — 2026-05-04**
-
-**Priority:** P2
-**Effort:** XS
-**Version impact:** patch
-**Depends on:** spec-29
-**Touches:** `scripts/format-hook.mjs`
-
-## Context
-
-`lib/config.mjs` exists after spec-29 and exports `DEFAULTS` and `loadConfig`. This spec removes
-the now-duplicated code from `format-hook.mjs` and wires in the new module. External behaviour
-is unchanged.
-
-## Current behaviour
-
-`format-hook.mjs` defines `DEFAULTS` (lines 25–57) and `loadProjectConfig()` (lines 65–93).
-`const CONFIG = loadProjectConfig()` calls the inline function.
-
-## Target behaviour
-
-`format-hook.mjs` imports `loadConfig` from `./lib/config.mjs`. The `DEFAULTS` constant and
-`loadProjectConfig` function are deleted. `const CONFIG = loadConfig(PROJECT_DIR)`.
-
-`tests/format-hook.test.mjs` passes without modification.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `scripts/format-hook.mjs` | Remove `DEFAULTS` + `loadProjectConfig`; add import; update call site |
-
-## Implementation steps
-
-### Step 1 — Add import
-
-At the top of `format-hook.mjs`, after the existing Node.js imports, add:
-
-```js
-import { loadConfig } from './lib/config.mjs'
-```
-
-Also add the type-import tag alongside existing `@import` tags:
-
-```js
-/** @import { ProjectConfig, FormatterName } from './lib/types.mjs' */
-```
-
-can be trimmed — `ProjectConfig` and `FormatterName` are no longer used directly in the file after
-this spec. Keep only `HookEvent`:
-
-```js
-/** @import { HookEvent } from './lib/types.mjs' */
-```
-
-### Step 2 — Delete `DEFAULTS` constant
-
-Remove lines 24–57 (the `/** @type {ProjectConfig} */` JSDoc and the `const DEFAULTS = { ... }`
-block including its closing `}`).
-
-### Step 3 — Delete `loadProjectConfig` function
-
-Remove lines 60–93 (the JSDoc comment, the `function loadProjectConfig() { ... }` block including
-its closing `}`).
-
-### Step 4 — Update the call site
-
-Change:
-
-```js
-const CONFIG = loadProjectConfig()
-```
-
-to:
-
-```js
-const CONFIG = loadConfig(PROJECT_DIR)
-```
-
-### Step 5 — Verify and commit
-
-```sh
-# No old code remains
-grep -n "const DEFAULTS\|function loadProjectConfig" scripts/format-hook.mjs
-
-# Full suite passes
-pnpm test
-
-# Type-check passes
-pnpm typecheck
-
-git add scripts/format-hook.mjs
-git commit -m "refactor(spec-30): use lib/config.mjs in format-hook.mjs"
-```
-
-## Verification
-
-```sh
-grep -n "const DEFAULTS\|function loadProjectConfig" scripts/format-hook.mjs
-grep -n "loadConfig" scripts/format-hook.mjs
-pnpm test
-pnpm typecheck
-```
-
-## Acceptance criteria
-
-- [ ] `grep -n "const DEFAULTS\|function loadProjectConfig" scripts/format-hook.mjs` → 0 matches
-- [ ] `grep -n "loadConfig" scripts/format-hook.mjs` → match on import line and call site
-- [ ] `pnpm test` passes without any changes to `tests/format-hook.test.mjs`
-- [ ] `pnpm typecheck` passes
-
-## Out of scope
-
-- Changes to any file other than `scripts/format-hook.mjs`
-- Changing skip logic, extension sets, or any other hook behaviour
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/31-version-bump.md b/apps/claude-code/auto-format/docs/plans/31-version-bump.md
deleted file mode 100644
index d09beb7c..00000000
--- a/apps/claude-code/auto-format/docs/plans/31-version-bump.md
+++ /dev/null
@@ -1,66 +0,0 @@
-# 31. Version bump and CHANGELOG entry
-**Status: done — 2026-05-04**
-
-**Priority:** P2
-**Effort:** XS
-**Version impact:** patch
-**Depends on:** spec-30
-**Touches:** `.claude-plugin/plugin.json`, `.claude-plugin/marketplace.json`, `package.json`, `CHANGELOG.md`
-
-## Context
-
-Specs 29–30 extract config loading into `lib/config.mjs`. Internal refactor; no user-visible
-behaviour change. Patch bump.
-
-## Current behaviour
-
-Version reflects the state before specs 29–30.
-
-## Target behaviour
-
-Version incremented by one patch level. CHANGELOG has a dated entry for the config extraction.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `.claude-plugin/plugin.json` | Version bumped |
-| `.claude-plugin/marketplace.json` | Synced by `pnpm bump` |
-| `package.json` | Synced by `pnpm bump` |
-| `CHANGELOG.md` | New dated entry |
-
-## Implementation steps
-
-### Step 1 — Add CHANGELOG entry under `[Unreleased]`
-
-```markdown
-### Changed
-- Internal: extracted `DEFAULTS` and `loadConfig` from `format-hook.mjs` into `lib/config.mjs`,
-  with ten unit tests covering the merge strategy — no behaviour change for consumers.
-```
-
-### Step 2 — Run bump
-
-```sh
-pnpm bump patch
-```
-
-### Step 3 — Verify and commit
-
-```sh
-pnpm verify:changelog
-git add .claude-plugin/plugin.json .claude-plugin/marketplace.json package.json CHANGELOG.md
-git commit -m "chore(release): bump auto-format patch version — config extraction refactor"
-```
-
-## Acceptance criteria
-
-- [ ] `pnpm verify:changelog` passes
-- [ ] CHANGELOG entry present for the new version
-- [ ] `plugin.json`, `marketplace.json`, `package.json` versions are in sync
-
-## Out of scope
-
-- Git tag (create with `pnpm tag` when ready to release)
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/auto-format/docs/plans/README.md b/apps/claude-code/auto-format/docs/plans/README.md
deleted file mode 100644
index bb8f8767..00000000
--- a/apps/claude-code/auto-format/docs/plans/README.md
+++ /dev/null
@@ -1,105 +0,0 @@
-# Plugin Roadmap
-
-This directory contains one Ralph-ready implementation spec per feature. Each file is a self-contained brief: context, affected files, step-by-step implementation, test cases, and acceptance criteria.
-
-## How to use (for Ralph)
-
-1. Pick the next unstarted spec in numerical order.
-2. Read the spec top to bottom **before** touching any code.
-3. If the "Current behaviour" section doesn't match what you find in the file, STOP and add a `## Deviations` section to the spec file documenting the discrepancy. Do NOT silently adapt.
-4. After landing the change, add `**Status: done — <YYYY-MM-DD>**` at the top of the spec.
-5. Read the spec's `**Version impact:** patch|minor|major` line and execute `PROMPT.md` Step 4.5 before committing.
-
-## Ground rules
-
-- **Package manager**: `pnpm` (workspace mode, after spec `00` lands). Before `00`: use `npm` only if unavoidable.
-- **pnpm version pinning**: `package.json#packageManager` (e.g. `pnpm@10.33.0`) is the single source of truth.
-- **Indentation**: tabs. Line endings: LF.
-- **Commits**: conventional commits — `feat(scope): description`, `fix(scope): description`, `chore(scope): description`.
-- **Dep versioning**: `devDependencies` go through the `catalog:` section in `pnpm-workspace.yaml`, pinned exactly. No `^` or `~` ranges.
-- **No runtime deps**: the hook script uses only Node built-ins (`node:child_process`, `node:fs`, `node:path`). Keep it that way — no external npm deps in `dependencies`.
-- **Node version**: Node ≥24.15.0 (Active LTS). `useNodeVersion` in `pnpm-workspace.yaml` pins the patch.
-- **Plugin versioning**: `.claude-plugin/plugin.json` is the single source of truth for the version number. **Never hand-edit** `.claude-plugin/marketplace.json` — use `pnpm bump` (spec `07`). After spec `13`, `sync-version.mjs` propagates to `marketplace.json` and `package.json` automatically.
-- **Release workflow** (after spec `14`): `pnpm bump <type>` → commit → `pnpm tag` → `git push --follow-tags`.
-- **Scope guard**: this plugin formats and lints. Do NOT add MCP servers, skills, agents, or any other Claude Code plugin features unless a spec explicitly calls for it.
-
-## Versioning policy
-
-**SemVer, strictly.**
-
-| Change type | Bump |
-|---|---|
-| Breaking change to hook contract, plugin manifest, or on-disk config schema | MAJOR |
-| New feature: new config option, new extension support, new hook matcher | MINOR |
-| Bug fix, refactor, docs, internal tooling | PATCH |
-
-## Execution order
-
-| # | Spec | Priority | Effort | Status |
-|---|---|---|---|---|
-| 00 | [Bootstrap plugin tooling](./00-bootstrap-plugin-tooling.md) | P0 | S | done |
-| 01 | [Plugin manifest](./01-plugin-manifest.md) | P0 | S | done |
-| 02 | [Hook registration](./02-hook-registration.md) | P0 | S | done |
-| 03 | [Format hook script](./03-format-hook-script.md) | P0 | M | done |
-| 04 | [Per-project config support](./04-per-project-config.md) | P1 | S | done |
-| 05 | [Plugin README](./05-readme.md) | P1 | S | done |
-| 06 | [CHANGELOG and versioning docs](./06-changelog-and-versioning.md) | P1 | S | done |
-| 07 | [pnpm bump tooling](./07-pnpm-bump-tooling.md) | P1 | M | done |
-| 08 | [pnpm verify:changelog](./08-pnpm-verify-changelog.md) | P1 | S | done |
-| 09 | [Smoke tests](./09-smoke-tests.md) | P1 | M | done |
-| 10 | [CI workflow](./10-ci-workflow.md) | P1 | S | done |
-| 11 | [CLAUDE.md](./11-claude-md.md) | P2 | S | done |
-| 12 | [Diff-based verify:changelog](./12-diff-based-verify-changelog.md) | P1 | S | done |
-| 13 | [Version source of truth + sync-version](./13-version-source-of-truth-and-sync.md) | P1 | M | done |
-| 14 | [pnpm tag](./14-pnpm-tag.md) | P2 | S | done |
-| 15 | [Hoist extension Sets out of main()](./15-hoist-extension-sets.md) | P1 | S | done |
-| 16 | [Windows path-separator normalisation](./16-windows-path-normalisation.md) | P0 | S | done |
-| 17 | [spawnSync timeout guard](./17-spawnsync-timeout-guard.md) | P1 | S | done |
-| 18 | [notebook_path + skip-prefix test matrix](./18-notebook-and-skip-prefix-tests.md) | P1 | S | done |
-| 19 | [Optional Biome support](./19-biome-support.md) | P1 | M | done |
-| 20 | [additionalSkipPrefixes config key](./20-additional-skip-prefixes.md) | P2 | S | done |
-| 21 | [JSON 2-space indentation](./21-json-indentation.md) | P1 | S | done |
-| 22 | [JSDoc types and --checkJs type-checking](./22-jsdoc-types-ts-check.md) | P2 | M | done |
-| 23 | [Backfill historical git tags](./23-backfill-tags.md) | P2 | S | done |
-| 24 | [CI auto-tag on version bump](./24-ci-auto-tag.md) | P2 | S | done |
-| 25 | [Add FormatterDescriptor typedef](./25-formatter-descriptor-type.md) | P2 | XS | todo |
-| 26 | [Extract lib/runners.mjs with runFormatter](./26-runner-module.md) | P2 | S | todo |
-| 27 | [Replace runner functions with descriptors](./27-replace-runner-functions.md) | P2 | S | todo |
-| 28 | [Version bump — runner extraction refactor](./28-version-bump.md) | P2 | XS | todo |
-| 29 | [Extract lib/config.mjs with DEFAULTS and loadConfig](./29-config-module.md) | P2 | S | todo |
-| 30 | [Update format-hook.mjs to use lib/config.mjs](./30-wire-up-config-module.md) | P2 | XS | todo |
-| 31 | [Version bump — config extraction refactor](./31-version-bump.md) | P2 | XS | todo |
-
-## Cross-cutting dependencies
-
-- **`01` → `00`**: Manifest needs `package.json` with `name` and `version` already set.
-- **`02` → `01`**: Hook registration references the plugin manifest structure.
-- **`03` → `02`**: The hook script is the implementation of what `02` registers.
-- **`04` → `03`**: Per-project config is a feature of the hook script; must be layered on top of spec `03`.
-- **`05` → `03` + `04`**: README documents the hook behaviour and config format — land those first.
-- **`06` → `01`**: CHANGELOG documents plugin version history; needs the versioning source of truth from `01`.
-- **`07` → `06`**: `pnpm bump` automates what spec `06` sets up manually.
-- **`08` → `07`**: `pnpm verify:changelog` references `pnpm bump` in its error messages.
-- **`09` → `03` + `04`**: Smoke tests cover core hook logic from specs `03` and `04`.
-- **`10` → `09`**: CI runs tests; land tests first.
-- **`11`**: Can land at any time; no code dependencies.
-- **`12` → `08` + `10`**: Diff-based gate extends the verify:changelog script from spec `08`; CI from spec `10` must exist to test it end-to-end.
-- **`13` → `07`**: sync-version.mjs is called by the refactored bump.mjs.
-- **`14` → `13`**: tag.mjs calls sync-version.mjs as a safety step.
-- **`15` → `04`**: Hoisting Sets requires `CONFIG` (from spec `04`) to be module-level.
-- **`16` → `03`**: Adds `toPosix` helper to `format-hook.mjs`; does not conflict with `15`.
-- **`17` → `03`**: Adds timeout to `runPrettier` / `runEslint` in `format-hook.mjs`.
-- **`18` → `09`**: Extends `tests/format-hook.test.mjs`; land the original test file first.
-- **`19` → `03` + `04` + `09` + `17`**: Biome support adds a new runner to `format-hook.mjs` using the same timeout pattern; tests build on spec `09`'s helpers.
-- **`20` → `04` + `09`**: Additive-merge logic is in `loadProjectConfig` (spec `04`); tests build on spec `09`.
-- **`21` → `07` + `13`**: Changes indentation in both `bump.mjs` (spec `07`) and `sync-version.mjs` (spec `13`), and reformats their output files.
-- **`22` → `19`**: Biome support (spec `19`) should be committed before JSDoc annotations to keep formatting diffs and type-annotation diffs in separate commits.
-- **`23` → `14`**: backfill-tags.mjs relies on the same `vX.Y.Z` tag convention as tag.mjs.
-- **`24` → `10` + `14`**: release.yml extends the CI setup from spec `10` and uses the same tag convention as tag.mjs from spec `14`.
-- **`25` → `22`**: `FormatterDescriptor` typedef lives in `lib/types.mjs`, which spec `22` creates.
-- **`26` → `25`**: `runners.mjs` imports `FormatterDescriptor` via JSDoc `@import` from `types.mjs`.
-- **`27` → `26`**: replaces the three runner functions in `format-hook.mjs` with descriptors + `runFormatter`.
-- **`28` → `27`**: version bump after all three refactor specs land.
-- **`29` → `22`**: `lib/config.mjs` uses `ProjectConfig` and `FormatterName` from `lib/types.mjs` (spec `22`).
-- **`30` → `29`**: wires `format-hook.mjs` to the new config module.
-- **`31` → `30`**: version bump after config extraction specs land.
diff --git a/apps/claude-code/auto-format/package.json b/apps/claude-code/auto-format/package.json
index c7254121..17931d45 100644
--- a/apps/claude-code/auto-format/package.json
+++ b/apps/claude-code/auto-format/package.json
@@ -1,28 +1,26 @@
 {
-  "name": "auto-format",
-  "version": "0.5.15",
-  "private": true,
-  "license": "LGPL-3.0-or-later",
-  "type": "module",
-  "packageManager": "pnpm@10.33.0",
+  "devDependencies": {
+    "@types/node": "catalog:",
+    "@unic/release-tools": "workspace:*",
+    "@unic/tsconfig": "workspace:*",
+    "typescript": "catalog:"
+  },
   "engines": {
     "node": ">=22",
     "pnpm": ">=10"
   },
+  "license": "LGPL-3.0-or-later",
+  "name": "auto-format",
+  "packageManager": "pnpm@10.33.0",
+  "private": true,
   "scripts": {
-    "test": "node --test tests/format-hook.test.mjs scripts/lib/runners.test.mjs scripts/lib/config.test.mjs",
-    "typecheck": "tsc --noEmit --project tsconfig.json",
     "bump": "unic-bump",
     "sync-version": "unic-sync-version",
     "tag": "unic-tag",
-    "verify:changelog": "unic-verify-changelog",
-    "ralph": "ralph run -c ralph.yml -H builtin:code-assist"
+    "test": "node --test tests/format-hook.test.mjs scripts/lib/runners.test.mjs scripts/lib/config.test.mjs",
+    "typecheck": "tsc --noEmit --project tsconfig.json",
+    "verify:changelog": "unic-verify-changelog"
   },
-  "devDependencies": {
-    "@ralph-orchestrator/ralph-cli": "catalog:",
-    "@types/node": "catalog:",
-    "@unic/release-tools": "workspace:*",
-    "@unic/tsconfig": "workspace:*",
-    "typescript": "catalog:"
-  }
+  "type": "module",
+  "version": "0.5.15"
 }
diff --git a/apps/claude-code/auto-format/ralph.yml b/apps/claude-code/auto-format/ralph.yml
deleted file mode 100644
index a221284f..00000000
--- a/apps/claude-code/auto-format/ralph.yml
+++ /dev/null
@@ -1,35 +0,0 @@
-# Ralph Orchestrator Configuration
-# Generated by: ralph init --backend claude
-# Project: unic-auto-format
-# Docs: https://github.com/mikeyobrien/ralph-orchestrator
-
-cli:
-  backend: 'claude'
-
-event_loop:
-  prompt_file: 'PROMPT.md'
-  completion_promise: 'LOOP_COMPLETE'
-  max_iterations: 100 # 15 specs * generous retry budget
-  max_runtime_seconds: 28800 # 8 hours; abort if a spec stalls
-
-# ─────────────────────────────────────────────────────────────────────────────
-# Additional Configuration (uncomment to customize)
-# ─────────────────────────────────────────────────────────────────────────────
-
-# core:
-#   scratchpad: ".ralph/agent/scratchpad.md"
-#   specs_dir: ".ralph/specs/"
-
-# Custom hats for multi-agent workflows:
-# hats:
-#   builder:
-#     name: "Builder"
-#     triggers: ["build.task"]
-#     publishes: ["build.done", "build.blocked"]
-#
-#   reviewer:
-#     name: "Reviewer"
-#     triggers: ["review.request"]
-#     publishes: ["review.approved", "review.changes_requested"]
-
-# Create PROMPT.md with your task, then run: ralph run
diff --git a/apps/claude-code/pr-review/AGENTS.md b/apps/claude-code/pr-review/AGENTS.md
new file mode 100644
index 00000000..a5dbeb24
--- /dev/null
+++ b/apps/claude-code/pr-review/AGENTS.md
@@ -0,0 +1,82 @@
+# AGENTS.md — pr-review
+
+Guidance for any AI agent working inside this Plugin directory. `CLAUDE.md` in this directory is a symlink to this file.
+
+## What this Plugin is
+
+`pr-review` is a Claude Code Plugin in the [`unic-agents-plugins`](../../../AGENTS.md) monorepo. It adds the `/pr-review:review-pr` slash command, which Reviews and Re-reviews Azure DevOps pull requests across multiple Review Aspects in parallel and posts Inline Comments plus a Review Summary back to ADO. See [`CONTEXT.md`](CONTEXT.md) for the domain vocabulary (Platform, Revision, Review, Re-review, Review Aspect, Inline Comment, Review Summary, Review Thread).
+
+## Where to start
+
+Root docs (monorepo-wide conventions live here — pnpm scripts, Gitflow, SemVer, Conventional Commits, code conventions, LICENSE policy, cross-platform requirement):
+
+- [Root AGENTS.md](../../../AGENTS.md) — source of truth for cross-cutting rules
+- [Root CONTEXT.md](../../../CONTEXT.md) — monorepo-wide vocabulary (Plugin, Workspace Package, Release, Feature, Consumer)
+- [Root CONTEXT-MAP.md](../../../CONTEXT-MAP.md) — index of all bounded contexts in the repo
+- [Root docs/adr/](../../../docs/adr/) — monorepo-wide architecture decisions
+- [Root docs/process/](../../../docs/process/) — process and workflow guides
+
+This Plugin's own decisions:
+
+- [Plugin docs/adr/](docs/adr/) — Plugin-specific architecture decisions
+
+## Commands
+
+Plugin-specific pnpm scripts (run from this directory or with `pnpm --filter pr-review <script>` from the repo root):
+
+```sh
+pnpm test                 # run node:test suite (scripts/ + agents/ coverage)
+pnpm bump <patch|minor|major>   # bump plugin.json version + promote CHANGELOG
+pnpm sync-version         # mirror plugin.json version into marketplace.json + package.json
+pnpm tag                  # create the pr-review@<version> git tag locally
+pnpm verify:changelog     # check CHANGELOG entry for the current version
+```
+
+Monorepo-wide commands (`pnpm install`, `pnpm check`, `pnpm format`, `pnpm ci:check`) are documented in the [root AGENTS.md](../../../AGENTS.md).
+
+## Layout
+
+```tree
+.claude-plugin/           # Plugin manifest (plugin.json) and marketplace listing
+commands/                 # Slash command definitions — review-pr.md is the thin orchestrator
+agents/                   # Specialised review and ADO-interaction agents
+scripts/                  # Pure helpers shared by the command and the agents (ado-fetcher, ado-writer, pre-pr, re-review/*)
+tests/                    # node:test suites covering scripts/ and agents/
+docs/                     # Plugin-specific documentation
+  adr/                    # Plugin Architecture Decision Records
+```
+
+## Plugin doctrines
+
+Load-bearing invariants. These either originate in a Plugin ADR or are policy decisions that are not obvious from the code.
+
+- **Canonical bot signature.** Every comment posted to a Platform ends with the exact signature `---\n🤖 *Reviewed by Claude Code* — Iteration N` (where `N` is the Revision id). This wording is load-bearing for signature-based detection. See [ADR-0001](docs/adr/0001-canonical-bot-signature.md).
+- **Signature-based prior-Review detection.** The Plugin discovers prior Reviews exclusively via the canonical signature — no metadata fields, no separate state file. See [ADR-0002](docs/adr/0002-signature-based-prior-review-detection.md).
+- **Target the latest Revision.** Every Review targets the latest Revision (`LATEST_ITERATION_ID`); `iterationId=1` is never used. See [ADR-0003](docs/adr/0003-target-latest-pr-iteration.md).
+- **Incremental diff baseline.** A Re-review computes its baseline from the prior Review's signature, producing an incremental diff rather than re-reviewing the whole PR. See [ADR-0004](docs/adr/0004-incremental-diff-baseline.md).
+- **Four-state Review Thread classification.** Each existing Review Thread is classified into one of four states (`addressed`, `disputed`, `pending`, `obsolete`) before any write. See [ADR-0005](docs/adr/0005-four-state-thread-classification.md).
+- **Reply, not duplicate; auto-resolve when addressed.** Re-reviews reply to existing Review Threads rather than opening duplicates; threads classified as `addressed` are auto-resolved. See [ADR-0006](docs/adr/0006-reply-not-duplicate-auto-resolve.md).
+- **Review Summary is rewritten, not appended.** The Review Summary General Comment is updated in place across Re-reviews; new findings are not appended. See [ADR-0007](docs/adr/0007-summary-rewritten-not-appended.md).
+- **Soft dependency on `pr-review-toolkit`.** The command checks for the toolkit Plugin at startup and aborts with installation instructions if missing — there is no bundled copy. See [ADR-0008](docs/adr/0008-soft-dependency-pr-review-toolkit.md).
+- **Orchestrator stays thin (≤ 200 lines).** `commands/review-pr.md` delegates all Platform interaction and coordination logic to the agents in `agents/`. See [ADR-0013](docs/adr/0013-orchestrator-split-for-review-pr.md).
+- **ADO read/write split.** All read operations go through the ADO Fetcher agent; all write operations go through the ADO Writer agent. The orchestrator never calls `az devops invoke` directly. See [ADR-0016](docs/adr/0016-fold-thread-fetch-into-ado-fetcher.md).
+- **Dry-run is a fourth peer mode.** Dry-run sits alongside `review`, `re-review`, and `summary-delta` as a peer Review mode, not a flag on another mode. See [ADR-0017](docs/adr/0017-dry-run-as-fourth-peer-mode.md).
+
+## External dependencies
+
+- **Azure CLI** with the `azure-devops` extension (`az extension add --name azure-devops`), authenticated against the target ADO organisation (`az devops login`). The Plugin shells out via `az devops invoke` for all Platform interaction.
+- **`pr-review-toolkit`** Plugin from `anthropics/claude-plugins-official`. Soft dependency — the command checks at startup and aborts with instructions if absent.
+
+All `az` commands the plugin uses are enumerated in `tests/fixtures/ado-cli-inventory.mjs`. The smoke test (`tests/ado-cli-smoke.test.mjs`) asserts every `az ` invocation in `agents/`/`commands/`/`scripts/` has a matching inventory entry — modulo the allowlist in `tests/fixtures/ado-cli-allowlist.mjs`. Register a new ADO call in the inventory before invoking it.
+
+## Do not add
+
+- **GitHub or GitLab Platform support before an ADR locks the abstraction.** The vocabulary in [`CONTEXT.md`](CONTEXT.md) is Platform-agnostic, but the implementation today is ADO-only; broadening it without an ADR will leak ADO assumptions into shared code.
+- **Vote-on-PR actions** (approve / reject after a Review). Out of scope until requested with a concrete use case.
+- **PR description generation from diff.** Out of scope; the Plugin reviews PRs, it does not author them.
+- **Inline `az devops invoke` calls in the orchestrator.** All Platform calls go through the ADO Fetcher / ADO Writer agents (see doctrines above).
+- **Bypassing the canonical signature** in any comment the Plugin posts. Signature-based detection (ADR-0002) depends on the exact wording from ADR-0001.
+
+## Plugin ADRs
+
+Plugin-specific architecture decisions live in [docs/adr/](docs/adr/).
diff --git a/apps/claude-code/pr-review/CHANGELOG.md b/apps/claude-code/pr-review/CHANGELOG.md
index ba6e3a58..7adda2e3 100644
--- a/apps/claude-code/pr-review/CHANGELOG.md
+++ b/apps/claude-code/pr-review/CHANGELOG.md
@@ -8,8 +8,18 @@
 ### Added
 - (none)
 
+### Changed
+- Preflight now verifies `az devops invoke` is callable in addition to the existing `az --version` and `azure-devops` extension checks; failures surface a re-install hint. CI smoke test (`tests/ado-cli-smoke.test.mjs`) asserts every ADO subcommand the plugin actually invokes still exists, sourced from a single inventory at `tests/fixtures/ado-cli-inventory.mjs`. Same failure class as the Step 4 bug — guarded offline now.
+
 ### Fixed
-- (none)
+- Step 4 mode detection was calling a non-existent `az repos pr thread list` subcommand and failing fatally on every ADO PR review. Thread fetching now lives in the ADO Fetcher and uses `az devops invoke --resource pullRequestThreads`; the orchestrator's Step 4 captures PR metadata via `az repos pr show` and the Fetcher's result block now emits `RAW_THREADS_JSON`, `MODE`, `IS_REREVIEW`, `PRIOR_ITERATION_ID`, and `SUMMARY_THREAD_ID`. Per ADR 0016.
+- `tests/plugin-structure.test.mjs` frontmatter regex now tolerates CRLF line endings, fixing a Windows-only failure that surfaced when `pr-review` was added to the CI test matrix.
+- `scripts/ado/cli-completeness.mjs` now emits an `az <flag>` shape for root-flag invocations (e.g. `az --version`) so the allowlist actually covers them; previously the scanner silently dropped these.
+- `tests/ado-cli-smoke.test.mjs` normalises path separators before checking `SKIP_FRAGMENTS`, so `scratchpad/` and `node_modules/` are also skipped on Windows.
+- `tests/ado-cli-smoke.test.mjs` `walk()` now only swallows `ENOENT`; other read errors (permissions, FD exhaustion) propagate so a broken scan can't silently report "no uninventoried commands".
+- `tests/ado-cli-smoke.test.mjs` pre-warms the `azure-devops` CLI extension in a `before()` hook so the per-entry 5s timeout measures the actual subcommand rather than the one-time extension bootstrap. Fixes a flaky `az repos pr show --help` failure on cold Ubuntu CI runners.
+- `agents/ado-fetcher.md` Step 2 now logs a `WARN` when the HTTP status parser comes up empty on a non-zero `az` exit, so a real 404 with an unparseable error body isn't silently routed to the DEGRADED branch.
+- `tests/ado-fetcher.test.mjs` now asserts `MODE`/`IS_REREVIEW`/`PRIOR_ITERATION_ID`/`SUMMARY_THREAD_ID`/`RAW_THREADS_JSON` appear specifically inside the `ADO_FETCHER_RESULT_START`/`_END` block, not just anywhere in the markdown.
 
 ## [1.2.11] — 2026-05-19
 
diff --git a/apps/claude-code/pr-review/CLAUDE.md b/apps/claude-code/pr-review/CLAUDE.md
deleted file mode 100644
index 0554a834..00000000
--- a/apps/claude-code/pr-review/CLAUDE.md
+++ /dev/null
@@ -1,75 +0,0 @@
-# CLAUDE.md
-
-This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
-
-## What this repo is
-
-A Claude Code plugin (`pr-review`) that adds a `/pr-review:review-pr` command. When invoked with an ADO PR URL it:
-
-1. Fetches the PR diff via `az devops` CLI + ADO REST API
-2. Launches specialized review agents from the `pr-review-toolkit` plugin in parallel
-3. Posts each finding as an inline thread at the exact file/line in Azure DevOps
-4. Posts a severity-grouped summary comment
-
-## Repository layout
-
-```
-.claude-plugin/
-  plugin.json          # Plugin manifest (name, version, description)
-  marketplace.json     # Marketplace listing metadata
-commands/
-  review-pr.md        # The slash command definition — thin orchestrator
-agents/
-  ado-fetcher.md          # ADO Fetcher — all Azure DevOps REST API fetches
-  re-review-coordinator.md # Re-review Coordinator — thread classification + incremental diff
-  ado-writer.md           # ADO Writer — posts inline threads and summary comment to ADO
-  doc-context-orchestrator.md  # Doc Context Orchestrator — work item + Confluence fetching
-  doc-context-synthesizer.md   # Doc Context Synthesizer — produces business-context narrative
-```
-
-`commands/review-pr.md` is a thin orchestrator (≤ 200 lines per PRD acceptance criterion). It delegates ADO API calls and coordination logic to the focused agents in `agents/`. Pure helpers used by both the orchestrator and the agents live under `scripts/` (`ado-fetcher.mjs`, `ado-writer.mjs`, `pre-pr.mjs`, `mode-detection.mjs`, `confluence-client.mjs`, `re-review/*.mjs`) with tests under `tests/`. There are no build steps, no transpilation, no dependencies to install.
-
-## Plugin metadata
-
-When bumping the version, update it in **both** files:
-
-- `.claude-plugin/plugin.json` → `version`
-- `.claude-plugin/marketplace.json` → `plugins[0].version`
-
-## Command conventions (`commands/review-pr.md`)
-
-- YAML frontmatter declares `allowed-tools` — add any new tools the command needs there
-- Auto-generated files are skipped during file-content reading by the `shouldSkipFile` helper in `scripts/pre-pr.mjs` (serialization YAMLs, `*.g.cs`, generated types output, `swagger.md`)
-- All comments posted to ADO **must** end with the exact signature: `---\n🤖 *Reviewed by Claude Code* — Iteration N` (where N = LATEST_ITERATION_ID)
-- ADO REST calls (`pullRequestThreads`, thread replies, iteration fetches) are handled by the focused agents in `agents/`, not inline in the orchestrator command
-- ADO Fetcher (`ado-fetcher.md`) owns all read operations; ADO Writer (`ado-writer.md`) owns all write operations; Re-review Coordinator (`re-review-coordinator.md`) owns thread classification and incremental diff logic
-- Inline threads use ADO REST `pullRequestThreads` via `az devops invoke`; file paths must match ADO format (leading `/`, forward slashes)
-- Always use the latest iteration of the PR. `iterationId=1` is never used. Re-reviews additionally compute `PRIOR_ITERATION_ID` from the prior review's signature — see spec 02.
-- If `az devops invoke` returns a `threadContext` error, fall back to posting without `threadContext` (general comment)
-
-## External dependencies
-
-- **`pr-review-toolkit` plugin** — soft dependency; the command checks for it at startup and aborts with instructions if missing
-- **Azure CLI** with `azure-devops` extension (`az extension add --name azure-devops`)
-- **`az devops login`** authenticated to the target org
-
-## Install for local development
-
-Add to `~/.claude/settings.json`:
-
-```json
-"extraKnownMarketplaces": {
-  "unic": {
-    "source": { "source": "directory", "path": "/path/to/unic-agents-plugins" },
-    "autoUpdate": true
-  }
-}
-```
-
-Then add `"pr-review@unic": true` to `enabledPlugins` and restart Claude Code.
-
-## Roadmap (not yet implemented)
-
-- GitHub PR support
-- Vote on PR (approve/reject) after review
-- PR description generation from diff
diff --git a/apps/claude-code/pr-review/CLAUDE.md b/apps/claude-code/pr-review/CLAUDE.md
new file mode 120000
index 00000000..47dc3e3d
--- /dev/null
+++ b/apps/claude-code/pr-review/CLAUDE.md
@@ -0,0 +1 @@
+AGENTS.md
\ No newline at end of file
diff --git a/apps/claude-code/pr-review/CONTEXT.md b/apps/claude-code/pr-review/CONTEXT.md
index 1b896d68..7f534924 100644
--- a/apps/claude-code/pr-review/CONTEXT.md
+++ b/apps/claude-code/pr-review/CONTEXT.md
@@ -91,10 +91,14 @@ _Avoid_: initial review, fresh review
 A Review run against an ADO PR where a prior **Bot Signature** is found in the PR's threads. Focuses on commits since the last Review, performs Thread Classification, and replies to or resolves existing Review Threads rather than duplicating them.
 _Avoid_: incremental review, follow-up review, second pass
 
+**Dry-run mode**:
+A Review run against an ADO PR with the `--dry-run` flag. Identical to first-review / re-review for every read-side step — **ADO Fetcher**, **Doc Context Orchestrator**, all Review Aspect agents, and (when prior signature is found) the **Re-review Coordinator's** Thread Classification — but the **ADO Writer** is never invoked. Findings are rendered in the Claude interface using the same severity-grouped format as Pre-PR mode. Used to preview what a Review or Re-review would post before committing to write-back.
+_Avoid_: preview review, simulated review, no-post review
+
 ### Orchestration agents
 
 **ADO Fetcher**:
-A plugin agent that retrieves PR metadata, iterations, changed files, and the raw diff from Azure DevOps. Used by first-review and re-review modes; not invoked in pre-PR mode.
+A plugin agent that retrieves PR metadata, iterations, PR threads, changed files, and the raw diff from Azure DevOps — and determines the Review mode from the thread data. All ADO read operations for a PR review are owned here; the orchestrator makes no inline ADO read calls.
 _Avoid_: fetcher, data agent, ADO client
 
 **Re-review Coordinator**:
@@ -158,9 +162,9 @@ A single end-of-run line printed by the orchestrator to the Claude interface, re
 - A **Doc Context** is assembled via a three-tier pipeline: the **Doc Context Orchestrator** spawns **Work Item Summarizer** and **Confluence Fetcher** agents (Doc Context Sub-agents) in parallel, then delegates their outputs to the **Doc Context Synthesizer**, which produces the final `DOC_CONTEXT` narrative injected into every Review Aspect agent
 - A **Doc Context Sub-agent** operates on a single source (work item or Confluence page) and receives the changed files list and the local diff when available
 - The **Doc Context Orchestrator** returns the **Doc Context Synthesizer**'s output verbatim; it does not rewrite or reformat the narrative
-- The **ADO Fetcher** is invoked by first-review and re-review modes; **Pre-PR mode** skips it entirely and goes directly to Review Aspect agents
-- The **Re-review Coordinator** is invoked only when the mode is re-review; first-review and pre-PR modes never load it
-- The **ADO Writer** is invoked by first-review and re-review modes; **Pre-PR mode** does not write back to ADO
+- The **ADO Fetcher** is invoked by first-review, re-review, and dry-run modes; **Pre-PR mode** skips it entirely and goes directly to Review Aspect agents
+- The **Re-review Coordinator** is invoked when the mode is re-review or a dry-run that detected a prior **Bot Signature**; first-review, pre-PR, and dry-run-on-fresh-PR modes never load it. In dry-run mode the Coordinator performs Thread Classification but its results feed only the rendered output — no Replies are posted
+- The **ADO Writer** is invoked by first-review and re-review modes; **Pre-PR mode** and **Dry-run mode** do not write back to ADO
 - Every operation in an orchestration agent terminates in one of the four **Notice Tiers**. **DEGRADED** and **EMPTY-BY-DESIGN**-with-message operations emit a **Notice** that flows from the agent's structured result block, through the orchestrator's merge step, into the **Review Summary** (for ADO modes) or the printed pre-findings block (for **Pre-PR mode**). The end-of-run **Trailer** carries Notice counts so the invoker sees them without opening the PR.
 
 ## Example dialogue
diff --git a/apps/claude-code/pr-review/PROMPT.md b/apps/claude-code/pr-review/PROMPT.md
deleted file mode 100644
index 18ad81d4..00000000
--- a/apps/claude-code/pr-review/PROMPT.md
+++ /dev/null
@@ -1,113 +0,0 @@
-# Ralph driver — `unic-pr-review` re-review feature
-
-You are implementing the **re-review** capability for the `unic-pr-review` plugin, one spec at a time.
-
-## Step 1 — Determine what's next
-
-Open `docs/plans/README.md`. Scan spec files in order and find the first file whose status is **not** `done` nor `ignore`.
-
-If **all** specs contain `**Status: done**`, output exactly:
-
-```
-<promise>LOOP_COMPLETE</promise>
-```
-
-…and stop. Do nothing else.
-
-## Step 2 — Read the spec completely
-
-Read the entire spec file before writing any code. Pay special attention to:
-
-- **Current behaviour** — verify the code actually works this way before starting; if it doesn't, add a `## Deviations` section at the bottom of the spec file documenting the discrepancy
-- **Out of scope** — list of things you must NOT change in this iteration
-- **Depends on** — if a dependency spec is not yet marked done, skip this spec and move to the next one that can be done
-
-## Step 3 — Implement
-
-Check the spec's `**Version impact:**` line to choose the implementation approach:
-
-**`none` (workspace/infrastructure spec)** — implement directly. Make the smallest set of edits that fully satisfies the spec's _Acceptance criteria_.
-
-**`patch` / `minor` / `major` (plugin spec)** — use `/tdd` to drive implementation. Treat the spec's "Implementation steps" as guidance (key files to touch, rough order), not a recipe. The spec's **Acceptance criteria** are the target; the red-green-refactor cycle drives the path there.
-
-Ground rules:
-
-- Use `pnpm` for all package operations
-- Tabs for indentation, LF line endings (per `.editorconfig`)
-- Conventional commits: `feat(scope): description`, `fix(scope): description`, `chore(scope): description`
-- Cross-platform: use Node.js APIs (`node:path`, `node:fs`, `node:os`) instead of shell commands; no bash/sh assumptions
-- Don't refactor outside the spec's _Touches_ list
-- Don't invent new specs; if you find missing work, add a `## Follow-ups` line to the current spec and surface it in `docs/plans/README.md` under "Discovered work"
-- If something can't be followed as written: document it in `## Deviations`, don't silently deviate
-
-**Supporting skill:** `/diagnose` — if a failure persists and the root cause appears to be outside the spec's scope (an existing module, tooling, or dependency the spec doesn't touch), use `/diagnose` to isolate it before continuing.
-
-## Step 4 — Verify
-
-Run the exact commands in the spec's **Verification** section. Fix any failures before proceeding.
-
-Check every item in **Acceptance criteria**. If any item fails, fix it.
-
-Then run the always-on repo hygiene checks regardless of spec type:
-
-```sh
-pnpm -w check   # Biome + Prettier (workspace root) — fix any failures before proceeding
-```
-
-## Step 4.5 — Version bump (conditional)
-
-Check the spec's `**Version impact:**` line:
-
-**`none`** — skip this step entirely. No CHANGELOG update. Proceed to Step 5.
-
-**`patch` / `minor` / `major`** — bump the plugin:
-
-1. Append one bullet under the matching subsection of `## [Unreleased]` in `CHANGELOG.md`:
-
-   - `### Breaking` — CLI flag change, exit-code change, on-disk schema change
-   - `### Added` — new flag, subcommand, or user-visible feature
-   - `### Fixed` — bug fix, refactor, docs, internal tooling
-
-2. Run: `pnpm bump <patch|minor|major>`
-
-   This atomically increments `plugin.json` version, mirrors into `marketplace.json`, and promotes `[Unreleased]` → a new dated section.
-
-3. Run `pnpm verify:changelog` to confirm the check passes.
-
-## Step 5 — Mark done and commit
-
-1. Add this line immediately after the spec's `# NN. Title` heading (as the second line):
-
-```
-**Status: done — YYYY-MM-DD**
-```
-
-Replace `YYYY-MM-DD` with today's date.
-
-2. Stage and commit all changes:
-
-- **`none` specs**: `git commit -m "chore(spec-NN): <short description>"`
-- **`patch`/`minor`/`major` specs**: `git commit -m "feat(spec-NN): <short description> (vX.Y.Z)"`
-
-Replace `NN` with the spec number (e.g. `00`, `03`).
-Replace `X.Y.Z` with the version output by `pnpm bump`.
-
-3. **Do not push.** Commits only.
-
-## Step 6 — Stop for this iteration
-
-Output a brief summary of what was implemented and committed. Then stop — Ralph will feed this prompt again for the next spec.
-
----
-
-## Important constraints
-
-- Implement **one spec per iteration**. Do not implement multiple specs in a single run.
-- If a spec requires human judgment (e.g. choosing between two design approaches not covered by the spec), stop, document the question in a `## Questions` section at the bottom of the spec file, and output:
-  ```
-  <promise>LOOP_COMPLETE</promise>
-  ```
-  This pauses the loop for human review.
-- Do not create any files outside of what the spec describes.
-- Do not modify any spec files except to add `**Status: done**`, `## Deviations`, `## Questions`, or `## Follow-ups` sections.
-- Never edit `CLAUDE.md`'s rule list silently — propose the change in the spec instead.
diff --git a/apps/claude-code/pr-review/agents/ado-fetcher.md b/apps/claude-code/pr-review/agents/ado-fetcher.md
index 291344cc..20263642 100644
--- a/apps/claude-code/pr-review/agents/ado-fetcher.md
+++ b/apps/claude-code/pr-review/agents/ado-fetcher.md
@@ -1,7 +1,7 @@
 ---
 name: ado-fetcher
 allowed-tools: ['Bash']
-description: 'Fetch all Azure DevOps read data required for a PR review: PR metadata, latest iteration, changed files, raw diff, and linked work-item IDs. Read-only — no write operations.'
+description: 'Fetch all Azure DevOps read data required for a PR review: latest iteration, PR threads, mode detection, changed files, raw diff, and linked work-item IDs. Read-only — no write operations.'
 ---
 
 # ADO Fetcher
@@ -14,37 +14,23 @@ You receive all required context in this prompt as literal strings. Do not read
 
 ## Inputs
 
-You receive:
+You receive (all as literal strings — agents do not inherit environment variables):
 
 - `ORG_URL` — the Azure DevOps organisation URL (e.g. `https://dev.azure.com/myorg`)
 - `PROJECT` — the ADO project name
 - `PR_ID` — the pull request ID (integer as string)
-- `PRIOR_ITERATION_ID` — the iteration ID from the prior review (integer as string, or empty string for first-review)
+- `REPO_ID` — the ADO repository ID (GUID), captured by the orchestrator from PR metadata
+- `SOURCE_BRANCH` — the PR source branch name (no `refs/heads/` prefix)
+- `TARGET_BRANCH` — the PR target branch name (no `refs/heads/` prefix)
+- `PR_TITLE` — the PR title (for downstream agents; not used internally)
+- `PR_DESCRIPTION` — the PR description text (for downstream agents; not used internally)
 - `PLUGIN_ROOT` — absolute path to this plugin's directory (for Node.js helper scripts)
 
----
-
-## Step 1 — Fetch PR metadata
-
-```bash
-az repos pr show --id {PR_ID} --org {ORG_URL} --output json
-```
-
-Capture and remember:
-
-- `repository.id` → `REPO_ID`
-- `repository.project.name` → `PROJECT` (update if it differs from the input)
-- `sourceRefName` → `SOURCE_REF` (e.g. `refs/heads/feature/my-branch`)
-- `targetRefName` → `TARGET_REF` (e.g. `refs/heads/develop`)
-- `title` → `PR_TITLE`
-- `description` → `PR_DESCRIPTION`
-- `status` — note if already merged (`mergeStatus: succeeded`); continue without error — comments are still useful as a review record
-
-Strip `refs/heads/` prefix from `SOURCE_REF` and `TARGET_REF` to get plain branch names (`SOURCE_BRANCH`, `TARGET_BRANCH`).
+PR metadata (`REPO_ID`, branches, title, description) is gathered by the orchestrator's PR-metadata call and passed in — this agent never re-fetches it. If the upstream PR was already merged at fetch time (`mergeStatus: succeeded`), the orchestrator continues without error and that detail does not need to flow into this agent — comments are still useful as a review record.
 
 ---
 
-## Step 2 — Fetch PR iterations and resolve latest
+## Step 1 — Fetch PR iterations and resolve latest
 
 ```bash
 ITERATIONS_JSON=$(az devops invoke \
@@ -93,6 +79,94 @@ LATEST_COMMIT_SHA=$(echo "$ITER_RESULT"   | node -e "process.stdout.write(JSON.p
 
 ---
 
+## Step 2 — Fetch PR threads and detect mode
+
+Fetch the full thread list once via `az devops invoke` against the `pullRequestThreads` resource — the `azure-devops` CLI extension has no `az repos pr thread` subcommand, so the raw REST resource is the only option. Apply the HTTP-tier classification from ADR 0015: `401/403` → ABORTED with an `az devops login` hint; `404` → OK, treat as empty threads (`{"value":[]}`); `5xx` and network errors → DEGRADED warning Notice with `kind: thread-fetch`, proceed with empty threads.
+
+```bash
+THREADS_RESPONSE=$(az devops invoke \
+  --area git \
+  --resource pullRequestThreads \
+  --route-parameters "project=$PROJECT" "repositoryId=$REPO_ID" "pullRequestId=$PR_ID" \
+  --org "$ORG_URL" \
+  --api-version "7.1" \
+  --output json 2>/tmp/ado_fetcher_threads.err)
+THREADS_EXIT=$?
+THREADS_ERR_BODY=$(cat /tmp/ado_fetcher_threads.err 2>/dev/null || true)
+
+# Parse HTTP status from the CLI's error body (best-effort — falls back to exit-code-only classification).
+THREADS_STATUS=$(printf '%s' "$THREADS_ERR_BODY" | node -e "
+const txt = require('fs').readFileSync('/dev/stdin','utf8')
+const m = txt.match(/HTTP\\s+(?:status\\s+)?(\\d{3})/i) || txt.match(/\\b(4\\d{2}|5\\d{2})\\b/)
+process.stdout.write(m ? m[1] : '')
+")
+
+THREADS_TIER=$(
+  TH_STATUS="$THREADS_STATUS" \
+  TH_EXIT="$THREADS_EXIT" \
+  TH_BODY="$THREADS_ERR_BODY" \
+  PLUGIN_R="$PLUGIN_ROOT" \
+  node --input-type=module << 'EOJS'
+const { classifyHttpError } = await import(`file://${process.env.PLUGIN_R}/scripts/ado/classify-http-error.mjs`)
+const result = classifyHttpError({
+  status: process.env.TH_STATUS ? Number(process.env.TH_STATUS) : 0,
+  body: process.env.TH_BODY ?? '',
+  exitCode: Number(process.env.TH_EXIT),
+})
+process.stdout.write(JSON.stringify(result))
+EOJS
+)
+
+THREADS_FETCH_FAILED=false
+THREADS_FETCH_FAIL_MESSAGE=""
+
+if [ "$THREADS_EXIT" != "0" ]; then
+  # Surface the parser miss so a real 404 with an unparseable error body isn't silently routed
+  # to the DEGRADED branch (which would proceed with empty threads + a thread-fetch Notice).
+  if [ -z "$THREADS_STATUS" ]; then
+    echo "WARN: could not parse HTTP status from az stderr; falling back to exit-code classification." >&2
+  fi
+  TIER=$(echo "$THREADS_TIER" | node -e "process.stdout.write(JSON.parse(require('fs').readFileSync('/dev/stdin','utf8')).tier)")
+  TMSG=$(echo "$THREADS_TIER" | node -e "process.stdout.write(JSON.parse(require('fs').readFileSync('/dev/stdin','utf8')).message)")
+  if [ "$THREADS_STATUS" = "404" ]; then
+    # 404 on the threads endpoint is treated as OK with an empty thread list.
+    RAW_THREADS_JSON='{"value":[]}'
+  elif [ "$TIER" = "aborted" ]; then
+    # 401/403 — surface the az devops login hint and abort.
+    echo "ERROR: $TMSG" >&2
+    rm -f /tmp/ado_fetcher_threads.err
+    exit 1
+  else
+    # 5xx / network — DEGRADED. Proceed with empty threads and emit a thread-fetch Notice in Step 6.
+    RAW_THREADS_JSON='{"value":[]}'
+    THREADS_FETCH_FAILED=true
+    THREADS_FETCH_FAIL_MESSAGE="$TMSG"
+  fi
+else
+  RAW_THREADS_JSON="$THREADS_RESPONSE"
+fi
+rm -f /tmp/ado_fetcher_threads.err
+```
+
+Run `detectMode` against `.value` of `RAW_THREADS_JSON` to classify the run and capture the prior-iteration / summary-thread IDs:
+
+```bash
+eval "$(
+  RAW_T="$RAW_THREADS_JSON" \
+  PLUGIN_R="$PLUGIN_ROOT" \
+  node --input-type=module << 'EOJS'
+const { detectMode, formatModeEnv, SIGNATURE_PREFIX } = await import(`file://${process.env.PLUGIN_R}/scripts/mode-detection.mjs`)
+const parsed = JSON.parse(process.env.RAW_T || '{"value":[]}')
+const threads = Array.isArray(parsed) ? parsed : (parsed.value ?? [])
+process.stdout.write(formatModeEnv(detectMode({ threads, signaturePrefix: SIGNATURE_PREFIX })))
+EOJS
+)"
+```
+
+Sets `MODE`, `IS_REREVIEW`, `PRIOR_ITERATION_ID`, `SUMMARY_THREAD_ID`. `PRIOR_ITERATION_ID` is consumed by Step 4 (diff strategy); the other three flow through the result block to the orchestrator and downstream agents.
+
+---
+
 ## Step 3 — List changed files
 
 ```bash
@@ -240,12 +314,15 @@ rm -f /tmp/ado_fetcher_wi.err
 
 Initialise the per-agent Notices array. Emission sites:
 
+- **DEGRADED warning** (`kind: thread-fetch`) — when the threads endpoint returned 5xx or a network error (`THREADS_FETCH_FAILED=true`); the run proceeds with `MODE: first-review` and an empty thread list, but downstream re-review detection is unavailable for this run.
 - **DEGRADED warning** (`kind: diff-range`) — when `DIFF_RANGE_FALLBACK=true` (prior commit unreachable; fell back to full diff).
 - **DEGRADED warning** (`kind: work-items`) — when the work-item fetch failed (`WI_OK=false`); message comes from the helper.
 - **EMPTY-BY-DESIGN info** (`kind: doc-context`) — when `WORK_ITEM_IDS=[]` and the fetch succeeded (no work items linked to the PR).
 
 ```bash
 NOTICES=$(
+  THREADS_FAIL="${THREADS_FETCH_FAILED:-false}" \
+  THREADS_MSG="$THREADS_FETCH_FAIL_MESSAGE" \
   DIFF_RANGE_FB="${DIFF_RANGE_FALLBACK:-false}" \
   WI_IDS="$WORK_ITEM_IDS" \
   WI_OK="$WI_OK" \
@@ -255,6 +332,9 @@ NOTICES=$(
 const { createNotice } = await import(`file://${process.env.PLUGIN_R}/scripts/ado/notices.mjs`)
 const ids = JSON.parse(process.env.WI_IDS || '[]')
 const notices = []
+if (process.env.THREADS_FAIL === 'true') {
+  notices.push(createNotice('warning', 'thread-fetch', process.env.THREADS_MSG || 'Failed to fetch PR threads. Proceeded with empty thread list; re-review detection is unavailable for this run.'))
+}
 if (process.env.DIFF_RANGE_FB === 'true') {
   notices.push(createNotice('warning', 'diff-range', 'Incremental diff unavailable — Coordinator will classify against the full PR diff with conservative downgrades.'))
 }
@@ -288,9 +368,16 @@ TARGET_BRANCH: {TARGET_BRANCH}
 LATEST_ITERATION_ID: {LATEST_ITERATION_ID}
 LATEST_COMMIT_SHA: {LATEST_COMMIT_SHA}
 DIFF_RANGE: {DIFF_RANGE}
+MODE: {MODE}
+IS_REREVIEW: {IS_REREVIEW}
+PRIOR_ITERATION_ID: {PRIOR_ITERATION_ID}
+SUMMARY_THREAD_ID: {SUMMARY_THREAD_ID}
 WORK_ITEM_IDS: {WORK_ITEM_IDS}
 NOTICES: {NOTICES}
 
+RAW_THREADS_JSON:
+{RAW_THREADS_JSON}
+
 CHANGED_FILES:
 {CHANGED_FILES}
 
@@ -301,11 +388,16 @@ ADO_FETCHER_RESULT_END
 
 Where:
 
+- `MODE` is `first-review` or `re-review` as classified by `detectMode` in Step 2.
+- `IS_REREVIEW` is `true` or `false` (the boolean form of `MODE === 're-review'`).
+- `PRIOR_ITERATION_ID` is the iteration ID of the prior signed Review, or empty string for first-review.
+- `SUMMARY_THREAD_ID` is the thread ID of the prior Review Summary, or empty string for first-review.
+- `RAW_THREADS_JSON` is the unfiltered ADO `pullRequestThreads` response (a JSON object with a `.value` array). When the threads endpoint failed with 5xx / network, this is `{"value":[]}` and `NOTICES` contains a `warning`-severity `thread-fetch` entry.
 - `DIFF_RANGE` is `full` when the diff ran against `origin/${TARGET_BRANCH}...HEAD` (first-review or fallback), or `incremental` when it ran against `${PRIOR_COMMIT_SHA}..${LATEST_COMMIT_SHA}`. When `full` due to a fallback, the `NOTICES` array also contains a `warning`-severity `diff-range` entry.
 - `WORK_ITEM_IDS` is the JSON array from Step 5, e.g. `[42, 7]` or `[]`
 - `NOTICES` is the JSON array from Step 6, e.g. `[{"severity":"info","kind":"doc-context","message":"..."}]` or `[]`
 - `CHANGED_FILES` is the newline-separated list from Step 3, e.g. `edit: /src/api.ts`
 - `RAW_DIFF` is the full diff text from Step 4 (may be empty if no new commits)
-- `LATEST_COMMIT_SHA` is the latest source-branch commit SHA captured in Step 2; reserved for future diff-range debugging and not consumed by any current downstream agent — the diff-range logic that needed it is now self-contained in Step 4 above.
+- `LATEST_COMMIT_SHA` is the latest source-branch commit SHA captured in Step 1; reserved for future diff-range debugging and not consumed by any current downstream agent — the diff-range logic that needed it is now self-contained in Step 4 above.
 
 **Never add any ADO write operations (POST, PATCH, DELETE) to this agent.**
diff --git a/apps/claude-code/pr-review/agents/re-review-coordinator.md b/apps/claude-code/pr-review/agents/re-review-coordinator.md
index 8b3d8d78..ed2e440a 100644
--- a/apps/claude-code/pr-review/agents/re-review-coordinator.md
+++ b/apps/claude-code/pr-review/agents/re-review-coordinator.md
@@ -24,7 +24,7 @@ You receive:
   - `LATEST_ITERATION_ID`
   - `RAW_DIFF` — the raw git diff text (may be empty)
   - `DIFF_RANGE` — `full` or `incremental`; controls the γ-downgrade in Step 5
-- `RAW_THREADS_JSON` — the full unfiltered ADO thread list as a JSON array (fetched by the orchestrator via `az repos pr thread list`; not re-fetched here)
+- `RAW_THREADS_JSON` — the full unfiltered ADO thread list (the raw `pullRequestThreads` response, an object with a `.value` array of threads). Fetched by the ADO Fetcher agent and forwarded to this agent verbatim by the orchestrator; not re-fetched here.
 - `FINDINGS` — a JSON array of new findings: `{ severity, filePath, startLine, endLine, title, body }[]`
 - `SIGNATURE_PREFIX` — always `🤖 *Reviewed by Claude Code*`
 - `PLUGIN_ROOT` — absolute path to this plugin's directory (for Node.js helper scripts)
@@ -84,7 +84,8 @@ DETECT_JSON=$(
   node --input-type=module << 'EOJS'
 import { writeFileSync } from 'node:fs'
 const { detectPriorReview } = await import('file://' + process.env.PLUGIN_R + '/scripts/re-review/detect-prior-review.mjs')
-const threads = JSON.parse(process.env.RAW_THREADS)
+const parsed = JSON.parse(process.env.RAW_THREADS || '{"value":[]}')
+const threads = Array.isArray(parsed) ? parsed : (parsed.value ?? [])
 const r = detectPriorReview({ threads, signaturePrefix: process.env.SIG_P })
 writeFileSync(process.env.THREADS_OUT_F, JSON.stringify(r.priorThreads))
 process.stdout.write(JSON.stringify({
@@ -468,7 +469,7 @@ Where:
 ## Important invariants
 
 - **No ADO reads**: do not call `az devops invoke` for GET operations. All data is passed as inputs.
-- **No re-fetch of threads**: the orchestrator already captured `RAW_THREADS_JSON` during mode detection — do not call `az repos pr thread list` again.
+- **No re-fetch of threads**: the ADO Fetcher already captured `RAW_THREADS_JSON` via `az devops invoke --resource pullRequestThreads` and the orchestrator forwarded it here — do not re-issue that call.
 - **Early exit has no ADO writes**: the no-new-revisions path (Step 4) only prints to console and returns the result block — it never posts replies or PATCHes threads.
 - **All four count fields are always present** in the result block, even when zero.
 - **Matched findings are consumed**: a finding matched to any classified prior thread is excluded from `freshFindings`, regardless of whether a reply was posted.
diff --git a/apps/claude-code/pr-review/commands/review-pr.md b/apps/claude-code/pr-review/commands/review-pr.md
index 962ed647..a40fafb1 100644
--- a/apps/claude-code/pr-review/commands/review-pr.md
+++ b/apps/claude-code/pr-review/commands/review-pr.md
@@ -43,35 +43,36 @@ Extract a PR URL from `$ARGUMENTS`. Expected format: `https://dev.azure.com/{org
 
 ## Step 3 — Azure CLI check (PR modes only)
 
-Run `az --version` and `az extension list | grep azure-devops`. If missing: `az extension add --name azure-devops`.
+Run `az --version` and `az extension list | grep azure-devops`. If missing: `az extension add --name azure-devops`. Then verify `az devops invoke` itself is callable — the extension is sometimes installed but broken after partial upgrades (this is the failure class the smoke test in `tests/ado-cli-smoke.test.mjs` guards offline):
 
-## Step 4 — Re-review detection
+```bash
+if ! az devops invoke --help >/dev/null 2>&1; then
+  echo "ERROR: az devops invoke unavailable. Re-install: az extension remove --name azure-devops && az extension add --name azure-devops" >&2
+  exit 1
+fi
+```
 
-Fetch the thread list **once**; never re-fetch downstream.
+## Step 4 — Fetch PR metadata
+
+Gather the PR's repository, branches, title, and description from the Azure CLI so the Fetcher can be launched with full context. This is the only Azure CLI call the orchestrator makes — thread fetching and mode detection live inside the Fetcher (see ADR 0016). Branch refs are stripped of the `refs/heads/` prefix.
 
 ```bash
-PR_THREADS_ERR="${TMPDIR:-/tmp}/pr_threads.err"
-RAW_THREADS_JSON=$(az repos pr thread list --id "$PR_ID" --org "$ORG_URL" --output json 2>"$PR_THREADS_ERR") || {
-  echo "ERROR: failed to fetch PR threads via Azure CLI. Try \`az devops login\` to re-authenticate." >&2
-  cat "$PR_THREADS_ERR" >&2; exit 1; }
-
-eval "$(
-  RAW_T="$RAW_THREADS_JSON" SIG_P="🤖 *Reviewed by Claude Code*" PLUGIN_R="${CLAUDE_PLUGIN_ROOT}" \
-  node --input-type=module << 'EOJS'
-const { detectMode, formatModeEnv } = await import(`file://${process.env.PLUGIN_R}/scripts/mode-detection.mjs`)
-const threads = JSON.parse(process.env.RAW_T || '[]')
-process.stdout.write(formatModeEnv(detectMode({ threads, signaturePrefix: process.env.SIG_P })))
-EOJS
-)"
-
-echo "Mode detected: $MODE"
+PR_META_JSON=$(az repos pr show --id "$PR_ID" --org "$ORG_URL" --output json) || {
+  echo "ERROR: failed to fetch PR metadata. Try \`az devops login\` to re-authenticate." >&2; exit 1; }
+
+eval "$(printf '%s' "$PR_META_JSON" | node -e "
+const m=JSON.parse(require('fs').readFileSync(0,'utf8'))
+const q=s=>String(s||'').replace(/'/g,\"'\\\\''\")
+const strip=r=>String(r||'').replace(/^refs\/heads\//,'')
+for(const[k,v]of[['REPO_ID',m.repository?.id],['PROJECT',m.repository?.project?.name],['SOURCE_BRANCH',strip(m.sourceRefName)],['TARGET_BRANCH',strip(m.targetRefName)],['PR_TITLE',m.title],['PR_DESCRIPTION',m.description]])process.stdout.write(\`\${k}='\${q(v)}'\\n\`)
+")"
 ```
 
-Sets `MODE`, `IS_REREVIEW`, `PRIOR_ITERATION_ID`, `SUMMARY_THREAD_ID`.
+Sets `REPO_ID`, `PROJECT`, `SOURCE_BRANCH`, `TARGET_BRANCH`, `PR_TITLE`, `PR_DESCRIPTION`. The Fetcher (Step 5) detects mode and emits `MODE`, `IS_REREVIEW`, `PRIOR_ITERATION_ID`, `SUMMARY_THREAD_ID` in its result block.
 
 ## Step 5 — ADO Fetcher
 
-Launch the ADO Fetcher agent and **wait for its result** before anything else (the PRD requires the Fetcher to complete before downstream agents run).
+Launch the ADO Fetcher agent and **wait for its result** before anything else (the PRD requires the Fetcher to complete before downstream agents run). The Fetcher fetches threads and runs mode detection internally.
 
 ```txt
 Agent(
@@ -80,12 +81,16 @@ Agent(
   ORG_URL: {ORG_URL}
   PROJECT: {PROJECT}
   PR_ID: {PR_ID}
-  PRIOR_ITERATION_ID: {PRIOR_ITERATION_ID}
+  REPO_ID: {REPO_ID}
+  SOURCE_BRANCH: {SOURCE_BRANCH}
+  TARGET_BRANCH: {TARGET_BRANCH}
+  PR_TITLE: {PR_TITLE}
+  PR_DESCRIPTION: {PR_DESCRIPTION}
   PLUGIN_ROOT: {CLAUDE_PLUGIN_ROOT}"
 )
 ```
 
-Store the full output as `ADO_FETCHER_RESULT`. If the `ADO_FETCHER_RESULT_START`/`_END` block is absent (Fetcher exited non-zero), determine the abort kind from the output (output contains `az devops login` → `abortKind: 'auth'`; otherwise `abortKind: 'fetcher'`), call `formatTrailer({ mode: 'aborted', abortKind, abortReason: <first ERROR: line from output> })` from `scripts/ado/notices.mjs`, and stop. Otherwise parse `LATEST_ITERATION_ID`, `REPO_ID`, `CHANGED_FILES`, `RAW_DIFF`, `DIFF_RANGE`, `WORK_ITEM_IDS`, and `NOTICES` from the block. Store `DIFF_RANGE`; the Re-review Coordinator (Step 7) parses it from `ADO_FETCHER_RESULT` to apply the γ-downgrade when `DIFF_RANGE=full`. Set `NOTICES_JSON` to `mergeNotices(NOTICES)` via `scripts/ado/notices.mjs`.
+Store the full output as `ADO_FETCHER_RESULT`. If the `ADO_FETCHER_RESULT_START`/`_END` block is absent (Fetcher exited non-zero), determine the abort kind from the output (output contains `az devops login` → `abortKind: 'auth'`; otherwise `abortKind: 'fetcher'`), call `formatTrailer({ mode: 'aborted', abortKind, abortReason: <first ERROR: line from output> })` from `scripts/ado/notices.mjs`, and stop. Otherwise parse `LATEST_ITERATION_ID`, `CHANGED_FILES`, `RAW_DIFF`, `DIFF_RANGE`, `MODE`, `IS_REREVIEW`, `PRIOR_ITERATION_ID`, `SUMMARY_THREAD_ID`, `RAW_THREADS_JSON`, `WORK_ITEM_IDS`, and `NOTICES` from the block. Store `DIFF_RANGE`; the Re-review Coordinator (Step 7) parses it from `ADO_FETCHER_RESULT` to apply the γ-downgrade when `DIFF_RANGE=full`. Set `NOTICES_JSON` to `mergeNotices(NOTICES)` via `scripts/ado/notices.mjs`. Echo `Mode detected: $MODE`.
 
 ## Step 6 — Doc Context Orchestrator + review aspect agents (parallel)
 
diff --git a/apps/claude-code/pr-review/docs/adr/0012-plain-text-doc-context-agent-return.md b/apps/claude-code/pr-review/docs/adr/0012-plain-text-doc-context-agent-return.md
index 0de8572e..3c7af1c3 100644
--- a/apps/claude-code/pr-review/docs/adr/0012-plain-text-doc-context-agent-return.md
+++ b/apps/claude-code/pr-review/docs/adr/0012-plain-text-doc-context-agent-return.md
@@ -42,4 +42,4 @@ caller-side conditional.
 
 **See also:**
 
-- `docs/plans/11-doc-context-spawn-reliability.md`
+- `docs/plans/11-doc-context-spawn-reliability.md` (retired — see git history for content)
diff --git a/apps/claude-code/pr-review/docs/adr/0013-orchestrator-split-for-review-pr.md b/apps/claude-code/pr-review/docs/adr/0013-orchestrator-split-for-review-pr.md
index 0120b7d1..45058eec 100644
--- a/apps/claude-code/pr-review/docs/adr/0013-orchestrator-split-for-review-pr.md
+++ b/apps/claude-code/pr-review/docs/adr/0013-orchestrator-split-for-review-pr.md
@@ -1,6 +1,6 @@
 # 0013. Split review-pr.md into a thin orchestrator and focused agents
 
-**Status:** Accepted (2026-05)
+**Status:** Accepted (2026-05), amended by 0016
 
 ## Context
 
diff --git a/apps/claude-code/pr-review/docs/adr/0015-canonical-http-tier-mapping.md b/apps/claude-code/pr-review/docs/adr/0015-canonical-http-tier-mapping.md
index 282ab676..602b51db 100644
--- a/apps/claude-code/pr-review/docs/adr/0015-canonical-http-tier-mapping.md
+++ b/apps/claude-code/pr-review/docs/adr/0015-canonical-http-tier-mapping.md
@@ -1,6 +1,6 @@
 # ADR 0015 — Canonical HTTP-Tier Mapping
 
-**Status:** Accepted
+**Status:** Accepted, amended by 0018
 **Date:** 2026-05-13
 **Deciders:** Oriol Torrent Florensa
 **Context:** ADR 0014 (failure-classification helper layer)
@@ -55,6 +55,9 @@ Retries are not implemented. Reasons:
 Re-evaluate if 5xx Notices prove painful in practice; retries can be added behind the same Notice
 surface without changing the doctrine.
 
+**Update (2026-05-14):** this no-retry policy is extended to agent-spawn failures during orchestrator
+fan-out by ADR 0018. The reasoning is identical (AFK latency, retry-storm risk, accurate Notice).
+
 ### Implementation
 
 The canonical mapping is implemented in `scripts/ado/classify-http-error.mjs`:
diff --git a/apps/claude-code/pr-review/docs/adr/0016-fold-thread-fetch-into-ado-fetcher.md b/apps/claude-code/pr-review/docs/adr/0016-fold-thread-fetch-into-ado-fetcher.md
new file mode 100644
index 00000000..f0b04537
--- /dev/null
+++ b/apps/claude-code/pr-review/docs/adr/0016-fold-thread-fetch-into-ado-fetcher.md
@@ -0,0 +1,55 @@
+# 0016. Fold thread fetch into ADO Fetcher
+
+**Status:** Accepted (2026-05)
+**Context:** Amends ADR 0013 (orchestrator split for review-pr)
+**Date:** 2026-05-14
+
+## Context
+
+ADR 0013 split `review-pr.md` into a thin orchestrator and focused agents. As part of that split, ADR 0013 carved out **one allowed inline ADO call**: the mode-detection `az repos pr thread list` in the orchestrator's Step 4. The justification was that mode detection had to happen _before_ the ADO Fetcher could be launched, so the orchestrator needed to fetch and inspect threads itself.
+
+Two facts surfaced after shipping the split:
+
+1. **The carved-out command is invalid syntax.** `az repos pr thread list` does not exist in the `azure-devops` extension — valid `az repos pr` subcommands are `create / list / show / update / work-item / set-vote / reviewer / policy / checkout`. There is no `thread` group. The correct call is `az devops invoke --area git --resource pullRequestThreads --route-parameters …`. Every ADO PR review since the orchestrator-split shipped has been failing at Step 4. The first observed-in-the-wild failure is the 2026-05-14 dry-run captured in `docs/conversations/pr-review-dry-run-01.txt`.
+
+2. **The "fetch threads → derive mode" pair is naturally adjacent to the rest of the Fetcher's work.** The Fetcher already calls `az devops invoke` for `pullRequestIterations`, `pullRequestIterationChanges`, `pullRequestWorkItems`. Adding `pullRequestThreads` to that family keeps ADO read knowledge in one place — the agent that exists exactly to own it.
+
+The carve-out can go away if the Fetcher returns `MODE`, `IS_REREVIEW`, `PRIOR_ITERATION_ID`, `SUMMARY_THREAD_ID`, and `RAW_THREADS_JSON` in its structured result block. The orchestrator branches on those fields in Step 5 onwards — the same pattern it already uses for `WORK_ITEM_IDS`, `DIFF_RANGE`, etc.
+
+## Decision
+
+**Move thread fetching and mode detection from the orchestrator into the ADO Fetcher.** The Fetcher gains a new step (Step 2.5) between iterations fetch and changed-files fetch:
+
+1. Call `az devops invoke --area git --resource pullRequestThreads --route-parameters "project=$PROJECT" "repositoryId=$REPO_ID" "pullRequestId=$PR_ID"`.
+2. Apply ADR 0015's HTTP-tier mapping: 401/403 → ABORTED; 404 → OK (treat as empty threads; equivalent to first-review); 5xx / network → DEGRADED with `kind: thread-fetch`, treat as empty threads so mode detection still produces `first-review`.
+3. Run `detectMode` (from `scripts/mode-detection.mjs`, unchanged) on the response's `.value` array. The function returns `{ mode, isRereview, priorIterationId, summaryThreadId }`.
+4. Append `RAW_THREADS_JSON`, `MODE`, `IS_REREVIEW`, `PRIOR_ITERATION_ID`, `SUMMARY_THREAD_ID` to the Fetcher's result block.
+
+The orchestrator's Step 4 simplifies to a single `az repos pr show` call that captures PR metadata (`REPO_ID`, `PROJECT`, `SOURCE_BRANCH`, `TARGET_BRANCH`, `PR_TITLE`, `PR_DESCRIPTION`) and passes them into the Fetcher as inputs. The Fetcher no longer makes its own `az repos pr show` call — the data is already in its prompt.
+
+After this change, the orchestrator contains **no `az devops invoke` calls**. The only remaining inline `az` is `az --version`, `az extension list` (Step 3 preflight), and `az repos pr show` (Step 4 metadata).
+
+ADR 0013's carve-out (_"The one allowed inline ADO call is the mode-detection `az repos pr thread list` in the mode detection block"_) is **removed**. ADR 0013's core decision — thin orchestrator + focused agents — stands; only that one-sentence exception is superseded by this ADR.
+
+## Alternatives considered
+
+**Fix the inline command in Step 4 and leave it in the orchestrator.** Smallest change. Rejected because it preserves the architectural reason for the carve-out (mode detection must precede Fetcher launch) which doesn't actually exist — the orchestrator can equally well branch on Fetcher output. Keeping ADO knowledge in two places means future API changes require edits in two files.
+
+**Introduce a separate `mode-detector` agent that fetches threads.** Cleaner separation of concerns. Rejected because mode detection and the rest of the Fetcher's work share the same ADO call shape (`az devops invoke --area git --resource …`); a new agent would duplicate the auth/retry/error-classification setup the Fetcher already has.
+
+**Replace `az` with direct REST calls.** Long-discussed in ADR 0008's neighbourhood. Out of scope for this ADR; revisit if `az` proves a recurring failure source after spec 14's smoke test catches near-term drift.
+
+## Consequences
+
+- The plugin's mode-detection path actually works (Step 4 was previously broken on every run).
+- ADO read knowledge is centralised in `ado-fetcher.md`.
+- Adding new mode-affecting ADO data (e.g. a `mergeAttempt` field) requires editing one agent prompt.
+- The Fetcher's result block grows by five lines — downstream prompts (Coordinator, Writer) parse them via existing string-extraction patterns, no helper change.
+- ADR 0013's status line is updated to `Accepted (2026-05), amended by 0016`.
+
+## See also
+
+- ADR 0013 — Orchestrator split for review-pr (amended)
+- ADR 0014 — Notice Tier doctrine (governs the new `kind: thread-fetch` DEGRADED path)
+- ADR 0015 — Canonical HTTP-tier mapping (governs the 401/403/5xx behaviour for thread fetch)
+- Spec 12 — Implementation of this ADR
diff --git a/apps/claude-code/pr-review/docs/adr/0017-dry-run-as-fourth-peer-mode.md b/apps/claude-code/pr-review/docs/adr/0017-dry-run-as-fourth-peer-mode.md
new file mode 100644
index 00000000..ca4d5b8b
--- /dev/null
+++ b/apps/claude-code/pr-review/docs/adr/0017-dry-run-as-fourth-peer-mode.md
@@ -0,0 +1,46 @@
+# 0017. Dry-run as a fourth peer operating mode
+
+**Status:** Accepted (2026-05)
+**Date:** 2026-05-14
+
+## Context
+
+The 2026-05-14 dry-run captured in `docs/conversations/pr-review-dry-run-01.txt` was invoked with the natural-language instruction _"Make a dry-run PR review. DO NOT POST ANY COMMENT TO THE PR! Only report inline"_. The plugin had no formal dry-run mechanism. The LLM-as-orchestrator improvised: inlined `az` data fetches (compounding the Step 4 bug), skipped the `pr-review:ado-fetcher` agent entirely, and never invoked the ADO Writer. The user got useful findings but only by accident — the documented spec was bypassed.
+
+Two design problems:
+
+1. **No deterministic dry-run path.** Asking the LLM to "not post comments" works most of the time but invites free-form deviation from the spec. A documented mode replaces persuasion with branching.
+2. **Dry-run is genuinely useful and not just a debug helper.** Previewing what a re-review would post — before letting it edit threads in a customer-facing PR — is a legitimate user need.
+
+Two shapes for the formalisation: a fourth peer mode, or an orthogonal boolean flag.
+
+## Decision
+
+**Dry-run is a fourth peer mode, alongside pre-PR, first-review, and re-review.** CLI surface: `/pr-review:review-pr <url> --dry-run`. The orchestrator's mode resolution applies the flag after URL parsing; `IS_REREVIEW` (whether a prior bot signature exists) is captured separately and drives Coordinator inclusion within dry-run mode.
+
+Dry-run runs every read-side step identically to first-review / re-review (preflight, metadata fetch, ADO Fetcher, Doc Context Orchestrator, Review Aspects fan-out, and Re-review Coordinator's Thread Classification when `IS_REREVIEW=true`). The **ADO Writer is never invoked**. Findings render to the Claude interface using pre-PR Step E's severity-grouped format. Notices from Fetcher and Coordinator render via `formatNoticesAsPrePrPreamble`. The Trailer reads `🔍 Dry-run complete: <N> findings (<criticals> critical, <importants> important) · <warnings> warning notices · would have posted to <PR URL>`.
+
+Coordinator runs in dry-run mode when `IS_REREVIEW=true`, but its reply-posting branch is short-circuited. Its `freshFindings` output still feeds `FINDINGS_JSON` so the user previews exactly the incremental finding set the real re-review would have posted.
+
+## Alternatives considered
+
+**Orthogonal `DRY_RUN` boolean flag.** Any mode (first-review / re-review) could be dry-run. Conceptually cleaner — dry-run is "do the work, don't post" — but produces awkward mode names internally (`first-review-dry-run`, `re-review-dry-run`) and forces every branch in the orchestrator to read two state variables instead of one. Rejected: peer-mode keeps the orchestrator's mode-switch readable as a single discriminator.
+
+**Document a "DO NOT POST" convention and trust the LLM.** Zero implementation cost. Rejected because it is exactly what failed on 2026-05-14: the LLM honoured the instruction by skipping not just Writer but also the entire structured fetch path, masking the Step 4 bug and producing inconsistent behaviour across runs.
+
+**Make dry-run a flag that disables only the Writer agent invocation.** Halfway between the other two. Rejected because the orchestrator's other branches (Step 8 trailer text, Notice rendering) still need to know whether to use the pre-PR rendering path or the ADO Summary path. A single mode value carries that decision uniformly.
+
+## Consequences
+
+- The orchestrator branches on `MODE ∈ {pre-pr, dry-run, first-review, re-review}` — four cases, no nested flags.
+- Re-review-eligible PRs can be previewed safely: classification runs and reports its plan, but no replies are posted and no thread statuses are PATCHed.
+- The Pre-PR rendering helpers (`formatNoticesAsPrePrPreamble`, severity-grouped finding print) are reused by dry-run — confirming that "render to interface" is a coherent capability, not a pre-PR-only quirk.
+- The Trailer line gets a fourth shape (`🔍 Dry-run complete: ...`). `formatTrailer` in `scripts/ado/notices.mjs` grows one branch + test cases.
+- `CONTEXT.md` gains a `Dry-run mode` term and updates three relationship lines documenting which agents run in which modes.
+
+## See also
+
+- Spec 13 — Implementation of this ADR
+- Spec 12 — Step 4 fix (the upstream bug that exposed the no-formal-dry-run gap)
+- ADR 0014 — Notice Tier doctrine (`formatTrailer`, `formatNoticesAsPrePrPreamble` are governed here)
+- `CONTEXT.md` — `Dry-run mode` definition and the agent-invocation matrix
diff --git a/apps/claude-code/pr-review/docs/adr/0018-fan-out-resilience-trust-harness.md b/apps/claude-code/pr-review/docs/adr/0018-fan-out-resilience-trust-harness.md
new file mode 100644
index 00000000..3fb2815e
--- /dev/null
+++ b/apps/claude-code/pr-review/docs/adr/0018-fan-out-resilience-trust-harness.md
@@ -0,0 +1,60 @@
+# 0018. Fan-out resilience — trust the harness, surface partial failure as a Notice
+
+**Status:** Accepted (2026-05)
+**Context:** Amends ADR 0015 (canonical HTTP-tier mapping — no-retry doctrine)
+**Date:** 2026-05-14
+
+## Context
+
+The 2026-05-14 dry-run captured in `docs/conversations/pr-review-dry-run-01.txt` produced a transient `API Error: The socket connection was closed unexpectedly` from the Claude Code HTTP client during Step 6's parallel agent fan-out. Step 6 is the plugin's largest concurrent spawn: Doc Context Orchestrator plus 2–5 review-aspect agents in a single message.
+
+The error printed in the output but all spawned agents completed normally. The Claude Code harness either auto-retried the launch transparently or the server-side launch had already been accepted before the client socket dropped. The user-visible outcome was a successful run that _looked_ broken.
+
+Two unaddressed risks:
+
+1. **Cosmetic confusion.** A scary API error appeared with no framing. AFK users skimming the output can't tell whether a partial failure occurred.
+2. **Real partial-failure invisibility.** If a future fan-out _does_ lose one agent (e.g. the harness fails to recover), the orchestrator currently has no post-condition that detects the missing result block. The Review posts with N-1 aspects' findings and the user has no signal that coverage was reduced.
+
+ADR 0015 already forbids retries for HTTP failures in v1 with three reasons: (1) AFK latency, (2) retry-storm failure mode, (3) DEGRADED Notices carrying accurate information. All three reasons apply identically to agent-spawn failures during fan-out. This ADR extends ADR 0015's doctrine to the orchestrator's fan-out layer.
+
+## Decision
+
+**Trust the harness. Don't retry. Surface partial fan-out as a DEGRADED Notice.**
+
+After Step 6's fan-out completes, the orchestrator runs a post-condition check: every launched Review Aspect agent must have returned a parseable JSON findings array. Agents that return nothing (no result block, no JSON, empty stdout) are recorded as missing.
+
+If any aspects are missing, the orchestrator emits a single Notice:
+
+```js
+{ severity: 'warning', kind: 'agent-spawn',
+  message: 'Review Aspects did not return findings: <name>, <name>. Coverage is reduced for this run.' }
+```
+
+The Notice flows through the same pipeline as Fetcher and Coordinator notices: merged via `mergeNotices`, rendered in the Review Summary (or pre-PR / dry-run preamble), counted in the Trailer's `warning notices` field.
+
+The Review still posts with whatever findings did arrive. No retry. No double-spawn risk.
+
+The `agent-spawn` kind is added to the `NoticeKind` enum in `scripts/ado/notices.mjs`. Per ADR 0014's invariant _"each `kind` has exactly one source agent"_, the source agent is **the orchestrator itself** — it is the only place that knows which agents were launched.
+
+## Alternatives considered
+
+**Retry missing agents sequentially after fan-out.** The orchestrator detects the missing ones and re-spawns them in a second pass. Rejected: the harness may already have launched them server-side (the dry-run example demonstrates the harness recovers transparently). Re-spawning risks double inferences, double findings, and confusing the user with duplicate output. No idempotency guarantee.
+
+**Catch the socket error itself and retry the whole fan-out.** Guaranteed double-spawn of agents that did succeed — explicitly worse than the previous option.
+
+**Add a feature-flagged single retry behind a configuration option.** Postponed. ADR 0015's clause _"Re-evaluate if 5xx Notices prove painful in practice"_ applies equivalently here. If `agent-spawn` Notices become a frequent reality, this ADR can be revisited with the same opt-in approach.
+
+**Treat partial fan-out as ABORTED.** Rejected: one missing aspect (e.g. `comment-analyzer`) is not run-corrupting. The Notice tells the user; they can rerun if needed. ABORTED is reserved for failures that would corrupt cross-run state.
+
+## Consequences
+
+- The orchestrator gains a single post-condition check after Step 6 — a few lines of bash + a `createNotice` call. No new helper, no new test infrastructure beyond extending the existing `tests/notices.test.mjs` cases.
+- The user always sees a documented status: either findings (success), findings + `agent-spawn` Notice (partial), or aborted Trailer (full failure upstream).
+- Future failure modes affecting fan-out (e.g. a `pr-review-toolkit` upgrade introducing a non-JSON response) become observable via the same Notice, without a new failure-classification path.
+- ADR 0015's no-retry doctrine is documented as load-bearing across two layers (HTTP and agent-spawn). Its status line gains `, amended by 0018` and its "No retries in v1" section closes with a cross-reference to this ADR.
+
+## See also
+
+- ADR 0014 — Notice Tier doctrine (`agent-spawn` is a new `kind` in the enum)
+- ADR 0015 — Canonical HTTP-tier mapping (amended)
+- Spec 15 — Implementation of this ADR
diff --git a/apps/claude-code/pr-review/docs/adr/README.md b/apps/claude-code/pr-review/docs/adr/README.md
index 75cb0c98..e4120420 100644
--- a/apps/claude-code/pr-review/docs/adr/README.md
+++ b/apps/claude-code/pr-review/docs/adr/README.md
@@ -5,19 +5,23 @@ See the root `docs/adr/README.md` for format and numbering conventions.
 
 ## Index
 
-| ID   | Title                                                                                   | Status     |
-| ---- | --------------------------------------------------------------------------------------- | ---------- |
-| 0001 | Canonical bot signature                                                                 | Accepted   |
-| 0002 | Signature-based prior-review detection                                                  | Accepted   |
-| 0003 | Target latest PR iteration                                                              | Accepted   |
-| 0004 | Incremental diff baseline                                                               | Accepted   |
-| 0005 | Four-state thread classification                                                        | Accepted   |
-| 0006 | Reply to existing threads instead of opening duplicates; auto-resolve addressed threads | Accepted   |
-| 0007 | Summary comment is rewritten on re-review, not appended                                 | Superseded |
-| 0008 | Soft dependency on pr-review-toolkit                                                    | Accepted   |
-| 0009 | Re-review summary delta is posted as a reply to the existing summary thread             | Accepted   |
-| 0010 | Inline Confluence client                                                                | Accepted   |
-| 0011 | Additive parallel paths for doc-context extensibility                                   | Accepted   |
-| 0012 | Plain-text Doc-Context agent return                                                     | Accepted   |
-| 0013 | Orchestrator split for review-pr                                                        | Accepted   |
-| 0014 | Notice Tier doctrine and failure-classification helpers                                 | Accepted   |
+| ID   | Title                                                                                   | Status                     |
+| ---- | --------------------------------------------------------------------------------------- | -------------------------- |
+| 0001 | Canonical bot signature                                                                 | Accepted                   |
+| 0002 | Signature-based prior-review detection                                                  | Accepted                   |
+| 0003 | Target latest PR iteration                                                              | Accepted                   |
+| 0004 | Incremental diff baseline                                                               | Accepted                   |
+| 0005 | Four-state thread classification                                                        | Accepted                   |
+| 0006 | Reply to existing threads instead of opening duplicates; auto-resolve addressed threads | Accepted                   |
+| 0007 | Summary comment is rewritten on re-review, not appended                                 | Superseded                 |
+| 0008 | Soft dependency on pr-review-toolkit                                                    | Accepted                   |
+| 0009 | Re-review summary delta is posted as a reply to the existing summary thread             | Accepted                   |
+| 0010 | Inline Confluence client                                                                | Accepted                   |
+| 0011 | Additive parallel paths for doc-context extensibility                                   | Accepted                   |
+| 0012 | Plain-text Doc-Context agent return                                                     | Accepted                   |
+| 0013 | Orchestrator split for review-pr                                                        | Accepted (amended by 0016) |
+| 0014 | Notice Tier doctrine and failure-classification helpers                                 | Accepted                   |
+| 0015 | Canonical HTTP-tier mapping                                                             | Accepted (amended by 0018) |
+| 0016 | Fold thread fetch into ADO Fetcher                                                      | Accepted                   |
+| 0017 | Dry-run as a fourth peer operating mode                                                 | Accepted                   |
+| 0018 | Fan-out resilience — trust the harness                                                  | Accepted                   |
diff --git a/apps/claude-code/pr-review/docs/issues/pr-review-ado-fetcher-step4-fix/01-fold-thread-fetch-into-fetcher.md b/apps/claude-code/pr-review/docs/issues/pr-review-ado-fetcher-step4-fix/01-fold-thread-fetch-into-fetcher.md
new file mode 100644
index 00000000..f0eff7ff
--- /dev/null
+++ b/apps/claude-code/pr-review/docs/issues/pr-review-ado-fetcher-step4-fix/01-fold-thread-fetch-into-fetcher.md
@@ -0,0 +1,55 @@
+# 01. Fold thread fetch + mode detection into ADO Fetcher
+
+**Status:** ready-for-agent
+**Category:** bug
+**Plugin:** `apps/claude-code/pr-review`
+**Type:** AFK
+
+## Parent
+
+`apps/claude-code/pr-review/docs/issues/pr-review-ado-fetcher-step4-fix/PRD.md`
+
+## What to build
+
+Move thread fetching and mode detection from the orchestrator's Step 4 into the ADO Fetcher, per [ADR 0016](../../adr/0016-fold-thread-fetch-into-ado-fetcher.md) (already merged). The bug is currently live: every ADO PR review fails at Step 4's invalid `az repos pr thread list` invocation; the LLM-as-orchestrator improvises around it, masking the failure.
+
+Implementation cuts through every layer:
+
+- **Orchestrator (`commands/review-pr.md`)** — replace Step 4's `az repos pr thread list` block with a single `az repos pr show --id "$PR_ID" --org "$ORG_URL" --output json` call. Parse `repository.id`, `repository.project.name`, `sourceRefName`, `targetRefName`, `title`, `description` into `REPO_ID`, `PROJECT`, `SOURCE_BRANCH`, `TARGET_BRANCH`, `PR_TITLE`, `PR_DESCRIPTION` (stripping `refs/heads/`). Pass these into the Fetcher's prompt. Step 5 parses `MODE`, `IS_REREVIEW`, `PRIOR_ITERATION_ID`, `SUMMARY_THREAD_ID`, `RAW_THREADS_JSON` from the Fetcher's result block. The orchestrator must contain **no `az devops invoke` calls** after this slice — only `az --version`, `az extension list`, `az repos pr show`.
+
+- **ADO Fetcher (`agents/ado-fetcher.md`)** — accept the new inputs (`REPO_ID`, `PROJECT`, `SOURCE_BRANCH`, `TARGET_BRANCH`, `PR_TITLE`, `PR_DESCRIPTION`) as literal-string inputs in the prompt. Remove the existing Step 1 (`az repos pr show`) since metadata is now passed in. Insert a new step between iterations and changed-files: fetch threads via `az devops invoke --area git --resource pullRequestThreads --route-parameters "project=$PROJECT" "repositoryId=$REPO_ID" "pullRequestId=$PR_ID" --org "$ORG_URL" --api-version "7.1" --output json`. Apply ADR 0015's HTTP-tier mapping via `scripts/ado/classify-http-error.mjs`: `401/403` → ABORTED with `az devops login` hint; `404` → OK, treat `RAW_THREADS_JSON` as `{"value":[]}`; `5xx / network` → DEGRADED Notice with `kind: thread-fetch`, treat as empty threads. Run `detectMode` from `scripts/mode-detection.mjs` against `.value`. Extend the `ADO_FETCHER_RESULT_START/END` block with `RAW_THREADS_JSON`, `MODE`, `IS_REREVIEW`, `PRIOR_ITERATION_ID`, `SUMMARY_THREAD_ID`.
+
+- **`scripts/mode-detection.mjs`** — export a `SIGNATURE_PREFIX` constant: `export const SIGNATURE_PREFIX = '🤖 *Reviewed by Claude Code*'`. Update the Fetcher to import it directly. Remove the stale `az repos pr thread list` reference in the module's comment (line 27).
+
+- **`scripts/ado/notices.mjs`** — add `thread-fetch` to the `NoticeKind` union.
+
+- **Re-review Coordinator (`agents/re-review-coordinator.md`)** — update the `RAW_THREADS_JSON` description (line 27) and the "No re-fetch of threads" note (line 471) to reflect that the data now flows from the Fetcher's output block via the orchestrator, not from the (broken) orchestrator-level `az repos pr thread list`.
+
+- **ADR 0013 status line** — already not yet amended; update to `Accepted (2026-05), amended by 0016`.
+
+- **CHANGELOG** — `[Unreleased]` `### Fixed` entry: _Step 4 mode detection was calling a non-existent `az repos pr thread list` subcommand and failing fatally on every ADO PR review. Thread fetching now lives in the ADO Fetcher and uses `az devops invoke --resource pullRequestThreads`._
+
+End-to-end demoable: `/pr-review:review-pr <fresh PR URL>` runs cleanly with `MODE: first-review` in the Fetcher output. `/pr-review:review-pr <PR with prior bot signature>` runs cleanly with `MODE: re-review`, `IS_REREVIEW: true`, and a non-empty `PRIOR_ITERATION_ID`. No `az repos pr thread list` anywhere in the source tree outside `docs/conversations/`.
+
+## Acceptance criteria
+
+- [ ] `grep -rn "az repos pr thread list" agents/ commands/ scripts/ tests/` returns no matches.
+- [ ] `commands/review-pr.md` contains no `az devops invoke` lines.
+- [ ] `commands/review-pr.md` Step 4 captures `REPO_ID`, `PROJECT`, `SOURCE_BRANCH`, `TARGET_BRANCH`, `PR_TITLE`, `PR_DESCRIPTION` from `az repos pr show` and passes them into the Fetcher prompt.
+- [ ] `commands/review-pr.md` is ≤ 200 lines (orchestrator-thin invariant from ADR 0013).
+- [ ] `agents/ado-fetcher.md` contains a step that calls `az devops invoke --area git --resource pullRequestThreads` and invokes `detectMode`.
+- [ ] `agents/ado-fetcher.md` no longer calls `az repos pr show`.
+- [ ] The Fetcher's result block emits `RAW_THREADS_JSON`, `MODE`, `IS_REREVIEW`, `PRIOR_ITERATION_ID`, `SUMMARY_THREAD_ID`.
+- [ ] On 5xx/network on the threads endpoint, the Fetcher emits a `warning`-severity Notice with `kind: thread-fetch` and proceeds with `MODE: first-review`.
+- [ ] On 401/403 on the threads endpoint, the Fetcher aborts with an `az devops login` hint.
+- [ ] On 404 on the threads endpoint, the Fetcher proceeds with `RAW_THREADS_JSON={"value":[]}` and `MODE: first-review`.
+- [ ] `scripts/mode-detection.mjs` exports `SIGNATURE_PREFIX = '🤖 *Reviewed by Claude Code*'`.
+- [ ] `tests/mode-detection.test.mjs` includes a case asserting `SIGNATURE_PREFIX === '🤖 *Reviewed by Claude Code*'`.
+- [ ] `agents/re-review-coordinator.md` no longer references `az repos pr thread list` as the source of `RAW_THREADS_JSON`.
+- [ ] `apps/claude-code/pr-review/docs/adr/0013-orchestrator-split-for-review-pr.md` status line ends with `, amended by 0016`.
+- [ ] `CHANGELOG.md` under `[Unreleased]` has a `### Fixed` entry describing the Step 4 bug fix and the fold-into-Fetcher refactor.
+- [ ] `pnpm format`, `pnpm check`, `pnpm --filter pr-review test`, `pnpm --filter pr-review typecheck`, `pnpm --filter pr-review verify:changelog` all pass.
+
+## Blocked by
+
+None — can start immediately.
diff --git a/apps/claude-code/pr-review/docs/issues/pr-review-ado-fetcher-step4-fix/02-ado-cli-inventory-and-smoke-test.md b/apps/claude-code/pr-review/docs/issues/pr-review-ado-fetcher-step4-fix/02-ado-cli-inventory-and-smoke-test.md
new file mode 100644
index 00000000..6c0e85de
--- /dev/null
+++ b/apps/claude-code/pr-review/docs/issues/pr-review-ado-fetcher-step4-fix/02-ado-cli-inventory-and-smoke-test.md
@@ -0,0 +1,60 @@
+# 02. ADO CLI inventory + completeness test + smoke test + preflight
+
+**Status:** ready-for-agent
+**Category:** enhancement
+**Plugin:** `apps/claude-code/pr-review`
+**Type:** AFK
+
+## Parent
+
+`apps/claude-code/pr-review/docs/issues/pr-review-ado-fetcher-step4-fix/PRD.md`
+
+## What to build
+
+Add a single source of truth for every `az` command the plugin uses, plus two automated checks that prevent silent drift between source code and that inventory. The motivation is timeless: Microsoft renames or removes `az` subcommands without notice (the original Step 4 bug from Slice 01 is exactly this failure class, just hand-coded against an invalid command from the start).
+
+Implementation cuts through every layer:
+
+- **`tests/fixtures/ado-cli-inventory.mjs`** (new) — exports `adoCliInventory`, an array of `{ kind: 'invoke' | 'repos' | 'boards', area?, resource?, command: string[], helpKeywordsRequired: string[] }` entries. Initial entries cover everything the plugin uses post-Slice-01: `az repos pr show`, `az repos pr checkout`, `az boards work-item show`, and `az devops invoke --area git --resource {pullRequestThreads | pullRequestIterations | pullRequestIterationChanges | pullRequestWorkItems | pullRequestThreadComments}`.
+
+- **`tests/fixtures/ado-cli-allowlist.mjs`** (new) — exports `ADO_CLI_ALLOWLIST`, regex patterns for `az` invocations intentionally NOT in the inventory: `az --version`, `az extension list`, error-message-only hints like `az devops login` and `az extension add`, and the preflight `az devops invoke --help`. Each entry carries an inline comment explaining its exemption.
+
+- **`scripts/ado/cli-completeness.mjs`** (new — follows the `scripts/ado/` deep-helper pattern PRD A established) — exports `findUninventoriedCommands({ sources, inventory, allowlist }) → string[]`. Pure function; takes source-file contents (not paths) and returns the list of unregistered `az` command shapes. Must handle multi-line bash blocks, backslash continuations, and `--area`/`--resource` flags split across lines.
+
+- **`tests/ado-cli-completeness.test.mjs`** (new) — fixture-style unit tests for `findUninventoriedCommands`: empty source, single inline command, multi-line bash with backslash continuation, `--area`/`--resource` split across lines, allowlist filtering, mixed allowlisted + inventoried + uninventoried. Mirrors the `tests/parse-diff-hunks.test.mjs` style.
+
+- **`tests/ado-cli-smoke.test.mjs`** (new) — two test cases:
+
+  1. **Inventory completeness** — reads `agents/`, `commands/`, `scripts/` source files via `fs.readFileSync`; calls `findUninventoriedCommands` with the real inventory + allowlist; asserts the returned array is empty. Pure-JS; runs everywhere.
+  2. **CLI smoke** — iterates `adoCliInventory`; runs each entry's `--help` via `child_process.spawnSync` with a 5s timeout; asserts exit 0 and that every `helpKeywordsRequired` substring appears in stdout. On `ENOENT` (i.e. `az` not on PATH): `t.skip('az CLI not installed')`.
+
+- **Orchestrator Step 3 preflight (`commands/review-pr.md`)** — append after the existing `az --version` / `az extension list` checks: `if ! az devops invoke --help >/dev/null 2>&1; then echo "ERROR: az devops invoke unavailable. Re-install: az extension remove --name azure-devops && az extension add --name azure-devops" >&2; exit 1; fi`.
+
+- **CI workflow (`.github/workflows/ci.yml`)** — add a single conditional step (`if: matrix.os == 'ubuntu-latest' && matrix.node == 24`) that installs `azure-cli` via `apt` and adds the `azure-devops` extension. Other matrix cells skip the CLI smoke test cleanly via the `t.skip` path; the completeness test runs everywhere.
+
+- **`apps/claude-code/pr-review/AGENTS.md`** — append a single pointer paragraph to the "External dependencies" section: _"All `az` commands the plugin uses are enumerated in `tests/fixtures/ado-cli-inventory.mjs`. The smoke test (`tests/ado-cli-smoke.test.mjs`) asserts every `az ` invocation in `agents/`/`commands/`/`scripts/` has a matching inventory entry — modulo the allowlist in `tests/fixtures/ado-cli-allowlist.mjs`. Register a new ADO call in the inventory before invoking it."_ No mirrored command table.
+
+- **CHANGELOG** — `[Unreleased]` `### Changed` entry: _Preflight now verifies `az devops invoke` is callable; CI smoke test asserts every ADO subcommand the plugin uses exists._
+
+End-to-end demoable: `pnpm --filter pr-review test` passes locally with `az` installed (smoke test runs) and without `az` (smoke test skips, completeness test still runs). Deliberately adding an unregistered `az` call to any agent → completeness test fails. Adding a known-bad inventory entry (e.g. `az repos pr thread list`) → CLI smoke test fails with the `az` rename / removal as the surfaced reason.
+
+## Acceptance criteria
+
+- [ ] `tests/fixtures/ado-cli-inventory.mjs` exists and lists every `az` command the plugin uses post-Slice-01 (including `pullRequestThreads` introduced there).
+- [ ] `tests/fixtures/ado-cli-allowlist.mjs` exists; each entry has an inline comment explaining the exemption.
+- [ ] `scripts/ado/cli-completeness.mjs` exports `findUninventoriedCommands` as a pure function and handles multi-line bash, backslash continuations, and split `--area`/`--resource` flags.
+- [ ] `tests/ado-cli-completeness.test.mjs` covers: empty source, single inline command, multi-line block, split flags, allowlist filtering, mixed scenarios.
+- [ ] `tests/ado-cli-smoke.test.mjs` has two cases: completeness (runs everywhere) and CLI smoke (`t.skip` when `az` absent).
+- [ ] When `az` is on PATH, the CLI smoke test runs `--help` against every inventory entry and asserts exit 0 + every `helpKeywordsRequired` substring appears in stdout.
+- [ ] When `az` is absent, the CLI smoke test skips cleanly; the completeness test still runs.
+- [ ] `commands/review-pr.md` Step 3 asserts `az devops invoke --help` exits 0; on failure, exits with the re-install instructions.
+- [ ] `commands/review-pr.md` is ≤ 200 lines.
+- [ ] `.github/workflows/ci.yml` has a conditional install step (`matrix.os == 'ubuntu-latest' && matrix.node == 24`) that installs `azure-cli` + the `azure-devops` extension; no install on other matrix cells.
+- [ ] CI passes on Linux, macOS, Windows × Node 22, 24.
+- [ ] `apps/claude-code/pr-review/AGENTS.md` contains the pointer paragraph; no command table.
+- [ ] `CHANGELOG.md` under `[Unreleased]` has a `### Changed` entry describing the preflight expansion and the smoke test.
+- [ ] `pnpm format`, `pnpm check`, `pnpm --filter pr-review test`, `pnpm --filter pr-review typecheck`, `pnpm --filter pr-review verify:changelog` all pass.
+
+## Blocked by
+
+`apps/claude-code/pr-review/docs/issues/pr-review-ado-fetcher-step4-fix/01-fold-thread-fetch-into-fetcher.md` — the inventory must include the new `pullRequestThreads` invocation that Slice 01 introduces, and the completeness check would fail until the stale `az repos pr thread list` references that Slice 01 removes are gone.
diff --git a/apps/claude-code/pr-review/docs/issues/pr-review-ado-fetcher-step4-fix/PRD.md b/apps/claude-code/pr-review/docs/issues/pr-review-ado-fetcher-step4-fix/PRD.md
new file mode 100644
index 00000000..c71219de
--- /dev/null
+++ b/apps/claude-code/pr-review/docs/issues/pr-review-ado-fetcher-step4-fix/PRD.md
@@ -0,0 +1,95 @@
+# PRD: pr-review — ADO Fetcher Step 4 fix + ADO CLI smoke test
+
+**Status:** ready-for-agent
+**Category:** bug
+**Plugin:** `apps/claude-code/pr-review`
+**GitHub:** [#120](https://github.com/unic/unic-agents-plugins/issues/120)
+
+---
+
+> **Alignment note (2026-05-26): this PRD is aligned with [ADR 0016](../../adr/0016-fold-thread-fetch-into-ado-fetcher.md) as merged.** Three points in the original GitHub issue body diverged from the merged ADR and are corrected here:
+>
+> 1. **Thread-fetch 5xx/network handling.** Applies ADR 0015's standard mapping: `5xx / network → DEGRADED` with `kind: thread-fetch`, treating empty threads as first-review. The earlier "ABORTED exemption" framing is superseded.
+> 2. **ADR amendment chain.** ADR 0016 amends only ADR 0013. ADR 0015 was separately amended by ADR 0018 (fan-out resilience), unrelated to thread fetch.
+> 3. **`kind: thread-fetch` Notice.** A new `NoticeKind` IS added (DEGRADED tier for 5xx / network on the threads endpoint).
+>
+> **Foundation context.** The two prior pr-review PRDs (`pr-review-ado-fetcher-reliability`, `pr-review-platform-failure-handling`) have both landed. The 4-tier Notice doctrine, `scripts/ado/` helpers (`classify-http-error`, `notices`, `parse-write-response`), `DIFF_RANGE` sentinel, and Trailer are in place. Spec 12 plugs into that foundation — no overlap, clean dependency.
+
+## Problem Statement
+
+Since the pr-review orchestrator-split shipped (ADR 0013), every ADO PR Review fails at Step 4. The orchestrator calls `az repos pr thread list` — a CLI subcommand that does not exist in the `azure-devops` extension. The LLM-as-orchestrator silently improvises around the broken command instead of bailing, bypassing the ADO Fetcher entirely and producing inconsistent reviews. Observed symptoms: comments anchored to wrong iterations, mode misdetection on re-review-eligible PRs, and a fragile dependence on the LLM guessing the right metadata.
+
+Additionally, there is no test that asserts every `az` command the plugin uses actually exists in the installed extension — so a Microsoft rename or removal of any subcommand would surface only in production, exactly as `az repos pr thread list` did.
+
+## Solution
+
+Fold thread fetching and mode detection into the ADO Fetcher (the agent that already owns ADO read operations). The orchestrator's Step 4 becomes a single `az repos pr show` metadata call; threads are fetched via the canonical `az devops invoke --area git --resource pullRequestThreads` endpoint inside the Fetcher; the Fetcher returns mode-derived fields in its output block.
+
+Add a CI smoke test that runs `--help` on every `az` command the plugin uses, plus an inventory-completeness check that grep-validates every `az ` invocation in the source against a registered inventory file.
+
+## User Stories
+
+1. As a developer running `/pr-review:review-pr <ADO PR URL>`, I want Inline Comments to land at the correct file and line in Azure DevOps, so that my team gets the review feedback in the place it belongs.
+2. As a developer reviewing a PR for the second time, I want Re-review mode to be correctly detected, so that the plugin replies to existing Review Threads instead of duplicating them.
+3. As a developer hitting a transient 5xx on thread fetch, I want a DEGRADED Notice surfaced in the Review Summary and the run to proceed in first-review mode, consistent with ADR 0015's HTTP-tier mapping.
+4. As a developer running the plugin on a fresh ADO PR with no prior threads (404 on the threads endpoint), I want the run to proceed cleanly as first-review, so that "no threads yet" is not treated as an error.
+5. As a developer hitting an auth failure (401/403) on thread fetch, I want a clear `az devops login` hint and an aborted run, so that I know the recovery action.
+6. As a plugin maintainer, I want every `az` command the plugin uses to be registered in a single inventory file, so that there is one source of truth for the plugin's ADO CLI surface.
+7. As a plugin maintainer, I want CI to fail when an `az` subcommand the plugin uses no longer resolves to a real command, so that Microsoft renames don't ship undetected.
+8. As a plugin maintainer adding a new `az` call to any agent, I want a test to fail until I've registered the command in the inventory, so that I can't silently expand the plugin's CLI surface without test coverage.
+9. As a plugin maintainer, I want the `SIGNATURE_PREFIX` literal to live as an exported constant in `mode-detection.mjs`, so that there is no risk of string drift between the orchestrator and the Fetcher.
+10. As a CI operator, I want the smoke test to install `azure-cli` on exactly one Linux matrix cell, so that the protection is real without paying install cost on every cell.
+11. As a developer running tests locally without `azure-cli` installed, I want the CLI smoke test to skip cleanly while the inventory-completeness test still runs, so that I can verify the inventory contract without installing `az`.
+
+## Implementation Decisions
+
+**Architectural changes (spec 12 — Slice 01):**
+
+- Thread fetch + mode detection move from the orchestrator's Step 4 into the ADO Fetcher. Recorded in [ADR 0016](../../adr/0016-fold-thread-fetch-into-ado-fetcher.md) (already merged).
+- The Fetcher loses its current Step 1 (`az repos pr show`). That call moves to the orchestrator's Step 4, and the captured fields (`REPO_ID`, `PROJECT`, `SOURCE_BRANCH`, `TARGET_BRANCH`, `PR_TITLE`, `PR_DESCRIPTION`) are passed to the Fetcher as literal-string inputs in its prompt.
+- The Fetcher gains a new step between iterations and changed-files: fetch threads via `az devops invoke --area git --resource pullRequestThreads --route-parameters "project=… repositoryId=… pullRequestId=…"`, then call `detectMode` on the response.
+- The Fetcher's output block grows by five fields: `RAW_THREADS_JSON`, `MODE`, `IS_REREVIEW`, `PRIOR_ITERATION_ID`, `SUMMARY_THREAD_ID`. The orchestrator parses these from the output block and uses them in Step 7's Coordinator/Writer branching.
+- **Thread-fetch failure mapping (aligned with ADR 0016):** `401/403` → ABORTED (re-auth hint); `404` → OK (treat as "no threads yet" → first-review); `5xx` / network → DEGRADED with `kind: thread-fetch`, treat as empty threads → first-review. This follows ADR 0015's standard HTTP-tier mapping; no exemption.
+- `scripts/mode-detection.mjs` gains an exported constant `SIGNATURE_PREFIX = '🤖 *Reviewed by Claude Code*'`. The Fetcher imports it directly. No prompt-literal `SIGNATURE_PREFIX` input.
+- `agents/re-review-coordinator.md` lines 27 and 471 lose their references to `az repos pr thread list` as the data source — `RAW_THREADS_JSON` now flows from the Fetcher's output block via the orchestrator, not from a (broken) orchestrator-level call.
+
+**Hardening (spec 14 — Slice 02):**
+
+- A new inventory file enumerates every `az` command the plugin uses, including `--area`/`--resource` combinations. Initial entries: `az repos pr show`, `az repos pr checkout`, `az devops invoke --area git --resource {pullRequestThreads | pullRequestIterations | pullRequestIterationChanges | pullRequestWorkItems | pullRequestThreadComments}`, `az boards work-item show`.
+- A new allowlist file enumerates intentional exceptions to the inventory: `az --version`, `az extension list`, and error-message-only hints like `az devops login` and `az extension add`. Each entry is documented inline with the reason for exemption.
+- A new deep module `scripts/ado/cli-completeness.mjs` (following the `scripts/ado/` pattern PRD A established) exports `findUninventoriedCommands({ sources, inventory, allowlist }) → string[]`. Pure function. Handles multi-line bash blocks, backslash continuations, and `--area`/`--resource` split across lines.
+- A new smoke test has two cases:
+  1. **Inventory-completeness** — pure-JS, runs everywhere. Reads `agents/`, `commands/`, `scripts/` source files; calls `findUninventoriedCommands`; asserts the result is empty.
+  2. **CLI smoke** — integration test, requires `az` on PATH. Iterates the inventory; runs each entry's `--help` via `child_process.spawnSync` with a 5s timeout; asserts exit 0 and that `helpKeywordsRequired` substrings appear in stdout. `t.skip` when `az` is absent.
+- CI installs `azure-cli` + the `azure-devops` extension on a single matrix cell only (`ubuntu-latest × Node 24`), via a conditional workflow step. Other matrix cells skip the CLI smoke test cleanly.
+- The orchestrator's Step 3 preflight gains an assertion: `az devops invoke --help` must exit 0. On failure, exit with re-install instructions.
+- `AGENTS.md` gets a single pointer paragraph to the inventory file. No mirrored command table — the inventory is the only source of truth.
+
+## Testing Decisions
+
+Assert external behaviour against module contracts, not internal regex shape. Feed fixture markdown into `findUninventoriedCommands`, assert the returned array. Don't assert on internal helper functions or intermediate parser state.
+
+**Modules to test:**
+
+- `scripts/ado/cli-completeness.mjs` (new). Fixture-style cases: empty source; single inline `az` command; multi-line bash block with backslash continuation; `--area`/`--resource` split across lines; allowlist filtering; uninventoried command surfaces correctly; mixed allowlisted + inventoried + uninventoried.
+- `scripts/mode-detection.mjs`. Extend `tests/mode-detection.test.mjs` with one case asserting `SIGNATURE_PREFIX` equals the canonical literal `🤖 *Reviewed by Claude Code*`. Catches accidental edits to the constant.
+
+**Integration (not unit):**
+
+- The CLI smoke test (`tests/ado-cli-smoke.test.mjs`'s second case) is an integration test that requires `az`. It skips on machines/cells without the CLI installed.
+
+**Prior art:** existing tests follow the `node:test` ESM pattern with `// @ts-check`, fixture inputs, and `assert.deepStrictEqual`-style assertions. `tests/parse-diff-hunks.test.mjs` is the closest reference for the completeness parser tests; `tests/mode-detection.test.mjs` is the file to extend with the `SIGNATURE_PREFIX` assertion.
+
+## Out of Scope
+
+- The duplicate-comment bug from issue #46. Different root cause (ADO Writer's multi-script batching + spurious threadContext-rejection fallback). Independent investigation, independent fix.
+- Replacing `az` with direct REST calls. Considered and rejected: `az` handles token acquisition and corporate TLS-inspecting proxies that bare REST does not.
+- Invoking the global `azure-devops-cli` skill from the plugin. ADR 0008 keeps pr-review self-contained.
+- Validating ADO REST response shapes against schemas. If response drift becomes a real failure mode, that's future work.
+- Running the CLI smoke test on macOS/Windows in CI. The smoke test's purpose is catching Microsoft API changes, which surface identically across OSes — one Linux runner is sufficient.
+
+## Further Notes
+
+- **Severity.** Slice 01 is P0. Every ADO PR review currently runs through an improvised LLM workaround that bypasses the documented architecture. The user's most recent run produced Inline Comments anchored to wrong locations.
+- **Why combined PRD.** Slice 02's inventory-completeness check immediately validates Slice 01's new `pullRequestThreads` registration. Shipping them together is one CHANGELOG entry, one review cycle, and one patch bump.
+- **Related but independent.** Issue #46 (duplicate threads) was filed against the same plugin and looked like it might share a root cause. Investigation confirmed they are independent: the PR in issue #46 had two iterations (so "Iteration 2" in the signature was correct), and the duplication pattern matches an ADO Writer concern (multi-script restart + spurious fallback), not a Fetcher concern.
diff --git a/apps/claude-code/pr-review/docs/plans/00-preflight.md b/apps/claude-code/pr-review/docs/plans/00-preflight.md
deleted file mode 100644
index bff01f6f..00000000
--- a/apps/claude-code/pr-review/docs/plans/00-preflight.md
+++ /dev/null
@@ -1,97 +0,0 @@
-# 00. Pre-flight: ADR-0007 supersession + .prettierignore CHANGELOG guard
-
-**Status: done — 2026-05-06**
-
-- Priority: P0 (blocks every other spec in this feature)
-- Effort: XS
-- Version impact: none
-- Depends on: —
-- Touches: monorepo `.prettierignore`, `apps/claude-code/pr-review/docs/adr/0007-summary-rewritten-not-appended.md`,
-  new `apps/claude-code/pr-review/docs/adr/0009-summary-delta-as-reply.md`,
-  `apps/claude-code/pr-review/docs/adr/README.md`
-
-## Context
-
-Specs 01–09 implement the re-review feature. Two preconditions must be in place
-before any of them runs:
-
-1. ADR 0007 records the now-reversed "rewrite summary in place" decision. Spec 07
-   replies to the existing summary thread instead. Implementing 01–09 without first
-   superseding 0007 leaves the repo with two contradictory decisions of record.
-2. The monorepo-root `.prettierignore` does not exclude `**/CHANGELOG.md`. The
-   `verify:changelog` script enforces a literal em-dash (`## [X.Y.Z] — YYYY-MM-DD`);
-   any `pnpm format` run between now and spec 08 will reflow the em-dash and silently
-   break the next bump.
-
-## Current behaviour
-
-- ADR 0007: `**Status:** Accepted (2025-04)`, decision = rewrite-in-place.
-- `.prettierignore` (monorepo root): no entry for CHANGELOG files.
-
-## Target behaviour
-
-- ADR 0007: `**Status:** Superseded by 0009 (2026-05)`. Body untouched.
-- New `apps/claude-code/pr-review/docs/adr/0009-summary-delta-as-reply.md` created with status
-  `Accepted (2026-05) — Supersedes 0007`, recording:
-  - **Decision:** the re-review delta is posted as a reply to the existing summary
-    thread, identified via `isSummaryThread = true` (set by spec 02).
-  - **Fall-back:** if the prior summary thread is missing or deleted, post a fresh
-    full summary as a new thread — never attempt to edit an existing comment.
-  - **Consequences:** the PR shows one summary thread with reply entries across
-    iterations; edit-history of summary comments is no longer load-bearing.
-- ADR README updated: 0009 row added, 0007 row marked as superseded.
-- `.prettierignore` (monorepo root) gains `**/CHANGELOG.md` under a
-  `# Plugin changelogs — em-dash dates must not be reflowed` comment.
-
-## Implementation steps
-
-1. Add the following block to the monorepo-root `.prettierignore`, after the
-   existing sections:
-
-   ```
-   # Plugin changelogs — em-dash dates must not be reflowed
-   **/CHANGELOG.md
-   ```
-
-2. Create `apps/claude-code/pr-review/docs/adr/0009-summary-delta-as-reply.md`
-   using the ADR format. Key fields:
-   - **Status:** `Accepted (2026-05) — Supersedes 0007`
-   - **Context:** references ADR 0007; explains that the grilling session (2026-05)
-     reversed the "rewrite-in-place" decision in favour of a reply model.
-   - **Decision:** reply-to-existing-summary-thread + fall-back to new full summary
-     thread when the prior thread is missing (never edit).
-   - **Consequences:** as above.
-
-3. Edit `apps/claude-code/pr-review/docs/adr/0007-summary-rewritten-not-appended.md`:
-   change **only the Status line** from `Accepted (2025-04)` to
-   `Superseded by 0009 (2026-05)`. Leave Context, Decision, and Consequences
-   verbatim.
-
-4. Update `apps/claude-code/pr-review/docs/adr/README.md`: add the 0009 entry and
-   note 0007 as superseded.
-
-## Verification
-
-- `grep -F '**/CHANGELOG.md' .prettierignore` returns the new line.
-- `pnpm format` is a no-op on any `CHANGELOG.md`; `pnpm --filter pr-review verify:changelog` passes.
-- First line group of `apps/claude-code/pr-review/docs/adr/0007-…md` shows
-  `**Status:** Superseded by 0009 (2026-05)`.
-- `apps/claude-code/pr-review/docs/adr/0009-summary-delta-as-reply.md` exists and
-  references ADR 0007.
-- `apps/claude-code/pr-review/docs/adr/README.md` lists 0009.
-
-## Acceptance criteria
-
-- All four changes land in a single conventional commit:
-  `docs(pr-review): supersede ADR 0007 with 0009; guard CHANGELOGs from prettier`
-- No plugin version bump, no CHANGELOG entry (docs/policy only).
-- Specs 01–09 unblocked.
-
-## Out of scope
-
-- Any code change to `commands/review-pr.md` (specs 01–09 own that).
-- The `Re-review` README/CLAUDE.md updates (spec 08 owns those).
-
-## Follow-ups
-
-— none —
diff --git a/apps/claude-code/pr-review/docs/plans/01-normalize-signature.md b/apps/claude-code/pr-review/docs/plans/01-normalize-signature.md
deleted file mode 100644
index c0f9b07e..00000000
--- a/apps/claude-code/pr-review/docs/plans/01-normalize-signature.md
+++ /dev/null
@@ -1,66 +0,0 @@
-# 01. Normalize Claude Code signature
-
-**Status: done — 2026-05-06**
-
-- Priority: P0 (blocks detection)
-- Effort: XS
-- Version impact: patch
-- Depends on: 00
-- Touches: `commands/review-pr.md`
-
-## Context
-
-Re-review detection works by substring-matching the canonical signature prefix in existing PR threads. The current command emits `🤖 *Reviewed by Claude Code*` in some places and `🤖 _Reviewed by Claude Code_` in others (asterisk vs underscore italics). Markdown renders both identically, but detection must pick a single canonical prefix first.
-
-Additionally, every comment must now embed the iteration number so future re-reviews can determine which iteration each finding was raised on. The canonical form becomes `🤖 *Reviewed by Claude Code* — Iteration N` where N is the value of `LATEST_ITERATION_ID` at post time.
-
-## Current behaviour
-
-`commands/review-pr.md` writes the signature in multiple places. At least one location (the summary template markdown block, ~line 286) uses underscore italics; the rest use asterisks. No location includes an iteration number.
-
-## Target behaviour
-
-Every emitted signature is exactly `🤖 *Reviewed by Claude Code* — Iteration {N}` (asterisk italics, iteration suffix, single trailing newline before any extra metadata).
-
-Detection logic uses the invariant prefix `🤖 *Reviewed by Claude Code*` as a substring match, so it works against both old-format comments (no iteration suffix) and new-format comments.
-
-## Affected locations in `commands/review-pr.md`
-
-Two distinct types of location must be updated:
-
-1. **Runtime-emitted signatures** — inside JSON payloads written to `/tmp/pr_thread_N.json` and `/tmp/pr_summary.json` (Step 10 and Step 11). These must emit the full `— Iteration {LATEST_ITERATION_ID}` suffix.
-2. **Documentation/example blocks** — the summary structure markdown example (~line 283–287) that uses underscore italics. Must be updated to asterisk form. The example may show a placeholder like `— Iteration {N}` rather than a live variable.
-
-## Implementation steps
-
-1. Define two constants near the top of `commands/review-pr.md`:
-   - `SIGNATURE_PREFIX` = `🤖 *Reviewed by Claude Code*`
-   - `SIGNATURE` = `🤖 *Reviewed by Claude Code* — Iteration {LATEST_ITERATION_ID}` (resolved at post time)
-2. Replace every runtime-emitted signature with `SIGNATURE`.
-3. Replace the underscore-italics occurrence in the documentation example with the asterisk form (using a `{N}` placeholder).
-4. Add a note in the *Notes* section: "The detection prefix is `🤖 *Reviewed by Claude Code*` (substring match). The full emitted form is `🤖 *Reviewed by Claude Code* — Iteration N`. Never alter the prefix — re-review detection depends on it."
-
-## Test cases
-
-- `grep -nF '🤖 *Reviewed by Claude Code*' commands/review-pr.md` prints matches at every signature location.
-- `grep -nF '🤖 _Reviewed by Claude Code_' commands/review-pr.md` prints 0 matches.
-- All runtime-emitted signatures include `— Iteration` followed by a variable or placeholder.
-
-## Acceptance criteria
-
-- All runtime-emitted signatures are byte-identical in structure.
-- The Notes section documents both the prefix and the full form.
-- No underscore-italics signature remains anywhere in the file.
-
-## Verification
-
-- Run both greps above.
-- Read the diff: confirm only signature characters changed and every runtime location now includes the iteration suffix.
-
-## Out of scope
-
-- Detection logic itself (spec 02).
-
-## Follow-ups
-
-— none —
diff --git a/apps/claude-code/pr-review/docs/plans/02-detect-prior-review.md b/apps/claude-code/pr-review/docs/plans/02-detect-prior-review.md
deleted file mode 100644
index dc01c461..00000000
--- a/apps/claude-code/pr-review/docs/plans/02-detect-prior-review.md
+++ /dev/null
@@ -1,82 +0,0 @@
-# 02. Detect prior review on PR
-
-**Status: done — 2026-05-06**
-
-- Priority: P0
-- Effort: S
-- Version impact: minor
-- Depends on: 01
-- Touches: `commands/review-pr.md`
-
-## Context
-
-A re-review must first know whether Claude Code already reviewed this PR. The signal is the canonical signature prefix (spec 01) inside any thread comment on the PR. Detection must also identify the existing summary thread and parse the prior iteration number from the most recent bot comment, so later specs can act on that information.
-
-## Current behaviour
-
-The command never inspects existing threads before posting. Re-runs duplicate every comment.
-
-## Target behaviour
-
-After Step 3 (PR metadata) and before Step 4 (iteration), the command runs a detection step that:
-
-1. Calls `az devops invoke --area git --resource pullRequestThreads --route-parameters project={project} repositoryId={REPO_ID} pullRequestId={PR_ID} --org {ORG_URL} --api-version 7.1`.
-2. Follows pagination: if the response includes a `continuationToken`, repeat the call with `--query-parameters continuationToken={token}` until no token is returned. Collect all threads before proceeding.
-3. For each thread, checks every comment's `content` for the signature prefix `🤖 *Reviewed by Claude Code*` (substring match — no `createdBy` identity check required).
-4. For threads where a bot comment is found, captures:
-   - `threadId`
-   - `filePath` (from `threadContext.filePath`, or `null` for general threads)
-   - `start: { line, offset }` (from `threadContext.rightFileStart`)
-   - `end: { line, offset }` (from `threadContext.rightFileEnd`)
-   - `comments` (full array)
-   - `status` (ADO thread status integer)
-   - `isSummaryThread` — `true` when `filePath` is `null` and the first comment begins with the prefix `## PR Review Summary` (substring match — the actual heading may include a ` — {PR_TITLE}` suffix)
-5. Identifies the most recent bot comment across all prior threads and parses `PRIOR_ITERATION_ID` from its signature suffix (`— Iteration N`). When the suffix is absent (legacy format), sets `PRIOR_ITERATION_ID=null` — spec 03 resolves the actual ID via timestamp comparison against the iterations API.
-6. Sets `IS_REREVIEW=true` and records `PRIOR_THREADS` and `SUMMARY_THREAD_ID`.
-7. Logs a one-line summary: `Detected N prior Claude Code threads — re-review mode ON` (or `…OFF`).
-
-## Edge cases
-
-- Thread `status` may be `closed`/`fixed` — keep in `PRIOR_THREADS` with status; downstream specs decide how to act.
-- General threads (no `threadContext`) have `filePath = null`, `start = null`, `end = null`. The summary thread is the general thread with `isSummaryThread = true`.
-- A thread can have multiple comments; scan all of them for the signature prefix to confirm bot authorship.
-- A human who quotes the signature string triggers a false positive — this is accepted and documented in Notes.
-- Threads from a prior run that was interrupted (no completion marker) are included in `PRIOR_THREADS` normally; spec 06 detects the partial-run state from the missing completion marker.
-
-## Implementation steps
-
-1. Insert a new section "Step 3.5: Detect prior review" with the paginated API call and parsing logic.
-2. Write thread JSON to a temp file under `$TMPDIR`; parse with `jq`.
-3. Export `IS_REREVIEW`, `PRIOR_THREADS_FILE` (path to the jq-readable JSON file — downstream steps consume it as `jq ... < "$PRIOR_THREADS_FILE"`), `SUMMARY_THREAD_ID`, and `PRIOR_ITERATION_ID` for use by later steps. Within spec descriptions, `PRIOR_THREADS` refers to the in-memory thread collection; `PRIOR_THREADS_FILE` is the on-disk path used by the command layer.
-
-## Test cases
-
-- Against a PR with no prior bot comments: `IS_REREVIEW=false`, `PRIOR_THREADS` empty.
-- Against a PR with prior bot comments across two pages (>100 threads): all threads collected, count matches total across both pages.
-- Against a PR with a summary thread: `SUMMARY_THREAD_ID` set, `isSummaryThread=true` on that entry.
-- Against a PR where a human quoted the signature: detected as prior review (accepted false positive).
-- Against a PR with a multi-line inline thread: `start.line != end.line`, offsets preserved.
-
-## Acceptance criteria
-
-- Step 3.5 runs unconditionally and prints the summary line.
-- All threads are fetched before proceeding (pagination loop complete).
-- No write actions occur during detection.
-- `isSummaryThread` is set on exactly one thread (or zero if no prior summary exists).
-
-## Verification
-
-- Dry-run on a PR with no prior review: confirm OFF path, command continues unchanged.
-- Dry-run on a PR with known prior threads: confirm ON path, thread count matches ADO UI.
-
-## Out of scope
-
-- Acting on the detection (specs 03–07).
-
-## Notes
-
-Detection is signature-prefix only. There is no `createdBy` bot-identity check. This makes detection PAT-agnostic: it works correctly regardless of which team member's credentials were used to post the original review.
-
-## Follow-ups
-
-— none —
diff --git a/apps/claude-code/pr-review/docs/plans/03-target-latest-iteration.md b/apps/claude-code/pr-review/docs/plans/03-target-latest-iteration.md
deleted file mode 100644
index f7521e9b..00000000
--- a/apps/claude-code/pr-review/docs/plans/03-target-latest-iteration.md
+++ /dev/null
@@ -1,71 +0,0 @@
-# 03. Target latest PR iteration
-
-**Status: done — 2026-05-06**
-
-- Priority: P0
-- Effort: S
-- Version impact: minor
-- Depends on: 02
-- Touches: `commands/review-pr.md`, `CLAUDE.md`
-
-## Context
-
-Re-reviews must reason about the latest pushed commits, not the initial iteration. The command currently hardcodes `iterationId=1` in two places — Step 4 (list changed files) and Step 5 (local diff) — and `CLAUDE.md` codifies that as a rule. Both must be updated.
-
-## Current behaviour
-
-`pullRequestIterationChanges` is fetched with `iterationId=1`, missing every change introduced after the first push. The local diff in Step 5 uses `git diff origin/{target}...HEAD` which implicitly covers all commits but does not track iteration boundaries.
-
-## Target behaviour
-
-1. Call `az devops invoke --area git --resource pullRequestIterations …` and pick the highest `id` as `LATEST_ITERATION_ID`.
-2. Use `LATEST_ITERATION_ID` for the `pullRequestIterationChanges` call in Step 4 (file list).
-3. Capture `sourceRefCommit.commitId` for `LATEST_ITERATION_ID` as `LATEST_COMMIT_ID`.
-4. When `IS_REREVIEW=true`:
-   - Use `PRIOR_ITERATION_ID` (from spec 02: parsed from signature suffix, with timestamp fallback for legacy comments) to look up that iteration's `sourceRefCommit.commitId` as `PRIOR_COMMIT_ID`.
-   - Export both `LATEST_COMMIT_ID` and `PRIOR_COMMIT_ID` for spec 04.
-5. In `CLAUDE.md`, replace the bullet beginning with `` `iterationId=1` is always used `` with: "Always use the latest iteration of the PR. `iterationId=1` is never used. Re-reviews additionally compute `PRIOR_ITERATION_ID` from the prior review's signature — see spec 02."
-
-## Timestamp fallback for legacy `PRIOR_ITERATION_ID`
-
-When the prior review's newest comment lacks the `— Iteration N` suffix (legacy format), derive `PRIOR_ITERATION_ID` as follows: fetch all iterations, find the max `createdDate` across all comments in `PRIOR_THREADS`, then select the iteration whose `createdDate` is the highest value still ≤ that timestamp.
-
-## Edge cases
-
-- A brand-new PR has only iteration 1 — this is valid; `LATEST_ITERATION_ID=1` and no fallback needed.
-- `pullRequestIterations` returns iterations unsorted — use `jq 'max_by(.id) | .id'` not array index.
-- `sourceRefCommit.commitId` may be absent on a very new PR before the first push completes — treat as null and fall back to full diff in spec 04.
-
-## Implementation steps
-
-1. Replace the hardcoded `iterationId=1` block in Step 4 with a fetch + `jq 'max_by(.id)'` lookup.
-2. Capture `sourceRefCommit.commitId` from the iterations response.
-3. Add the fallback: if zero iterations returned, default to `LATEST_ITERATION_ID=1` with a warning log.
-4. When `IS_REREVIEW=true`, resolve `PRIOR_COMMIT_ID` from `PRIOR_ITERATION_ID`.
-5. Update `CLAUDE.md` rule.
-
-## Test cases
-
-- Single-iteration PR: `LATEST_ITERATION_ID=1`, behaviour unchanged from before.
-- Multi-iteration PR: `LATEST_ITERATION_ID` equals the highest push ID (verify via ADO UI).
-- Re-review with iteration suffix in prior signature: `PRIOR_ITERATION_ID` parsed directly, no API cross-join.
-- Re-review with legacy prior comment (no suffix): `PRIOR_ITERATION_ID` derived from timestamp comparison.
-
-## Acceptance criteria
-
-- No `iterationId=1` literals remain in `commands/review-pr.md`.
-- Step 4 file list and Step 5 diff use the same iteration boundary.
-- `CLAUDE.md` rule updated.
-
-## Verification
-
-- `grep -n 'iterationId=1' commands/review-pr.md` → 0 matches.
-- Re-run the command on a PR with ≥ 2 iterations and confirm Step 4 logs the latest iteration ID.
-
-## Out of scope
-
-- Computing the diff baseline (spec 04).
-
-## Follow-ups
-
-— none —
diff --git a/apps/claude-code/pr-review/docs/plans/04-incremental-diff-baseline.md b/apps/claude-code/pr-review/docs/plans/04-incremental-diff-baseline.md
deleted file mode 100644
index b05c6847..00000000
--- a/apps/claude-code/pr-review/docs/plans/04-incremental-diff-baseline.md
+++ /dev/null
@@ -1,66 +0,0 @@
-# 04. Incremental diff baseline on re-review
-
-**Status: done — 2026-05-06**
-
-- Priority: P1
-- Effort: M
-- Version impact: minor
-- Depends on: 03
-- Touches: `commands/review-pr.md`
-
-## Context
-
-Step 5 currently runs `git diff origin/{target}...HEAD`, which on re-review shows the entire branch — including code already reviewed. The reviewer wastes tokens re-analysing untouched lines.
-
-## Current behaviour
-
-Full branch diff regardless of `IS_REREVIEW`.
-
-## Target behaviour
-
-- When `IS_REREVIEW=false`: keep existing full-branch diff.
-- When `IS_REREVIEW=true`: diff between `PRIOR_COMMIT_ID` and `LATEST_COMMIT_ID` (both provided by spec 03 via `sourceRefCommit.commitId`).
-- If the two commit IDs are identical (no new pushes since prior review), exit Step 5 early:
-  1. Print `No new commits since last review.`
-  2. Print a list of all `pending` threads from `PRIOR_THREADS` (file path and line range for each), so the user knows what is still outstanding.
-  3. Exit cleanly — do not proceed to Steps 6–11.
-- If `PRIOR_COMMIT_ID` is null (legacy comment with no parseable iteration), fall back to full diff with a warning: `Warning: could not resolve prior commit — falling back to full diff.`
-
-## Edge cases
-
-- Force-push rewrites history; `PRIOR_COMMIT_ID` may no longer exist locally. Attempt `git fetch origin {PRIOR_COMMIT_ID}` before diffing; if the fetch fails, fall back to full diff with a warning that includes both commit IDs: `Warning: prior commit {PRIOR_COMMIT_ID} unreachable; latest commit {LATEST_COMMIT_ID} — falling back to full diff.`
-- Files renamed between iterations: rely on `git diff -M` (already default) so renames map correctly.
-- `LATEST_COMMIT_ID` is null (spec 03 edge case): fall back to full diff with a warning.
-
-## Implementation steps
-
-1. Branch Step 5 on `IS_REREVIEW`.
-2. Add the early-exit path: print "No new commits" message, list pending threads, exit cleanly.
-3. Add the `git fetch` attempt with the detailed fallback warning message.
-4. The diff hunk output from this step is consumed by spec 05 for thread classification — ensure hunk boundaries (file path, start line, end line) are exported in a structured format (JSON or line-delimited) to `$TMPDIR`.
-
-## Test cases
-
-- Re-review with identical commit IDs: early-exit path fires, pending thread list printed to console, no ADO comments posted.
-- Re-review with one new commit: diff contains exactly that commit's changes.
-- Re-review after a force-push that rebased the branch: fallback warning fires with both commit IDs; full diff is used.
-- Re-review with `PRIOR_COMMIT_ID=null` (legacy): fallback warning fires; full diff used.
-
-## Acceptance criteria
-
-- First-time review behaviour unchanged.
-- Early exit lists all pending threads before stopping.
-- Fallback warnings always include both commit IDs when available.
-
-## Verification
-
-- Run the command on a PR with no new commits since last review — confirm early exit fires and pending threads are listed.
-- Run the command on a PR with one new commit — confirm diff output matches `git diff {PRIOR_COMMIT_ID}..{LATEST_COMMIT_ID}`.
-
-## Out of scope
-
-- Thread classification (spec 05).
-
-## Follow-ups
-
-— none —
diff --git a/apps/claude-code/pr-review/docs/plans/05-classify-existing-threads.md b/apps/claude-code/pr-review/docs/plans/05-classify-existing-threads.md
deleted file mode 100644
index 6d84b608..00000000
--- a/apps/claude-code/pr-review/docs/plans/05-classify-existing-threads.md
+++ /dev/null
@@ -1,78 +0,0 @@
-# 05. Classify existing threads
-
-**Status: done — 2026-05-06**
-
-- Priority: P1
-- Effort: M
-- Version impact: minor
-- Depends on: 02
-- Touches: `commands/review-pr.md`
-
-## Context
-
-Each prior bot thread must be tagged so Step 10 knows whether to reply, skip, or auto-resolve. Classification uses ADO thread status as the primary signal and diff-hunk intersection as a secondary signal. No comment-text analysis is performed.
-
-## Current behaviour
-
-Threads are not classified.
-
-## Target behaviour
-
-For every thread in `PRIOR_THREADS`, compute one of:
-
-- `addressed` — ADO thread `status` is one of `fixed` (2), `wontFix` (3), `closed` (4), or `byDesign` (5), **OR** the thread `status` is `active` (1) or `pending` (6) and the thread's line range intersects a changed hunk in the incremental diff (spec 04). ADO `pending` (6) is treated the same as `active` — it is not auto-addressed; diff intersection is still required.
-- `disputed` — `status` is `active` AND at least one comment in the thread does not contain the signature prefix `🤖 *Reviewed by Claude Code*`.
-- `pending` — `status` is `active` AND no comment in the thread lacks the signature prefix (i.e. only bot comments present).
-- `obsolete` — the thread's `filePath` does not appear in the PR diff at all (or `filePath` is null and the thread is not the summary thread).
-
-ADO status codes: 1 = active, 2 = fixed, 3 = wontFix, 4 = closed, 5 = byDesign, 6 = pending.
-
-**Line-range intersection** (for the secondary `addressed` signal): a thread is considered intersecting a hunk when `max(thread.start.line, hunk.startLine) ≤ min(thread.end.line, hunk.endLine)`. Use line numbers only; offsets are not used in intersection logic.
-
-**Human reply detection** (for `disputed`): a comment is human-authored when its `content` does not contain the substring `🤖 *Reviewed by Claude Code*`. No `createdBy` identity check is performed — this makes classification PAT-agnostic.
-
-## Edge cases
-
-- General threads (`filePath = null`) that are not the summary thread: `addressed` and `obsolete` do not apply. Classify as `disputed` or `pending` only.
-- The summary thread (`isSummaryThread = true`): skip classification entirely; it is handled by spec 07.
-- Multi-line threads: intersection check uses the full `[start.line, end.line]` range.
-- Threads where the entire file was deleted from the PR: `obsolete`.
-
-## Implementation steps
-
-1. Extract diff hunk boundaries from Step 5 output (file path, start line, end line per hunk).
-2. For each `PRIOR_THREADS` entry (skipping the summary thread), apply the classification logic above.
-3. Store results in `PRIOR_THREADS` under a new `classification` field.
-4. Print a one-line summary count: `Threads: N addressed, N disputed, N pending, N obsolete`.
-
-## Test cases
-
-- Thread whose line range intersects a changed hunk (status active) → `addressed`.
-- Thread with ADO status `fixed` (no diff intersection needed) → `addressed`.
-- Thread with a human reply (no signature prefix in that reply) → `disputed`.
-- Thread with only bot comments and no diff intersection → `pending`.
-- Thread on a file not in the diff → `obsolete`.
-- Thread spanning lines 10–15 with a hunk at lines 12–13 → `addressed`.
-- General thread (no file) with no human reply → `pending`.
-
-## Acceptance criteria
-
-- Every non-summary thread receives exactly one classification.
-- Summary thread is skipped.
-- Summary count line printed before Step 6.
-
-## Verification
-
-- Run the command against a PR where one thread was addressed, one disputed, one pending, and one is on a deleted file — confirm all four classifications appear in the summary count.
-
-## Out of scope
-
-- Posting replies (spec 06).
-
-## Notes
-
-The `disputed` reply template (spec 06) will remind the author to mark the thread resolved in ADO when they consider the conversation done.
-
-## Follow-ups
-
-— none —
diff --git a/apps/claude-code/pr-review/docs/plans/06-reply-to-threads.md b/apps/claude-code/pr-review/docs/plans/06-reply-to-threads.md
deleted file mode 100644
index ce3a8c5c..00000000
--- a/apps/claude-code/pr-review/docs/plans/06-reply-to-threads.md
+++ /dev/null
@@ -1,99 +0,0 @@
-# 06. Reply to threads instead of duplicating
-
-**Status: done — 2026-05-06**
-
-- Priority: P0
-- Effort: M
-- Version impact: minor
-- Depends on: 04, 05
-- Touches: `commands/review-pr.md`
-
-## Context
-
-Step 10 posts inline comments via `pullRequestThreads` (creates a new thread). On re-review we must reuse existing threads: reply via `pullRequestThreadComments` and PATCH thread status via `pullRequestThreads` when resolving.
-
-## Current behaviour
-
-Every finding becomes a fresh thread regardless of whether one already exists at that file/line range.
-
-## Target behaviour
-
-### Thread matching
-
-For each finding, attempt to match a prior thread by:
-
-1. `filePath` equality (exact match).
-2. Line-range overlap: `max(finding.start.line, thread.start.line) ≤ min(finding.end.line, thread.end.line)`, with a ±3 line drift tolerance applied to both endpoints (i.e. expand each range by 3 lines before testing overlap).
-
-If no match is found, the finding is new — create a fresh thread as in the current flow.
-
-### Reply actions per classification
-
-| Classification | Action |
-|---|---|
-| `pending` (unchanged finding) | **Skip** — do not post. |
-| `pending` (new evidence in finding) | **Reply** via `pullRequestThreadComments` with only the new evidence. |
-| `disputed` | **Reply** acknowledging the author's point; never re-assert without new evidence. Include: *"If you consider this resolved, please mark the thread as fixed in Azure DevOps."* |
-| `addressed` | **Reply** with `Resolved as of Iteration N — thanks!` then **PATCH** thread status to `fixed` via `pullRequestThreads`. |
-| `obsolete` | Leave alone — no action. |
-
-### ADO API resources
-
-- **Reply content** (add a comment to an existing thread): POST to `--resource pullRequestThreadComments` with `--route-parameters … threadId={id}`.
-- **Thread status PATCH** (mark as fixed): PATCH to `--resource pullRequestThreads` with `--route-parameters … threadId={id}`. Body: `{ "status": 2 }`. ADO status codes: 1 = active, 2 = fixed, 3 = wontFix, 4 = closed, 5 = byDesign, 6 = pending.
-- If the PATCH returns a 409 (concurrent resolution by human), log and continue.
-
-### New findings on re-review
-
-Fresh threads (no prior match) are created as in the current flow. The first comment body does **not** include a separate "Iteration N" header line — the iteration is carried by the signature suffix: `🤖 *Reviewed by Claude Code* — Iteration {LATEST_ITERATION_ID}`.
-
-### Run completion marker
-
-After all threads and replies are posted, post one final reply to the summary thread:
-
-```
-✅ Review complete — Iteration {LATEST_ITERATION_ID} ({N} findings posted)
-```
-
-This is the last action of every successful run (first review or re-review). Its absence for `LATEST_ITERATION_ID` signals a partial prior run; on the next run, treat the current iteration as a first-time review.
-
-## Edge cases
-
-- **No reply cap.** Post all replies regardless of count.
-- **`pending` general threads** (not the summary thread, no file): skip — same rule as inline `pending`.
-- **Concurrency:** PATCH to `status=fixed` may 409 if author resolved the thread mid-run — log and continue.
-- **Partial prior run detected** (no completion marker for `LATEST_ITERATION_ID`): treat as first-review mode; skip thread matching for this iteration.
-
-## Implementation steps
-
-1. Add the matching function (file path equality + range overlap with ±3 line drift).
-2. Branch Step 10 on `IS_REREVIEW`.
-3. Wire the five classification branches (pending-skip, pending-reply, disputed, addressed, obsolete).
-4. Add the completion marker reply as the final action after Step 11 (delta summary reply) completes — the posting order is: inline thread replies (Step 10) → delta summary reply (Step 11) → completion marker.
-
-## Test cases
-
-- Re-review with no new commits: spec 04 exits early before Step 10 is reached.
-- Re-review where author fixed one finding (ADO status fixed): one `addressed` reply posted, thread patched to fixed.
-- Re-review where author replied to one finding: one `disputed` acknowledgement with ADO nudge.
-- Re-review with one new commit introducing a new issue: one fresh thread posted with signature suffix.
-- Re-review with no partial-run marker: treated as first-review mode for this iteration.
-- Multi-line thread at lines 10–15 matched by finding at lines 12–13: match found, range overlap confirmed.
-
-## Acceptance criteria
-
-- No duplicate thread is ever created when a matching prior thread exists.
-- All replies carry the canonical signature on their last line.
-- Completion marker is the final comment posted on every successful run.
-
-## Verification
-
-- Inspect a PR after a fresh re-run — comment count must not have grown unless a new commit is present or an addressed/disputed reply was warranted.
-
-## Out of scope
-
-- Summary comment behaviour (spec 07).
-
-## Follow-ups
-
-— none —
diff --git a/apps/claude-code/pr-review/docs/plans/07-summary-comment-policy.md b/apps/claude-code/pr-review/docs/plans/07-summary-comment-policy.md
deleted file mode 100644
index c8ac5a04..00000000
--- a/apps/claude-code/pr-review/docs/plans/07-summary-comment-policy.md
+++ /dev/null
@@ -1,63 +0,0 @@
-# 07. Summary comment policy on re-review
-
-**Status: done — 2026-05-06**
-
-- Priority: P1
-- Effort: S
-- Version impact: minor
-- Depends on: 06
-- Touches: `commands/review-pr.md`
-
-## Context
-
-Step 11 posts a long summary comment. On re-review we do not want a second one. This spec encodes explicit user feedback: *"Refrain from publishing another general comment, only inline or file comments and only if the findings aren't yet commented."*
-
-The existing summary thread is identified by the `isSummaryThread = true` flag set in spec 02. The delta comment is posted as a **reply to that thread**, not as a new general thread. This keeps the PR conversation to a single summary entry.
-
-## Current behaviour
-
-Summary always posted at the end of the run.
-
-## Target behaviour
-
-- `IS_REREVIEW=false`: behaviour unchanged — post full summary as the first comment in a new thread.
-- `IS_REREVIEW=true` and at least one of {new threads created, addressed/disputed replies posted}: post a reply to the existing summary thread titled `🤖 *Reviewed by Claude Code* — Re-review delta (Iteration {N})` containing only:
-  - Counts: `X new findings, Y resolved, Z disputed, W pending`.
-  - Bullet list of new finding titles only (each linked to its thread).
-  - No prose, no section headings beyond the title.
-- `IS_REREVIEW=true` and nothing changed (early-exit from spec 04, or all matches were `pending` and unchanged): post **nothing** at the summary level.
-
-If the prior summary thread is not found (e.g. it was deleted by a human), fall back to first-review mode: post a full summary as a new thread.
-
-## Implementation steps
-
-1. Track counters during Step 10: new threads, addressed replies, disputed replies.
-2. Feed counters to Step 11.
-3. Branch Step 11 on `IS_REREVIEW` and the counters.
-4. When posting the delta, use `SUMMARY_THREAD_ID` from spec 02 to reply via `pullRequestThreadComments`.
-5. Update Step 11's signature line to use the canonical form from spec 01.
-
-## Test cases
-
-- First-time review: full summary posted in a new thread (regression check).
-- Re-review with zero changes: no summary or delta posted.
-- Re-review with one new commit and one resolved finding: delta reply posted to existing summary thread with `1 new, 1 resolved, 0 disputed`.
-- Prior summary thread deleted: falls back to full summary in a new thread.
-
-## Acceptance criteria
-
-- Re-review never produces a second full-length summary.
-- Delta reply, when posted, contains only counts and a bullet list — no prose sections.
-- Delta is a reply to the existing summary thread, not a new general thread.
-
-## Verification
-
-- Run a re-review against a PR with an existing summary thread — confirm the thread gains one reply (the delta) and no new general thread was created.
-
-## Out of scope
-
-- Versioning and docs (spec 08).
-
-## Follow-ups
-
-— none —
diff --git a/apps/claude-code/pr-review/docs/plans/08-version-bump-and-docs.md b/apps/claude-code/pr-review/docs/plans/08-version-bump-and-docs.md
deleted file mode 100644
index 2d97d76e..00000000
--- a/apps/claude-code/pr-review/docs/plans/08-version-bump-and-docs.md
+++ /dev/null
@@ -1,74 +0,0 @@
-# 08. Version bump, README, CLAUDE.md
-
-**Status: done — 2026-05-06**
-
-- Priority: P1
-- Effort: S
-- Version impact: minor (cumulative roll-up)
-- Depends on: 07
-- Touches: `.claude-plugin/plugin.json`, `.claude-plugin/marketplace.json`, `README.md`, `CLAUDE.md`, `CHANGELOG.md`
-
-## Context
-
-Specs 01–07 ship behaviour; this spec finalises metadata, versioning, and documentation so the release is coherent. Spec 09 ships the test harness in the same release.
-
-## Implementation steps
-
-1. **Add CHANGELOG entries first.** Before running the bump script, add bullets under `## [Unreleased]` in `CHANGELOG.md` (it already exists). Add entries under the appropriate subsections (`### Added`, `### Fixed`, `### Breaking`). The bump script fails verification if the versioned section contains only `(none)` placeholders.
-
-2. **Bump the version.** Run `pnpm --filter pr-review bump minor` (target: `0.X.0 → 0.(X+1).0`). This script updates both `.claude-plugin/plugin.json` and `marketplace.json` atomically. **Never hand-edit `marketplace.json`.**
-
-3. **Verify `.prettierignore`.** Confirm `**/CHANGELOG.md` is present in the monorepo-root `.prettierignore` (added by spec 00). Run `grep -F '**/CHANGELOG.md' .prettierignore` — must return a match. No action required; this is a pre-flight guard check.
-
-4. **Update `CLAUDE.md`.** Remove the roadmap line: *"Re-review: detect existing Claude Code threads and update instead of duplicating."* Also update line 37 (`All comments posted to ADO **must** end with the exact signature: …`) to reflect the new iteration-suffixed form: `---\n🤖 *Reviewed by Claude Code* — Iteration N`. The iteration-targeting rule (added by spec 03) should already be present.
-
-5. **Add a "Re-review" section to `README.md`:**
-   - Trigger: re-running `/unic-pr-review:review-pr` on a PR that already has Claude Code threads.
-   - What changes: detection, thread reuse, delta summary reply, completion marker.
-   - New signature format: `🤖 *Reviewed by Claude Code* — Iteration N`.
-   - Limitations: force-push fallback to full diff; partial-run recovery via missing completion marker.
-
-## CHANGELOG format (enforced by `verify:changelog`)
-
-```markdown
-## [Unreleased]
-
-### Breaking
-- (none)
-
-### Added
-- Re-review: detect prior Claude Code threads and reply instead of duplicating
-- …
-
-### Fixed
-- (none)
-```
-
-Versioned sections use an em-dash: `## [0.2.0] — YYYY-MM-DD`.
-
-## Test cases
-
-- `cat .claude-plugin/plugin.json | jq -r .version` → bumped minor version.
-- `pnpm --filter pr-review verify:changelog` → passes.
-- `grep -n 'Re-review' CLAUDE.md` → roadmap line gone; rule lines (signature, iteration) present.
-- `README.md` includes the new Re-review section.
-- `grep -F '**/CHANGELOG.md' .prettierignore` returns a match (guard from spec 00).
-
-## Acceptance criteria
-
-- All files updated in a single commit.
-- Versions match across `plugin.json` and `marketplace.json` (guaranteed by bump script).
-- CI `verify:changelog` passes on the PR.
-
-## Verification
-
-- Run `pnpm --filter pr-review verify:changelog` locally before opening the PR.
-- Open the plugin in Claude Code — `/help` shows the new version.
-
-## Out of scope
-
-- Marketplace publishing (separate manual step).
-
-## Follow-ups
-
-— none —
diff --git a/apps/claude-code/pr-review/docs/plans/09-test-harness.md b/apps/claude-code/pr-review/docs/plans/09-test-harness.md
deleted file mode 100644
index 645b782e..00000000
--- a/apps/claude-code/pr-review/docs/plans/09-test-harness.md
+++ /dev/null
@@ -1,146 +0,0 @@
-# 09. Test harness — node:test + extracted modules
-
-**Status: done — 2026-05-06**
-
-- Priority: P1
-- Effort: M
-- Version impact: none (test infrastructure only)
-- Depends on: 02, 05, 06
-- Touches: `scripts/re-review/`, `tests/`, `commands/review-pr.md`, `package.json`
-
-## Context
-
-The re-review feature (specs 01–07) contains complex logic for thread detection, classification, and matching. Without an automated test harness, regressions can only be caught during a live review run. This spec extracts the core logic into importable Node.js modules and provides a `node:test` suite with JSON fixtures — no live ADO connection required.
-
-All prior references to specific production PRs as test targets are replaced by named fixture scenarios. Production PRs must never be used as test dependencies.
-
-## Modules to extract
-
-Each module lives under `scripts/re-review/` and exposes a single exported function with a JSON-in / JSON-out contract. The `commands/review-pr.md` command calls these via Bash, piping JSON.
-
-### `parse-signature.mjs`
-
-```
-input:  string (comment body)
-output: { iterationId: number } | null
-```
-
-Extracts the iteration number from the canonical signature suffix `— Iteration N`. Returns null for legacy comments or non-bot comments.
-
-### `classify-thread.mjs`
-
-```
-input:  { thread: PriorThread, diffHunks: DiffHunk[], signaturePrefix: string }
-output: 'addressed' | 'disputed' | 'pending' | 'obsolete'
-```
-
-Applies the four-state classification rules (spec 05). Pure function — no I/O.
-
-### `match-finding.mjs`
-
-```
-input:  { finding: Finding, priorThreads: PriorThread[], driftLines?: number }
-output: PriorThread | null
-```
-
-Returns the best-matching prior thread using file path equality and line-range overlap with ±`driftLines` (default 3) tolerance. Pure function — no I/O.
-
-### `detect-prior-review.mjs`
-
-```
-input:  { threads: RawADOThread[], signaturePrefix: string }
-output: { isRereview: boolean, priorThreads: PriorThread[], summaryThread: PriorThread | null, priorIterationId: number | null }
-```
-
-Processes the full paginated thread list from ADO, identifies bot threads, tags the summary thread, parses the prior iteration ID. Does not call the ADO API — receives already-fetched thread data.
-
-## Fixture scenarios
-
-Fixtures live under `tests/fixtures/` as static JSON files. Each filename maps to a scenario name used in test descriptions.
-
-| Fixture file | Scenario |
-|---|---|
-| `threads-fresh-pr.json` | No prior bot threads — first-time review |
-| `threads-pending.json` | Bot threads present, no human replies, status active |
-| `threads-disputed.json` | Bot threads with human replies |
-| `threads-addressed-status.json` | Bot threads with ADO status `fixed` |
-| `threads-addressed-diff.json` | Bot threads with active status; line range in diff hunk |
-| `threads-obsolete.json` | Bot threads on a file not in the diff |
-| `threads-partial-run.json` | Bot threads present, no completion marker for current iteration |
-| `threads-paginated-p1.json` | First page (100 threads, includes continuationToken) |
-| `threads-paginated-p2.json` | Second page (remaining threads, no continuationToken) |
-| `diff-hunks-no-change.json` | No hunks (identical commits) |
-| `diff-hunks-with-changes.json` | Hunks covering lines 40–45 in one file |
-
-Fixture JSON shapes match the ADO `pullRequestThreads` API response format (including `threadContext`, `comments`, `status`, `id`).
-
-## Test files
-
-Each module has a corresponding test file under `tests/`:
-
-- `tests/parse-signature.test.mjs`
-- `tests/classify-thread.test.mjs`
-- `tests/match-finding.test.mjs`
-- `tests/detect-prior-review.test.mjs`
-
-## Representative test cases
-
-### parse-signature
-
-- Current format `🤖 *Reviewed by Claude Code* — Iteration 3` → `{ iterationId: 3 }`.
-- Legacy format `🤖 *Reviewed by Claude Code*` (no suffix) → `null`.
-- Human comment with no signature → `null`.
-
-### classify-thread
-
-- Thread with ADO status `fixed` (no diff needed) → `addressed`.
-- Thread with active status and line range intersecting a diff hunk → `addressed`.
-- Thread with active status, human reply present → `disputed`.
-- Thread with active status, no human replies, line not in diff → `pending`.
-- Thread on a file absent from diff → `obsolete`.
-- Multi-line thread (lines 10–15) with hunk at lines 12–13 → `addressed`.
-- General thread (filePath null) with human reply → `disputed`.
-- General thread (filePath null) with no human replies → `pending`.
-
-### match-finding
-
-- Finding at file A line 42 with prior thread at file A lines 42–42 → match.
-- Finding at file A line 44 with prior thread at file A lines 42–42 (within ±3 drift) → match.
-- Finding at file A line 50 with prior thread at file A lines 42–42 (outside drift) → no match.
-- Finding at file B with prior thread at file A → no match.
-- Multi-line finding at lines 10–15 with prior thread at lines 12–20 (overlapping) → match.
-
-### detect-prior-review
-
-- Fresh PR fixture → `isRereview=false`, `priorThreads=[]`.
-- Pending threads fixture → `isRereview=true`, count matches, `isSummaryThread` set.
-- Paginated fixture (p1 + p2 combined) → all threads collected.
-- Partial-run fixture (no completion marker) → `isRereview=true`, downstream caller handles partial-run logic.
-
-## Implementation steps
-
-1. Create `scripts/re-review/` directory; add the four `.mjs` modules.
-2. Create `tests/fixtures/` and author the fixture JSON files.
-3. Write the four test files using `node:test` and `node:assert/strict`.
-4. Add `"test": "node --test tests/parse-signature.test.mjs tests/classify-thread.test.mjs tests/match-finding.test.mjs tests/detect-prior-review.test.mjs"` to `package.json` scripts. Use explicit paths rather than a glob to avoid cross-platform globstar issues (consistent with the pattern in `packages/release-tools/package.json`).
-5. Update `commands/review-pr.md` to call the modules via Bash at the appropriate steps.
-
-## Acceptance criteria
-
-- `pnpm --filter pr-review test` passes with zero failures.
-- All four modules are exercised by at least one test per classification/scenario.
-- No test imports `az` or makes network calls.
-
-## Verification
-
-- Run `pnpm --filter pr-review test` from the monorepo root — all tests green.
-- Delete one fixture file and confirm the corresponding test fails with a clear message.
-
-## Out of scope
-
-- End-to-end tests against a live ADO organisation.
-- Testing the `commands/review-pr.md` Markdown orchestration layer directly.
-
-## Follow-ups
-
-— none —
diff --git a/apps/claude-code/pr-review/docs/plans/10-doc-context-enrichment.md b/apps/claude-code/pr-review/docs/plans/10-doc-context-enrichment.md
deleted file mode 100644
index 974f8bc6..00000000
--- a/apps/claude-code/pr-review/docs/plans/10-doc-context-enrichment.md
+++ /dev/null
@@ -1,257 +0,0 @@
-# 10. Doc Context Enrichment — work items + Confluence pages
-
-**Status: done — 2026-05-06**
-
-- Priority: P1
-- Effort: M
-- Version impact: minor (new user-visible capability; no breaking changes)
-- Depends on: —
-- Touches: `commands/review-pr.md`, new `scripts/confluence-client.mjs`,
-  `docs/adr/0010-inline-confluence-client.md` (already written),
-  `CONTEXT.md` (already updated), `docs/plans/README.md`, `CHANGELOG.md`
-
-## Context
-
-Review agents currently receive only the diff and changed file contents. They have no
-visibility into *why* a PR exists — what the ticket asked for, what design decisions
-were made, or what constraints were documented in Confluence.
-
-This spec adds a **Doc Context** phase that runs before the review agents and injects
-structured, diff-aware summaries of the PR's linked ADO work items and any Confluence
-pages referenced in those work items. The summaries are internal context only — no new
-comments are posted to the PR.
-
-Decisions recorded in ADR-0010: the Confluence fetch logic lives as a self-contained
-`.mjs` script inside `pr-review/` (not a shared package, not a dependency on
-`unic-confluence`), using the same credential lookup convention as `unic-confluence`
-(env vars first, `~/.unic-confluence.json` fallback).
-
-Extensibility approach recorded in ADR-0011: new work item sources and doc sources are
-added as parallel paths alongside the ADO/Confluence path — no plugin registry until a
-third distinct source type is introduced.
-
-## Current behaviour
-
-Steps 1–12 in `commands/review-pr.md` pass the PR title, description, diff, and key
-file contents to review agents. No work item or Confluence data is fetched or injected.
-
-## Target behaviour
-
-After step 4 (changed files list is known), a new **step 4a** runs:
-
-1. Fetch all work items linked to the PR from the ADO API.
-2. If none are linked, skip silently and proceed to step 5.
-3. For each linked work item, spawn a **Doc Context Sub-agent** in parallel.
-   Each sub-agent receives the work item ID, title, and description, plus the
-   changed files list and the local diff (if already available from step 5).
-4. Each Doc Context Sub-agent:
-   a. Summarises the work item description (diff-aware: focus on what is relevant
-      to the changed files).
-   b. Extracts any Confluence page URLs from the work item description.
-   c. For each Confluence URL, spawns a nested **Doc Context Sub-agent** (parallel)
-      that fetches the page and returns a diff-aware summary.
-5. Credential check for Confluence fetching:
-   - Env vars (`CONFLUENCE_URL`, `CONFLUENCE_USER`, `CONFLUENCE_TOKEN`) take
-     precedence; `~/.unic-confluence.json` is the fallback.
-   - If credentials are absent **and** at least one Confluence URL was found: emit
-     a console warning (never post to the PR):
-     ```
-     ⚠ Confluence pages not fetched — set CONFLUENCE_URL, CONFLUENCE_USER,
-       CONFLUENCE_TOKEN (or run `node scripts/push-to-confluence.mjs --setup`)
-       to enable doc-aware review.
-     ```
-   - If credentials are absent and no Confluence URLs were found: skip silently.
-6. If a Confluence page fetch fails (network error, 401, 403, etc.): skip that page,
-   emit a console warning (`⚠ Could not fetch Confluence page <url> — <reason>`),
-   continue with whatever other context was gathered.
-7. Steps 5–7 (local diff, read key files, determine aspects) run **in parallel** with
-   step 4a. Step 8 waits for all of step 4a to complete before launching review agents.
-8. Collected summaries are assembled into a **Doc Context block** (see format below)
-   and injected as a preamble into every review agent's prompt in step 8.
-
-Doc Context block format:
-
-```markdown
-## Business context for this PR
-
-### Work item: [{ID}] {Title}
-{diff-aware summary of work item description}
-
-### Confluence — {Page Title} ({URL})
-{diff-aware summary of page content}
-```
-
-If only work items are available (no Confluence links or credentials absent), the block
-contains only the work item sections. If no Doc Context is gathered at all, the preamble
-is omitted and review agents receive the same prompt as today.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `commands/review-pr.md` | Add step 4a; inject Doc Context block into step 8 agent prompts |
-| `scripts/confluence-client.mjs` | New — credential loading + Confluence v2 page fetch |
-| `docs/adr/0010-inline-confluence-client.md` | Already written (2026-05) |
-| `CONTEXT.md` | Already updated — Doc Context and Doc Context Sub-agent terms |
-| `docs/plans/README.md` | Add spec 10 row |
-| `CHANGELOG.md` | New `[Unreleased]` entry |
-
-## Implementation steps
-
-### 1. Create `scripts/confluence-client.mjs`
-
-A standalone Node.js script (`// @ts-check`, ESM, zero external deps) that:
-
-- Exports a `loadCredentials()` function: checks env vars first, falls back to
-  `~/.unic-confluence.json` (same logic as `unic-confluence/scripts/push-to-confluence.mjs`
-  `loadCredentials()`). Throws a descriptive error if neither source yields valid creds.
-- Exports a `fetchPageText(pageUrl, credentials)` function:
-  - Extracts the numeric page ID from the Confluence URL
-    (pattern: `/pages/{id}/` or `/pages/{id}`).
-  - Calls `GET {CONFLUENCE_URL}/wiki/api/v2/pages/{pageId}?body-format=storage`
-    with Basic auth (`username:token` base64-encoded).
-  - Returns the `body.storage.value` string (Confluence storage XML). The calling
-    agent reads through the XML directly — no dedicated parser needed.
-  - Throws on non-2xx response or network error (caller handles the warning).
-- CLI entry point (when called directly):
-  ```
-  node scripts/confluence-client.mjs <confluence-page-url>
-  ```
-  Prints the storage-format body to stdout; exits non-zero on error.
-
-### 2. Add step 4a to `commands/review-pr.md`
-
-Insert after the step 4 heading block, before step 5:
-
-```markdown
-## Step 4a — Gather Doc Context (work items + Confluence pages)
-
-Fetch work items linked to the PR:
-
-\```bash
-az devops invoke \
-  --area git \
-  --resource pullRequestWorkItems \
-  --route-parameters "repositoryId={REPO_ID}" "pullRequestId={PR_ID}" \
-  --org {ORG_URL} \
-  --api-version "7.1" \
-  --output json
-\```
-
-If the `value` array is empty, skip to step 5.
-
-For each work item ID returned, fetch its details:
-
-\```bash
-az boards work-item show --id {WI_ID} --org {ORG_URL} --output json
-\```
-
-Capture `fields.System.Title` and `fields.System.Description`.
-
-Spawn one **Doc Context Sub-agent** per work item in parallel (single message).
-Each sub-agent receives:
-- Work item ID, title, and description (HTML — read through the markup)
-- The changed files list from step 4
-- The local diff from step 5 (pass it if already available; otherwise omit)
-
-Each Doc Context Sub-agent must:
-1. Summarise the work item description, focusing only on what is relevant to the
-   changed files. Ignore sections that have no bearing on the diff.
-2. Extract all Confluence URLs from the description.
-3. Attempt to load Confluence credentials via
-   `node scripts/confluence-client.mjs --check-creds` (exit 0 = creds available).
-4. If creds available: spawn one nested Doc Context Sub-agent per Confluence URL
-   (parallel), each running:
-   `node scripts/confluence-client.mjs <url>`
-   Pass the storage-format output plus the changed files list to a sub-agent that
-   returns a diff-aware plain-text summary of the page.
-5. If creds absent and Confluence URLs were found: emit the console warning (see
-   Target behaviour §5). Do not spawn Confluence sub-agents.
-6. Return a structured Doc Context block (work item section + any Confluence
-   sections) using the format defined in Target behaviour.
-
-Collect all sub-agent outputs. Concatenate into a single Doc Context block.
-Store as `DOC_CONTEXT` for injection in step 8.
-```
-
-### 3. Inject Doc Context into step 8
-
-In the step 8 agent prompt template, prepend `DOC_CONTEXT` before the diff content:
-
-```
-{DOC_CONTEXT if non-empty}
-
-## Diff
-{diff content}
-
-## Changed files
-{file contents}
-```
-
-Add `WebFetch` to the `allowed-tools` frontmatter list (needed by the Confluence
-page sub-agents).
-
-### 4. Update `docs/plans/README.md`
-
-Add spec 10 row to the status table (no dependency on specs 00–09).
-
-### 5. Add `CHANGELOG.md` entry
-
-Under `[Unreleased]`, add:
-
-```markdown
-### Added
-- Doc Context enrichment: before review agents run, fetch linked ADO work items and
-  any Confluence pages referenced in their descriptions; inject structured,
-  diff-aware summaries as business context into every review agent's prompt.
-  Requires Confluence credentials (`CONFLUENCE_URL`, `CONFLUENCE_USER`,
-  `CONFLUENCE_TOKEN` or `~/.unic-confluence.json`) for Confluence page fetching;
-  degrades gracefully when absent or unreachable.
-```
-
-### 6. Bump version
-
-Check what version the re-review feature (`docs/issues/pr-review-rereview/09-version-bump-and-release.md`) shipped at before running the bump — this feature must take the next available minor version after re-review.
-
-```bash
-pnpm --filter pr-review bump minor
-```
-
-## Verification
-
-- PR with ≥1 linked work item, no Confluence links: review agents receive a
-  `## Business context` preamble containing the work item summary; no console warning.
-- PR with a linked work item whose description contains a Confluence URL, valid creds:
-  review agents receive work item summary + Confluence page summary in the preamble.
-- PR with no linked work items: step 4a exits silently; review agent prompts unchanged.
-- Confluence credentials absent + Confluence URLs found: console warning printed;
-  review continues with work item summaries only.
-- Confluence credentials absent + no Confluence URLs: no warning; no change in output.
-- Confluence page fetch returns 403: that page skipped with console warning; other
-  pages and work items still included in Doc Context.
-- No new threads or comments posted to the PR as a result of this spec.
-- `pnpm --filter pr-review verify:changelog` passes.
-- `node scripts/confluence-client.mjs --help` (or with a valid URL) runs without error.
-
-## Acceptance criteria
-
-- All changes land in a single conventional commit:
-  `feat(pr-review): add doc context enrichment from work items and Confluence pages`
-- Version bumped to the next available minor version after the re-review feature ships
-  (check `docs/issues/pr-review-rereview/09-version-bump-and-release.md` first).
-- `plugin.json` and `marketplace.json` versions match.
-- `CHANGELOG.md` has a dated entry for the new version when the bump is finalised.
-- `scripts/confluence-client.mjs` has zero external runtime dependencies.
-- Doc Context is never posted to the PR — it is internal agent context only.
-- Degradation is graceful at every tier: no work items → skip; no creds → warn + skip
-  Confluence; fetch error → warn + skip that page; local diff unavailable → sub-agents
-  work from changed files list only.
-
-## Out of scope
-
-- GitHub PR support (separate future spec).
-- Fetching Confluence pages linked from the PR description directly (not via a work
-  item) — a follow-up if demand exists.
-- Caching Confluence page content across runs.
-- Any changes to comment posting, thread management, or the Bot Signature.
-- Re-review integration (specs 00–09 own that).
diff --git a/apps/claude-code/pr-review/docs/plans/11-doc-context-spawn-reliability.md b/apps/claude-code/pr-review/docs/plans/11-doc-context-spawn-reliability.md
deleted file mode 100644
index a327649e..00000000
--- a/apps/claude-code/pr-review/docs/plans/11-doc-context-spawn-reliability.md
+++ /dev/null
@@ -1,200 +0,0 @@
-# 11. Doc Context Spawn Reliability
-
-**Status: done — 2026-05-08**
-
-- Priority: P1
-- Effort: S
-- Version impact: patch (bug fix; no new user-visible capability)
-- Depends on: 10
-- Touches: `commands/review-pr.md`, new `.agents/doc-context-orchestrator.md`, new `.agents/doc-context-synthesizer.md`, `docs/adr/0012-plain-text-doc-context-agent-return.md`, `docs/plans/README.md`, `CHANGELOG.md`
-
-## Context
-
-Spec 10 added the Doc Context phase and shipped it as done. In practice the phase is silently skipped on every run — review agents never receive business context. Three defects combine to produce this outcome:
-
-1. **Step 4a has no explicit Agent() spawn template.** The step describes what a Doc Context Sub-agent "must do" in prose bullets. The orchestrator satisfies the intent inline and proceeds to step 5 without spawning any agent.
-
-2. **`confluence-client.mjs` is called with a relative path.** `node scripts/confluence-client.mjs` resolves against the reviewed project's root, not the plugin directory. The path never finds the script.
-
-3. **`DOC_CONTEXT` is never initialised.** There is no `DOC_CONTEXT=''` before step 4a. When the phase is skipped the variable is undefined; step 8's `{DOC_CONTEXT if non-empty}` check produces no error, making the failure invisible.
-
-**Why extraction, not inline fix:** simply adding explicit `Agent()` templates inline in step 4a would fix the spawn defect, but it would cause the main orchestrator to accumulate all work item HTML, Confluence storage-format XML, and per-page summaries in its own context window. Token consumption would grow proportionally to PR complexity. Delegating to a dedicated orchestrator agent isolates that content in a fresh context window.
-
-## Target behaviour
-
-- `DOC_CONTEXT=''` is initialised at the very top of step 4a before any fetch.
-- Step 4a pre-fetches work item IDs in bash (to decide whether to spawn at all), then waits for the diff from step 5 before spawning the orchestrator agent.
-- The Doc Context Orchestrator agent handles the entire gathering phase in its own context: work item detail fetches, credential check, Confluence page fetches, Work Item Summarizer agents, and final synthesis.
-- The Doc Context Synthesizer agent produces a single coherent flat narrative — no work item headings, no redundant content.
-- Confluence credential check runs exactly once in the orchestrator agent (not per fetcher).
-- All absolute paths to `confluence-client.mjs` are resolved from `${CLAUDE_PLUGIN_ROOT}` in bash and injected into agent prompts as literal strings (agents do not inherit env vars).
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `commands/review-pr.md` | Rewrite step 4a: init `DOC_CONTEXT=''`, pre-fetch work item IDs, wait for diff, delegate to Doc Context Orchestrator agent via explicit `Agent()` call; update parallelism note |
-| `.agents/doc-context-orchestrator.md` | New agent: self-contained orchestrator for the entire Doc Context gathering phase |
-| `.agents/doc-context-synthesizer.md` | New agent: produces a single coherent flat Doc Context narrative from all work item and Confluence summaries |
-| `docs/adr/0012-plain-text-doc-context-agent-return.md` | New ADR: agent returns plain markdown, not JSON |
-| `docs/plans/README.md` | Add spec 11 row |
-| `CHANGELOG.md` | Add entry under `[Unreleased]` |
-
-No changes to `scripts/confluence-client.mjs`, tests, or any other file.
-
-## Implementation steps
-
-### 1. Write ADR-0012
-
-Create `docs/adr/0012-plain-text-doc-context-agent-return.md` recording the decision to have the Doc Context Orchestrator agent return plain markdown rather than a JSON wrapper (see ADR format below).
-
-### 2. Create `.agents/doc-context-orchestrator.md`
-
-This agent receives: `ORG_URL`, `PR_ID`, a list of work item IDs, the full diff, the changed files list, and `CONFLUENCE_CLIENT_PATH` (absolute, pre-expanded from `${CLAUDE_PLUGIN_ROOT}`). It does **not** receive the initial work item ID list fetch — that happened in bash.
-
-The agent must:
-
-1. For each work item ID, run:
-   ```bash
-   az boards work-item show --id {WI_ID} --org {ORG_URL} --output json
-   ```
-   If the command fails: emit `⚠ Could not fetch work item {WI_ID} — {error}` to console and skip that item.
-
-2. For each work item, extract the relevant description fields based on work item type:
-   - `Bug` → `Microsoft.VSTS.TCM.ReproSteps` + `Microsoft.VSTS.TCM.SystemInfo`
-   - `User Story`, `Task`, `Feature`, or any unrecognised type → `System.Description`
-
-3. Spawn one **Work Item Summarizer agent** per work item **in parallel** (single message). Each agent receives: work item ID, title, description (HTML), the changed files list, the diff.
-
-   Each Work Item Summarizer agent must:
-   - Summarise the description as plain text, focusing only on what is relevant to the changed files and diff. Ignore sections with no bearing on the diff.
-   - Extract all Confluence URLs from the description.
-   - Return structured output: work item summary text + list of `{url, relevantContext}` pairs (one per Confluence URL found).
-
-4. Run the Confluence credential check **once**:
-   ```bash
-   node {CONFLUENCE_CLIENT_PATH} --check-creds
-   ```
-   Exit 0 = creds available. Any other outcome = creds absent.
-
-   If creds are absent and any work item summarizer returned Confluence URLs: emit the following warning to console (never post to the PR) and skip all Confluence fetching:
-   ```
-   ⚠ Confluence pages not fetched — set CONFLUENCE_URL, CONFLUENCE_USER, CONFLUENCE_TOKEN (or create ~/.unic-confluence.json with { url, username, token }) to enable doc-aware review.
-   ```
-
-5. If creds are available: collect all unique Confluence URLs across all work item summarizer outputs. Spawn one **Confluence Fetcher agent** per unique URL **in parallel** (single message).
-
-   Each Confluence Fetcher agent must:
-   - Run: `node {CONFLUENCE_CLIENT_PATH} {URL}`
-   - If successful: return a plain-text summary of the page, focused on what is relevant to the changed files.
-   - If the fetch fails (network error, 401, 403, etc.): emit `⚠ Could not fetch Confluence page {URL} — {reason}` to console and return nothing.
-
-6. Pass all work item summarizer outputs and all Confluence fetcher outputs to the **Doc Context Synthesizer agent**.
-
-7. Return the Doc Context Synthesizer agent's output verbatim as the agent's final output.
-
-### 3. Create `.agents/doc-context-synthesizer.md`
-
-This agent receives all work item summaries and all Confluence page summaries (potentially overlapping across work items). It must produce a single coherent `## Business context for this PR` section with no redundant content.
-
-Rules:
-- Flat narrative — no work item headings, no per-ticket structure.
-- Synthesise overlapping content: if multiple work items or Confluence pages describe the same feature, merge them into one coherent description rather than repeating the same information.
-- Focus on business intent: what the PR is supposed to accomplish and why, from the specifications' perspective. Omit implementation details already visible in the diff.
-- If no meaningful context was gathered (all work items and pages failed or returned empty): return an empty string.
-
-Output format:
-```markdown
-## Business context for this PR
-
-{synthesised narrative}
-```
-
-### 4. Rewrite step 4a in `commands/review-pr.md`
-
-Replace the current step 4a content with:
-
-```
-DOC_CONTEXT=''
-
-Fetch work items linked to the PR:
-
-  az devops invoke \
-    --area git \
-    --resource pullRequestWorkItems \
-    --route-parameters "repositoryId={REPO_ID}" "pullRequestId={PR_ID}" \
-    --org {ORG_URL} \
-    --api-version "7.1" \
-    --output json
-
-If the `value` array is empty or the command fails, leave DOC_CONTEXT='' and proceed.
-
-Otherwise, wait for the diff from step 5 to be available (step 4a and step 5 run
-concurrently up to this point; only the orchestrator spawn waits for the diff).
-
-Resolve the plugin path:
-
-  CONFLUENCE_CLIENT_PATH="${CLAUDE_PLUGIN_ROOT}/scripts/confluence-client.mjs"
-
-Delegate to the Doc Context Orchestrator agent:
-
-  Agent(
-    subagent_type: "pr-review:doc-context-orchestrator",
-    prompt: "Orchestrate Doc Context gathering.
-
-    ORG_URL: {ORG_URL}
-    PR_ID: {PR_ID}
-    Work item IDs: {comma-separated list from value array}
-    Confluence client path: {CONFLUENCE_CLIENT_PATH}
-
-    Changed files:
-    {CHANGED_FILES_LIST}
-
-    Diff:
-    {RAW_DIFF}
-
-    Return the complete Doc Context markdown block, or an empty string if no
-    meaningful context could be gathered."
-  )
-
-Store the agent's output as DOC_CONTEXT.
-```
-
-Also update the parallelism note: "Step 4a pre-fetch (work item IDs) runs in parallel with step 5. The orchestrator agent spawn waits for the diff from step 5. Step 8 waits for the orchestrator agent to complete before launching review agents."
-
-### 5. Update `docs/plans/README.md`
-
-Add spec 11 row to the status table.
-
-### 6. Add `CHANGELOG.md` entry
-
-Under `[Unreleased]`, add a `Fixed` entry describing the three defects and the orchestrator extraction.
-
-## Verification
-
-- PR with ≥1 linked work item (non-Bug), no Confluence links: `DOC_CONTEXT` is non-empty; review agents receive the `## Business context` preamble.
-- PR with a Bug work item: `ReproSteps` and `SystemInfo` are extracted; `System.Description` is not used.
-- PR with a Confluence-linked work item, valid credentials: preamble includes synthesised content from both the work item and the Confluence page.
-- PR with two work items linking the same Confluence page: the page appears once in the output, not twice.
-- PR with no linked work items: step 4a exits with `DOC_CONTEXT=''`; agent prompts in step 8 unchanged.
-- Credentials absent + Confluence URLs found: console warning printed; `DOC_CONTEXT` contains work item summaries only.
-- No new PR threads or comments produced by the Doc Context phase.
-
-## Acceptance criteria
-
-- [ ] `DOC_CONTEXT=''` is the first statement in step 4a.
-- [ ] Step 4a contains an explicit `Agent(subagent_type: "pr-review:doc-context-orchestrator", ...)` call with all required context.
-- [ ] All paths to `confluence-client.mjs` in agent prompts are absolute strings resolved from `${CLAUDE_PLUGIN_ROOT}`.
-- [ ] The orchestrator agent runs the credential check exactly once.
-- [ ] Bug work items use `ReproSteps` + `SystemInfo`; all other types use `System.Description`.
-- [ ] The Doc Context Synthesizer agent produces a flat narrative with no per-work-item headings.
-- [ ] ADR-0012 is committed alongside the implementation.
-
-## Out of scope
-
-- Changes to `scripts/confluence-client.mjs`.
-- Changes to thread posting, re-review logic, or Bot Signature.
-- GitHub PR support.
-- Caching Doc Context across review runs.
-- Jira or other work item sources.
-- Re-review handling of Doc Context (no change to when/how DOC_CONTEXT is injected in step 8).
diff --git a/apps/claude-code/pr-review/docs/plans/README.md b/apps/claude-code/pr-review/docs/plans/README.md
deleted file mode 100644
index d05ee12b..00000000
--- a/apps/claude-code/pr-review/docs/plans/README.md
+++ /dev/null
@@ -1,20 +0,0 @@
-# Re-review feature roadmap
-
-Goal: when `/unic-pr-review:review-pr <url>` runs against a PR that already has Claude Code review threads, update those threads incrementally instead of duplicating them.
-
-## Status
-
-| #  | Spec                                  | Status  | Depends on |
-|----|---------------------------------------|---------|------------|
-| 00 | Pre-flight: ADR supersession + prettierignore guard | done    | — |
-| 01 | Normalize Claude Code signature       | done    | 00         |
-| 02 | Detect prior review on PR             | done    | 01         |
-| 03 | Target latest PR iteration            | done    | 02         |
-| 04 | Incremental diff baseline             | done    | 03         |
-| 05 | Classify existing threads             | done    | 02         |
-| 06 | Reply to threads instead of posting   | done    | 04, 05     |
-| 07 | Summary comment policy on re-review   | done    | 06         |
-| 08 | Version bump, README, CLAUDE.md       | done    | 07         |
-| 09 | Test harness — node:test + modules    | done    | 02, 05, 06 |
-| 10 | Doc Context enrichment — work items + Confluence pages | done | — |
-| 11 | Doc Context spawn reliability — fix silently skipped phase | done | 10 |
diff --git a/apps/claude-code/pr-review/package.json b/apps/claude-code/pr-review/package.json
index 45841e78..bf28ae4c 100644
--- a/apps/claude-code/pr-review/package.json
+++ b/apps/claude-code/pr-review/package.json
@@ -1,24 +1,22 @@
 {
-  "name": "pr-review",
-  "version": "1.2.10",
-  "private": true,
-  "license": "LGPL-3.0-or-later",
-  "type": "module",
-  "packageManager": "pnpm@10.33.0",
+  "devDependencies": {
+    "@unic/release-tools": "workspace:*"
+  },
   "engines": {
     "node": ">=22",
     "pnpm": ">=10"
   },
+  "license": "LGPL-3.0-or-later",
+  "name": "pr-review",
+  "packageManager": "pnpm@10.33.0",
+  "private": true,
   "scripts": {
-    "test": "node --test tests/parse-signature.test.mjs tests/classify-thread.test.mjs tests/match-finding.test.mjs tests/detect-prior-review.test.mjs tests/confluence-client.test.mjs tests/ado-fetcher.test.mjs tests/ado-writer.test.mjs tests/pre-pr.test.mjs tests/parse-diff-hunks.test.mjs tests/mode-detection.test.mjs tests/notices.test.mjs tests/classify-http-error.test.mjs tests/fetch-work-items.test.mjs tests/fetch-iterations.test.mjs tests/parse-write-response.test.mjs tests/detect-default-branch.test.mjs tests/plugin-structure.test.mjs",
     "bump": "unic-bump",
     "sync-version": "unic-sync-version",
     "tag": "unic-tag",
-    "verify:changelog": "unic-verify-changelog",
-    "ralph": "ralph run -c ralph.yml -H builtin:code-assist"
+    "test": "node --test tests/parse-signature.test.mjs tests/classify-thread.test.mjs tests/match-finding.test.mjs tests/detect-prior-review.test.mjs tests/confluence-client.test.mjs tests/ado-fetcher.test.mjs tests/ado-writer.test.mjs tests/pre-pr.test.mjs tests/parse-diff-hunks.test.mjs tests/mode-detection.test.mjs tests/notices.test.mjs tests/classify-http-error.test.mjs tests/fetch-work-items.test.mjs tests/fetch-iterations.test.mjs tests/parse-write-response.test.mjs tests/detect-default-branch.test.mjs tests/plugin-structure.test.mjs tests/ado-cli-completeness.test.mjs tests/ado-cli-smoke.test.mjs",
+    "verify:changelog": "unic-verify-changelog"
   },
-  "devDependencies": {
-    "@ralph-orchestrator/ralph-cli": "catalog:",
-    "@unic/release-tools": "workspace:*"
-  }
+  "type": "module",
+  "version": "1.2.10"
 }
diff --git a/apps/claude-code/pr-review/ralph.yml b/apps/claude-code/pr-review/ralph.yml
deleted file mode 100644
index 6b7525bb..00000000
--- a/apps/claude-code/pr-review/ralph.yml
+++ /dev/null
@@ -1,35 +0,0 @@
-# Ralph Orchestrator Configuration
-# Generated by: ralph init --backend claude
-# Project: unic-pr-review
-# Docs: https://github.com/mikeyobrien/ralph-orchestrator
-
-cli:
-  backend: 'claude'
-
-event_loop:
-  prompt_file: 'PROMPT.md'
-  completion_promise: 'LOOP_COMPLETE'
-  max_iterations: 100 # 15 specs * generous retry budget
-  max_runtime_seconds: 28800 # 8 hours; abort if a spec stalls
-
-# ─────────────────────────────────────────────────────────────────────────────
-# Additional Configuration (uncomment to customize)
-# ─────────────────────────────────────────────────────────────────────────────
-
-# core:
-#   scratchpad: ".ralph/agent/scratchpad.md"
-#   specs_dir: ".ralph/specs/"
-
-# Custom hats for multi-agent workflows:
-# hats:
-#   builder:
-#     name: "Builder"
-#     triggers: ["build.task"]
-#     publishes: ["build.done", "build.blocked"]
-#
-#   reviewer:
-#     name: "Reviewer"
-#     triggers: ["review.request"]
-#     publishes: ["review.approved", "review.changes_requested"]
-
-# Create PROMPT.md with your task, then run: ralph run
diff --git a/apps/claude-code/pr-review/scripts/ado/cli-completeness.mjs b/apps/claude-code/pr-review/scripts/ado/cli-completeness.mjs
new file mode 100644
index 00000000..5f99822b
--- /dev/null
+++ b/apps/claude-code/pr-review/scripts/ado/cli-completeness.mjs
@@ -0,0 +1,187 @@
+// @ts-check
+
+/**
+ * Pure-function CLI inventory check.
+ *
+ * Scans source-file contents for `az` invocations and returns the ones not
+ * registered in the inventory and not exempted by the allowlist. Knows nothing
+ * about the filesystem — callers pass file contents in.
+ */
+
+/**
+ * @typedef {{ path: string, content: string }} Source
+ *
+ * @typedef {{ kind: 'invoke', area: string, resource: string, command: string[], helpKeywordsRequired: string[] }} InvokeEntry
+ * @typedef {{ kind: 'repos' | 'boards', command: string[], helpKeywordsRequired: string[] }} SubcommandEntry
+ * @typedef {InvokeEntry | SubcommandEntry} InventoryEntry
+ */
+
+/**
+ * Find every `az` command in `sources` that is neither in `inventory` nor
+ * exempted by `allowlist`. Each returned entry reads
+ * `"<command-shape> (<source-path>)"` so test failures point at the offending file.
+ *
+ * @param {{ sources: Source[], inventory: InventoryEntry[], allowlist: RegExp[] }} args
+ * @returns {string[]}
+ */
+export function findUninventoriedCommands({ sources, inventory, allowlist }) {
+	const out = []
+	for (const { path, content } of sources) {
+		for (const shape of extractAzShapes(content)) {
+			if (matchesInventory(shape, inventory)) continue
+			if (matchesAllowlist(shape, allowlist)) continue
+			out.push(`${shape} (${path})`)
+		}
+	}
+	return out
+}
+
+/**
+ * Extract the normalized shape string for every `az` invocation in a source.
+ *
+ * Shape strings:
+ * - `az repos pr show` (leading non-flag tokens)
+ * - `az devops invoke --area git --resource pullRequestThreads` (invoke + area/resource flags)
+ *
+ * @param {string} content
+ * @returns {string[]}
+ */
+function extractAzShapes(content) {
+	const shapes = []
+	const text = content.replace(/\r\n/g, '\n')
+	// Delimiter class deliberately omits ` ' " — markdown inline code (`az ...`),
+	// JS string literals ("az ..."), and prose quotes ('az ...') wrap references,
+	// not real invocations. Real shell uses whitespace, $( ... ), &&, |, etc.
+	const matches = text.matchAll(/(^|[\s(&|])az\s/g)
+	for (const match of matches) {
+		const start = (match.index ?? 0) + match[1].length
+		if (insideShellCommentOrString(text, start)) continue
+		const tokens = tokenize(text, start)
+		if (tokens.length === 0) continue
+		const head = leadingNonFlagTokens(tokens)
+		if (head.length < 2) {
+			// Root-flag forms like `az --version` — emit `az <first-flag>` so the
+			// allowlist / inventory check still covers them.
+			const firstFlag = tokens.find((t) => t.startsWith('--'))
+			if (firstFlag) shapes.push(`az ${firstFlag}`)
+			continue
+		}
+		const shape = head[1] === 'devops' && head[2] === 'invoke' ? withAreaResource(tokens, head) : head.join(' ')
+		shapes.push(shape)
+	}
+	return shapes
+}
+
+/**
+ * Return true when `position` falls inside a shell comment, a markdown inline
+ * code span (`` `...` ``), or a quoted string (single or double) on the same line.
+ *
+ * Shell comments: a `#` preceded by start-of-line or whitespace before `position`.
+ * Quoted strings / inline code: an odd count of the matching quote / backtick char
+ * between the previous newline and `position`.
+ *
+ * @param {string} text
+ * @param {number} position
+ * @returns {boolean}
+ */
+function insideShellCommentOrString(text, position) {
+	const lineStart = text.lastIndexOf('\n', position - 1) + 1
+	const before = text.slice(lineStart, position)
+	if (/(^|\s)#/.test(before)) return true
+	for (const delim of ['"', "'", '`']) {
+		let count = 0
+		for (const ch of before) if (ch === delim) count++
+		if (count % 2 === 1) return true
+	}
+	return false
+}
+
+/**
+ * Read tokens starting at `start`, honouring shell backslash-newline continuations.
+ *
+ * @param {string} text
+ * @param {number} start
+ * @returns {string[]}
+ */
+function tokenize(text, start) {
+	let i = start
+	let buf = ''
+	while (i < text.length) {
+		const ch = text[i]
+		if (ch === '\\' && text[i + 1] === '\n') {
+			i += 2
+			buf += ' '
+			continue
+		}
+		if (ch === '\n' || ch === ';' || ch === '|') break
+		buf += ch
+		i++
+	}
+	return buf.split(/\s+/).filter(Boolean)
+}
+
+/**
+ * Take leading tokens until we hit a `-flag` or value position.
+ *
+ * @param {string[]} tokens
+ * @returns {string[]}
+ */
+function leadingNonFlagTokens(tokens) {
+	const head = []
+	for (const t of tokens) {
+		if (t.startsWith('-')) break
+		head.push(t)
+	}
+	return head
+}
+
+/**
+ * For `az devops invoke`, capture `--area X --resource Y` so the shape uniquely
+ * identifies the API endpoint. Missing values are surfaced as `<missing>`.
+ *
+ * @param {string[]} tokens
+ * @param {string[]} head
+ * @returns {string}
+ */
+function withAreaResource(tokens, head) {
+	const area = valueFor(tokens, '--area') ?? '<missing>'
+	const resource = valueFor(tokens, '--resource') ?? '<missing>'
+	return `${head.join(' ')} --area ${area} --resource ${resource}`
+}
+
+/**
+ * @param {string[]} tokens
+ * @param {string} flag
+ * @returns {string | undefined}
+ */
+function valueFor(tokens, flag) {
+	const idx = tokens.indexOf(flag)
+	if (idx === -1 || idx === tokens.length - 1) return undefined
+	const next = tokens[idx + 1]
+	return next.startsWith('-') ? undefined : next
+}
+
+/**
+ * @param {string} shape
+ * @param {InventoryEntry[]} inventory
+ * @returns {boolean}
+ */
+function matchesInventory(shape, inventory) {
+	for (const entry of inventory) {
+		if (entry.kind === 'invoke') {
+			if (shape === `${entry.command.join(' ')} --area ${entry.area} --resource ${entry.resource}`) return true
+		} else if (shape === entry.command.join(' ')) {
+			return true
+		}
+	}
+	return false
+}
+
+/**
+ * @param {string} shape
+ * @param {RegExp[]} allowlist
+ * @returns {boolean}
+ */
+function matchesAllowlist(shape, allowlist) {
+	return allowlist.some((re) => re.test(shape))
+}
diff --git a/apps/claude-code/pr-review/scripts/ado/notices.mjs b/apps/claude-code/pr-review/scripts/ado/notices.mjs
index 3a533e04..6bd4c505 100644
--- a/apps/claude-code/pr-review/scripts/ado/notices.mjs
+++ b/apps/claude-code/pr-review/scripts/ado/notices.mjs
@@ -2,7 +2,7 @@
 
 /**
  * @typedef {'info' | 'warning'} NoticeSeverity
- * @typedef {'doc-context' | 'diff-range' | 'work-items' | 'iterations' | 'default-branch' | 'partial-run-check' | 'thread-match' | 'thread-classify' | 'inline-post' | 'summary-post' | 'patch-to-fixed' | 'diff-parse' | 'delta-reply' | 'completion-marker'} NoticeKind
+ * @typedef {'doc-context' | 'diff-range' | 'work-items' | 'iterations' | 'default-branch' | 'partial-run-check' | 'thread-match' | 'thread-classify' | 'thread-fetch' | 'inline-post' | 'summary-post' | 'patch-to-fixed' | 'diff-parse' | 'delta-reply' | 'completion-marker'} NoticeKind
  * @typedef {{ severity: NoticeSeverity, kind: NoticeKind, message: string }} Notice
  * @typedef {'first-review' | 're-review' | 'pre-pr' | 'aborted'} TrailerMode
  * @typedef {{ critical: number, important: number, minor: number }} FindingCounts
diff --git a/apps/claude-code/pr-review/scripts/mode-detection.mjs b/apps/claude-code/pr-review/scripts/mode-detection.mjs
index 4665b7d0..571cc736 100644
--- a/apps/claude-code/pr-review/scripts/mode-detection.mjs
+++ b/apps/claude-code/pr-review/scripts/mode-detection.mjs
@@ -2,6 +2,13 @@
 
 import { detectPriorReview } from './re-review/detect-prior-review.mjs'
 
+/**
+ * Canonical bot-signature prefix used to detect prior Reviews via signature
+ * matching (ADR 0001 + ADR 0002). Load-bearing — the exact wording is part of
+ * the re-review detection contract.
+ */
+export const SIGNATURE_PREFIX = '🤖 *Reviewed by Claude Code*'
+
 /**
  * @typedef {{
  *   mode: 'first-review' | 're-review',
@@ -23,8 +30,8 @@ import { detectPriorReview } from './re-review/detect-prior-review.mjs'
  */
 export function detectMode({ threads, signaturePrefix }) {
 	const r = detectPriorReview({
-		// detect-prior-review accepts the raw ADO thread shape; the orchestrator
-		// passes whatever `az repos pr thread list` returned, untouched.
+		// detect-prior-review accepts the raw ADO thread shape; callers pass the
+		// `.value` array from the `pullRequestThreads` endpoint, untouched.
 		// @ts-expect-error -- runtime-validated by detectPriorReview's own guards
 		threads: Array.isArray(threads) ? threads : [],
 		signaturePrefix,
diff --git a/apps/claude-code/pr-review/tests/ado-cli-completeness.test.mjs b/apps/claude-code/pr-review/tests/ado-cli-completeness.test.mjs
new file mode 100644
index 00000000..ba8b08da
--- /dev/null
+++ b/apps/claude-code/pr-review/tests/ado-cli-completeness.test.mjs
@@ -0,0 +1,147 @@
+// @ts-check
+
+import assert from 'node:assert/strict'
+import { describe, it } from 'node:test'
+import { findUninventoriedCommands } from '../scripts/ado/cli-completeness.mjs'
+
+describe('findUninventoriedCommands', () => {
+	it('returns [] for empty sources', () => {
+		const result = findUninventoriedCommands({ sources: [], inventory: [], allowlist: [] })
+		assert.deepEqual(result, [])
+	})
+
+	it('returns single inline az repos pr show when not in inventory', () => {
+		const result = findUninventoriedCommands({
+			sources: [{ path: 'agents/x.md', content: 'PR=$(az repos pr show --id 1 --org $O)' }],
+			inventory: [],
+			allowlist: [],
+		})
+		assert.deepEqual(result, ['az repos pr show (agents/x.md)'])
+	})
+
+	it('returns [] when the command matches an inventory entry', () => {
+		const result = findUninventoriedCommands({
+			sources: [{ path: 'agents/x.md', content: 'az repos pr show --id 1' }],
+			inventory: [{ kind: 'repos', command: ['az', 'repos', 'pr', 'show'], helpKeywordsRequired: [] }],
+			allowlist: [],
+		})
+		assert.deepEqual(result, [])
+	})
+
+	it('returns [] when the command matches an allowlist regex', () => {
+		const result = findUninventoriedCommands({
+			sources: [{ path: 'agents/x.md', content: 'az --version && az extension list' }],
+			inventory: [],
+			allowlist: [/^az --version$/, /^az extension list$/],
+		})
+		assert.deepEqual(result, [])
+	})
+
+	it('handles multi-line bash with backslash continuations and split --area/--resource flags', () => {
+		const block = [
+			'RESP=$(az devops invoke \\',
+			'  --area git \\',
+			'  --resource pullRequestThreads \\',
+			'  --route-parameters "repositoryId=$REPO_ID" \\',
+			'  --org "$ORG_URL" \\',
+			'  --output json)',
+		].join('\n')
+		const result = findUninventoriedCommands({
+			sources: [{ path: 'agents/ado-fetcher.md', content: block }],
+			inventory: [],
+			allowlist: [],
+		})
+		assert.deepEqual(result, ['az devops invoke --area git --resource pullRequestThreads (agents/ado-fetcher.md)'])
+	})
+
+	it('returns only the uninventoried commands when source mixes allowlisted, inventoried, and uninventoried', () => {
+		const content = [
+			'az --version',
+			'az extension list | grep azure-devops',
+			'az repos pr show --id 1',
+			'az boards work-item show --id 42',
+			'az devops invoke --area git --resource pullRequestStatuses',
+		].join('\n')
+		const result = findUninventoriedCommands({
+			sources: [{ path: 'commands/review-pr.md', content }],
+			inventory: [
+				{ kind: 'repos', command: ['az', 'repos', 'pr', 'show'], helpKeywordsRequired: [] },
+				{ kind: 'boards', command: ['az', 'boards', 'work-item', 'show'], helpKeywordsRequired: [] },
+			],
+			allowlist: [/^az --version$/, /^az extension list$/],
+		})
+		assert.deepEqual(result, ['az devops invoke --area git --resource pullRequestStatuses (commands/review-pr.md)'])
+	})
+
+	it('ignores markdown inline-code prose references like `az devops invoke`', () => {
+		const content = [
+			'Use `az devops invoke` to call the API.',
+			'Do not call `az devops invoke --resource pullRequestThreads` for GET ops.',
+			'Tip: pass `az` flags carefully.',
+		].join('\n')
+		const result = findUninventoriedCommands({
+			sources: [{ path: 'agents/ado-writer.md', content }],
+			inventory: [],
+			allowlist: [],
+		})
+		assert.deepEqual(result, [])
+	})
+
+	it('ignores `az` mentions in shell comments (# ... az devops login hint and abort)', () => {
+		const content = ['az repos pr show --id 1', '# 401/403 — surface the az devops login hint and abort.'].join('\n')
+		const result = findUninventoriedCommands({
+			sources: [{ path: 'agents/x.md', content }],
+			inventory: [{ kind: 'repos', command: ['az', 'repos', 'pr', 'show'], helpKeywordsRequired: [] }],
+			allowlist: [],
+		})
+		assert.deepEqual(result, [])
+	})
+
+	it('ignores az references inside JS string literals (single or double quoted)', () => {
+		const content = `throw new Error("Try \\\`az devops login\\\` to re-authenticate.")`
+		const result = findUninventoriedCommands({
+			sources: [{ path: 'scripts/ado/x.mjs', content }],
+			inventory: [],
+			allowlist: [],
+		})
+		assert.deepEqual(result, [])
+	})
+
+	it('ignores az inside an echoed double-quoted error message even when whitespace precedes it', () => {
+		const content =
+			'echo "ERROR: az devops invoke unavailable. Re-install: az extension remove --name azure-devops" >&2'
+		const result = findUninventoriedCommands({
+			sources: [{ path: 'commands/review-pr.md', content }],
+			inventory: [],
+			allowlist: [],
+		})
+		assert.deepEqual(result, [])
+	})
+
+	it('emits `az --version` shape for root-flag invocations so the allowlist still applies', () => {
+		const result = findUninventoriedCommands({
+			sources: [{ path: 'commands/review-pr.md', content: 'az --version || exit 1' }],
+			inventory: [],
+			allowlist: [],
+		})
+		assert.deepEqual(result, ['az --version (commands/review-pr.md)'])
+	})
+
+	it('matches an az devops invoke entry against split --area/--resource flags', () => {
+		const block = ['az devops invoke \\', '  --area git \\', '  --resource pullRequestThreads'].join('\n')
+		const result = findUninventoriedCommands({
+			sources: [{ path: 'a.md', content: block }],
+			inventory: [
+				{
+					kind: 'invoke',
+					area: 'git',
+					resource: 'pullRequestThreads',
+					command: ['az', 'devops', 'invoke'],
+					helpKeywordsRequired: [],
+				},
+			],
+			allowlist: [],
+		})
+		assert.deepEqual(result, [])
+	})
+})
diff --git a/apps/claude-code/pr-review/tests/ado-cli-smoke.test.mjs b/apps/claude-code/pr-review/tests/ado-cli-smoke.test.mjs
new file mode 100644
index 00000000..80bb10fc
--- /dev/null
+++ b/apps/claude-code/pr-review/tests/ado-cli-smoke.test.mjs
@@ -0,0 +1,109 @@
+// @ts-check
+
+import assert from 'node:assert/strict'
+import { spawnSync } from 'node:child_process'
+import { readdirSync, readFileSync, statSync } from 'node:fs'
+import { dirname, join, relative, resolve } from 'node:path'
+import { before, describe, it } from 'node:test'
+import { fileURLToPath } from 'node:url'
+import { findUninventoriedCommands } from '../scripts/ado/cli-completeness.mjs'
+import { ADO_CLI_ALLOWLIST } from './fixtures/ado-cli-allowlist.mjs'
+import { adoCliInventory } from './fixtures/ado-cli-inventory.mjs'
+
+const PLUGIN_ROOT = resolve(dirname(fileURLToPath(import.meta.url)), '..')
+const SCAN_DIRS = ['agents', 'commands', 'scripts']
+const SCANNABLE_EXT = /\.(md|mjs|js)$/
+// scratchpad/ holds historical implementation notes, not active plugin code.
+const SKIP_FRAGMENTS = ['/scratchpad/', '/node_modules/']
+
+/**
+ * @returns {{ path: string, content: string }[]}
+ */
+function loadPluginSources() {
+	const out = []
+	for (const dir of SCAN_DIRS) {
+		const root = join(PLUGIN_ROOT, dir)
+		for (const file of walk(root)) {
+			if (!SCANNABLE_EXT.test(file)) continue
+			// Normalise so the POSIX-style SKIP_FRAGMENTS also match on Windows paths.
+			const normalised = file.replaceAll('\\', '/')
+			if (SKIP_FRAGMENTS.some((f) => normalised.includes(f))) continue
+			out.push({ path: relative(PLUGIN_ROOT, file), content: readFileSync(file, 'utf8') })
+		}
+	}
+	return out
+}
+
+/**
+ * @param {string} dir
+ * @returns {Generator<string>}
+ */
+function* walk(dir) {
+	let entries
+	try {
+		entries = readdirSync(dir)
+	} catch (err) {
+		// Only swallow "directory not present" — every other read failure (permissions, FD
+		// exhaustion, IO error) must surface, otherwise the smoke test reports a green
+		// "no uninventoried commands" even when the scan never ran.
+		if (/** @type {NodeJS.ErrnoException} */ (err).code === 'ENOENT') return
+		throw err
+	}
+	for (const name of entries) {
+		const full = join(dir, name)
+		const st = statSync(full)
+		if (st.isDirectory()) yield* walk(full)
+		else yield full
+	}
+}
+
+describe('ADO CLI smoke', () => {
+	// The first `az repos|devops|boards` invocation on a runner without the azure-devops extension
+	// cached pays a one-time lazy-load cost that exceeds the per-test 5s budget. Warm it up here
+	// so each per-entry assertion measures the actual subcommand, not the extension bootstrap.
+	before(() => {
+		const result = spawnSync('az', ['repos', '--help'], { timeout: 30000, encoding: 'utf8' })
+		if (result.error && /** @type {NodeJS.ErrnoException} */ (result.error).code === 'ENOENT') {
+			// az not installed — per-entry tests will self-skip.
+			return
+		}
+	})
+
+	it('every `az` invocation in agents/, commands/, scripts/ is in the inventory', () => {
+		const uninventoried = findUninventoriedCommands({
+			sources: loadPluginSources(),
+			inventory: adoCliInventory,
+			allowlist: ADO_CLI_ALLOWLIST,
+		})
+		assert.deepEqual(
+			uninventoried,
+			[],
+			`Uninventoried az invocations:\n  ${uninventoried.join('\n  ')}\n\n` +
+				`Add each to tests/fixtures/ado-cli-inventory.mjs, or to the allowlist if it is a preflight / error-message hint.`
+		)
+	})
+
+	for (const entry of adoCliInventory) {
+		const label =
+			entry.kind === 'invoke'
+				? `${entry.command.join(' ')} --area ${entry.area} --resource ${entry.resource}`
+				: entry.command.join(' ')
+		it(`\`${label} --help\` exits 0 and contains required keywords (skipped if az is absent)`, (t) => {
+			const result = spawnSync(entry.command[0], [...entry.command.slice(1), '--help'], {
+				timeout: 5000,
+				encoding: 'utf8',
+			})
+			if (result.error && /** @type {NodeJS.ErrnoException} */ (result.error).code === 'ENOENT') {
+				t.skip('az CLI not installed')
+				return
+			}
+			assert.equal(result.status, 0, `Exit ${result.status} — stderr: ${result.stderr}`)
+			for (const keyword of entry.helpKeywordsRequired) {
+				assert.ok(
+					result.stdout.includes(keyword),
+					`\`${label} --help\` output is missing the required keyword: ${keyword}`
+				)
+			}
+		})
+	}
+})
diff --git a/apps/claude-code/pr-review/tests/ado-fetcher.test.mjs b/apps/claude-code/pr-review/tests/ado-fetcher.test.mjs
index 986f3034..2b1ab51f 100644
--- a/apps/claude-code/pr-review/tests/ado-fetcher.test.mjs
+++ b/apps/claude-code/pr-review/tests/ado-fetcher.test.mjs
@@ -105,4 +105,72 @@ describe('ado-fetcher agent content', () => {
 			'Agent must delegate work-item fetching to fetchWorkItems helper'
 		)
 	})
+
+	it('fetches PR threads via az devops invoke --resource pullRequestThreads', () => {
+		assert.ok(
+			/az devops invoke[\s\S]*?--resource pullRequestThreads/.test(agentContent),
+			'Fetcher must call az devops invoke with --resource pullRequestThreads (folded in from orchestrator)'
+		)
+	})
+
+	it('invokes the detectMode helper from scripts/mode-detection.mjs', () => {
+		assert.ok(
+			agentContent.includes('detectMode') && agentContent.includes('mode-detection'),
+			'Fetcher must invoke detectMode from scripts/mode-detection.mjs after fetching threads'
+		)
+	})
+
+	it('does not call az repos pr show — metadata is now passed in from the orchestrator', () => {
+		assert.ok(
+			!/az repos pr show/.test(agentContent),
+			'Fetcher must accept REPO_ID/PROJECT/branches/title/description as inputs, not call `az repos pr show` itself'
+		)
+	})
+
+	it('accepts REPO_ID, PROJECT, branch refs, PR_TITLE and PR_DESCRIPTION as literal-string inputs', () => {
+		const requiredInputs = ['REPO_ID', 'SOURCE_BRANCH', 'TARGET_BRANCH', 'PR_TITLE', 'PR_DESCRIPTION']
+		const inputsSection = agentContent.split('## Inputs')[1] ?? ''
+		for (const field of requiredInputs) {
+			assert.ok(inputsSection.includes(field), `Inputs section must list ${field}`)
+		}
+	})
+
+	it('emits RAW_THREADS_JSON, MODE, IS_REREVIEW, PRIOR_ITERATION_ID, SUMMARY_THREAD_ID in the result block', () => {
+		const required = ['RAW_THREADS_JSON', 'MODE', 'IS_REREVIEW', 'PRIOR_ITERATION_ID', 'SUMMARY_THREAD_ID']
+		const blockMatch = agentContent.match(/ADO_FETCHER_RESULT_START\r?\n([\s\S]*?)\r?\nADO_FETCHER_RESULT_END/)
+		assert.ok(blockMatch, 'Fetcher must contain an ADO_FETCHER_RESULT_START/END block')
+		const block = blockMatch[1]
+		for (const field of required) {
+			assert.ok(block.includes(field), `Result block must emit ${field} between ADO_FETCHER_RESULT_START and _END`)
+		}
+	})
+
+	it('aborts with `az devops login` hint when threads endpoint returns 401/403', () => {
+		assert.ok(
+			/thread.*?(401|403|auth)[\s\S]*?az devops login/i.test(agentContent) ||
+				/(401|403|auth)[\s\S]*?az devops login[\s\S]*?thread/i.test(agentContent),
+			'Fetcher must abort on threads 401/403 with an az devops login hint'
+		)
+	})
+
+	it('treats 404 on the threads endpoint as empty threads (value:[])', () => {
+		assert.ok(
+			/404[\s\S]*?\{"value":\s*\[\]\}/.test(agentContent) || /\{"value":\s*\[\]\}[\s\S]*?404/.test(agentContent),
+			'Fetcher must treat 404 on threads as RAW_THREADS_JSON={"value":[]}'
+		)
+	})
+
+	it('emits a warning Notice with kind: thread-fetch on 5xx / network failure', () => {
+		assert.ok(
+			agentContent.includes('thread-fetch'),
+			'Fetcher must emit a warning Notice with kind: thread-fetch on 5xx / network on the threads endpoint'
+		)
+	})
+
+	it('applies HTTP-tier classification via classify-http-error helper', () => {
+		assert.ok(
+			agentContent.includes('classify-http-error') || agentContent.includes('classifyHttpError'),
+			'Fetcher must use the classifyHttpError helper from scripts/ado/classify-http-error.mjs for the threads endpoint'
+		)
+	})
 })
diff --git a/apps/claude-code/pr-review/tests/fixtures/ado-cli-allowlist.mjs b/apps/claude-code/pr-review/tests/fixtures/ado-cli-allowlist.mjs
new file mode 100644
index 00000000..55b8870f
--- /dev/null
+++ b/apps/claude-code/pr-review/tests/fixtures/ado-cli-allowlist.mjs
@@ -0,0 +1,28 @@
+// @ts-check
+
+/**
+ * Regex allowlist for `az` invocations intentionally NOT in `ado-cli-inventory.mjs`.
+ *
+ * Entries fall in two buckets:
+ * - **Preflight calls** — the orchestrator probes the CLI itself, not an ADO API,
+ *   so inventory-style coverage is overkill.
+ * - **Error-message hints** — strings the plugin prints to nudge the user; they
+ *   never run via the plugin, only via the human reading the error.
+ *
+ * Add a comment explaining the exemption alongside every entry.
+ */
+
+export const ADO_CLI_ALLOWLIST = [
+	// Preflight in commands/review-pr.md Step 3 — Azure CLI version check.
+	/^az --version$/,
+	// Preflight — confirms the azure-devops extension is installed.
+	/^az extension list$/,
+	// Preflight (added by Slice 02) — confirms `az devops invoke` is callable.
+	/^az devops invoke --area <missing> --resource <missing>$/,
+	// Error-message hint — never invoked by the plugin; suggested to the user.
+	/^az devops login$/,
+	// Error-message hint — recovery instruction when the extension is missing.
+	/^az extension add$/,
+	// Error-message hint — recovery instruction when `az devops invoke` is broken.
+	/^az extension remove$/,
+]
diff --git a/apps/claude-code/pr-review/tests/fixtures/ado-cli-inventory.mjs b/apps/claude-code/pr-review/tests/fixtures/ado-cli-inventory.mjs
new file mode 100644
index 00000000..e489b40a
--- /dev/null
+++ b/apps/claude-code/pr-review/tests/fixtures/ado-cli-inventory.mjs
@@ -0,0 +1,89 @@
+// @ts-check
+
+/**
+ * Single source of truth for every `az` command this plugin actually invokes.
+ *
+ * The smoke test (`tests/ado-cli-smoke.test.mjs`) does two things with this list:
+ * 1. Asserts every `az` shape found in `agents/`, `commands/`, and `scripts/` is
+ *    listed here (modulo `ado-cli-allowlist.mjs`).
+ * 2. Runs `<command> --help` against each entry to surface ADO CLI renames or
+ *    removals (the Step 4 bug class — `az repos pr thread list` was a phantom).
+ *
+ * Register a new ADO call here before invoking it.
+ *
+ * @typedef {{
+ *   kind: 'invoke',
+ *   area: string,
+ *   resource: string,
+ *   command: string[],
+ *   helpKeywordsRequired: string[],
+ * }} InvokeEntry
+ * @typedef {{
+ *   kind: 'repos' | 'boards',
+ *   command: string[],
+ *   helpKeywordsRequired: string[],
+ * }} SubcommandEntry
+ * @typedef {InvokeEntry | SubcommandEntry} InventoryEntry
+ */
+
+/** @type {InventoryEntry[]} */
+export const adoCliInventory = [
+	// Orchestrator Step 4 — PR metadata.
+	{
+		kind: 'repos',
+		command: ['az', 'repos', 'pr', 'show'],
+		helpKeywordsRequired: ['--id', '--org'],
+	},
+	// ADO Fetcher Step 4 — checkout PR source branch.
+	{
+		kind: 'repos',
+		command: ['az', 'repos', 'pr', 'checkout'],
+		helpKeywordsRequired: ['--id'],
+	},
+	// Doc Context Orchestrator — fetch linked work item.
+	{
+		kind: 'boards',
+		command: ['az', 'boards', 'work-item', 'show'],
+		helpKeywordsRequired: ['--id'],
+	},
+	// ADO Fetcher Step 1 — list iterations.
+	{
+		kind: 'invoke',
+		area: 'git',
+		resource: 'pullRequestIterations',
+		command: ['az', 'devops', 'invoke'],
+		helpKeywordsRequired: ['--area', '--resource'],
+	},
+	// ADO Fetcher Step 2 — fetch PR threads (introduced in Slice 01).
+	{
+		kind: 'invoke',
+		area: 'git',
+		resource: 'pullRequestThreads',
+		command: ['az', 'devops', 'invoke'],
+		helpKeywordsRequired: ['--area', '--resource'],
+	},
+	// ADO Fetcher Step 3 — list changed files.
+	{
+		kind: 'invoke',
+		area: 'git',
+		resource: 'pullRequestIterationChanges',
+		command: ['az', 'devops', 'invoke'],
+		helpKeywordsRequired: ['--area', '--resource'],
+	},
+	// ADO Fetcher Step 5 — list linked work items.
+	{
+		kind: 'invoke',
+		area: 'git',
+		resource: 'pullRequestWorkItems',
+		command: ['az', 'devops', 'invoke'],
+		helpKeywordsRequired: ['--area', '--resource'],
+	},
+	// ADO Writer + Re-review Coordinator — post replies to threads.
+	{
+		kind: 'invoke',
+		area: 'git',
+		resource: 'pullRequestThreadComments',
+		command: ['az', 'devops', 'invoke'],
+		helpKeywordsRequired: ['--area', '--resource'],
+	},
+]
diff --git a/apps/claude-code/pr-review/tests/mode-detection.test.mjs b/apps/claude-code/pr-review/tests/mode-detection.test.mjs
index aef8671f..641383d2 100644
--- a/apps/claude-code/pr-review/tests/mode-detection.test.mjs
+++ b/apps/claude-code/pr-review/tests/mode-detection.test.mjs
@@ -2,9 +2,13 @@
 
 import assert from 'node:assert/strict'
 import { describe, it } from 'node:test'
-import { detectMode, formatModeEnv } from '../scripts/mode-detection.mjs'
+import { detectMode, formatModeEnv, SIGNATURE_PREFIX } from '../scripts/mode-detection.mjs'
 
-const SIGNATURE_PREFIX = '🤖 *Reviewed by Claude Code*'
+describe('SIGNATURE_PREFIX', () => {
+	it('exports the canonical bot-signature prefix verbatim', () => {
+		assert.equal(SIGNATURE_PREFIX, '🤖 *Reviewed by Claude Code*')
+	})
+})
 
 describe('detectMode', () => {
 	it('no threads → first-review with empty fields', () => {
diff --git a/apps/claude-code/pr-review/tests/notices.test.mjs b/apps/claude-code/pr-review/tests/notices.test.mjs
index 299878c3..b9b7ea96 100644
--- a/apps/claude-code/pr-review/tests/notices.test.mjs
+++ b/apps/claude-code/pr-review/tests/notices.test.mjs
@@ -15,6 +15,16 @@ describe('createNotice', () => {
 		const n = createNotice('info', 'doc-context', 'hello')
 		assert.deepEqual(n, { severity: 'info', kind: 'doc-context', message: 'hello' })
 	})
+
+	it('accepts the thread-fetch kind (Fetcher emits this on 5xx/network)', () => {
+		const n = createNotice(
+			'warning',
+			'thread-fetch',
+			'Threads endpoint degraded — proceeding without prior-review context.'
+		)
+		assert.equal(n.kind, 'thread-fetch')
+		assert.equal(n.severity, 'warning')
+	})
 })
 
 describe('mergeNotices', () => {
diff --git a/apps/claude-code/pr-review/tests/plugin-structure.test.mjs b/apps/claude-code/pr-review/tests/plugin-structure.test.mjs
index e08f6057..4f0b39cb 100644
--- a/apps/claude-code/pr-review/tests/plugin-structure.test.mjs
+++ b/apps/claude-code/pr-review/tests/plugin-structure.test.mjs
@@ -42,9 +42,9 @@ describe('plugin structure', () => {
 	it('each agent file has a name: field in frontmatter matching its filename stem', () => {
 		for (const agent of EXPECTED_AGENTS) {
 			const content = readFileSync(join(AGENTS_DIR, `${agent}.md`), 'utf8')
-			const match = content.match(/^---\n([\s\S]*?)\n---/)
+			const match = content.match(/^---\r?\n([\s\S]*?)\r?\n---/)
 			assert.ok(match, `${agent}.md: missing YAML frontmatter`)
-			const frontmatter = match[1]
+			const frontmatter = match[1].replace(/\r\n/g, '\n')
 			assert.ok(
 				new RegExp(`^name: ${agent}$`, 'm').test(frontmatter),
 				`${agent}.md: frontmatter missing exact line "name: ${agent}"`
diff --git a/apps/claude-code/pr-review/tests/pre-pr.test.mjs b/apps/claude-code/pr-review/tests/pre-pr.test.mjs
index 99ce23f3..d94c4163 100644
--- a/apps/claude-code/pr-review/tests/pre-pr.test.mjs
+++ b/apps/claude-code/pr-review/tests/pre-pr.test.mjs
@@ -389,3 +389,74 @@ describe('review-pr command — Pre-PR mode', () => {
 		)
 	})
 })
+
+// ---------------------------------------------------------------------------
+// review-pr.md command content — Step 4 / Step 5 (post-fold-in of thread fetch)
+// ---------------------------------------------------------------------------
+
+describe('review-pr command — Step 4 (PR metadata only) + Fetcher delegation', () => {
+	it('no longer calls the non-existent `az repos pr thread` subcommand', () => {
+		// Built via concatenation so the test file itself does not contain the
+		// forbidden phrase (the spec's grep sweep treats any occurrence as a fail).
+		const forbidden = ['az', 'repos', 'pr', 'thread', 'list'].join(' ')
+		assert.ok(
+			!commandContent.includes(forbidden),
+			`Orchestrator must not call the non-existent ${forbidden} subcommand`
+		)
+	})
+
+	it('contains no `az devops invoke` REST calls (Fetcher owns all REST calls — `--help` probe is allowed)', () => {
+		// Strip markdown inline code (`...`) and quoted strings ("...", '...') before
+		// scanning each line — prose references and echoed error messages don't
+		// execute the command, only document or surface it.
+		const stripPassive = (line) =>
+			line
+				.replace(/`[^`]*`/g, '')
+				.replace(/"[^"]*"/g, '')
+				.replace(/'[^']*'/g, '')
+		const offending = commandContent.split('\n').filter(
+			(line) =>
+				/az devops invoke/.test(stripPassive(line)) &&
+				!line.trim().startsWith('-') &&
+				// The Step 3 preflight uses `az devops invoke --help` purely as a CLI capability
+				// probe — it makes no REST call, so ADR-0016 permits it.
+				!/az devops invoke --help/.test(line)
+		)
+		assert.deepEqual(offending, [], `Orchestrator must not call az devops invoke directly: ${offending.join(' | ')}`)
+	})
+
+	it('Step 4 captures REPO_ID, PROJECT, SOURCE_BRANCH, TARGET_BRANCH, PR_TITLE, PR_DESCRIPTION from `az repos pr show`', () => {
+		const step4 = commandContent.slice(commandContent.indexOf('## Step 4'), commandContent.indexOf('## Step 5'))
+		assert.ok(/az repos pr show/.test(step4), 'Step 4 must call az repos pr show to gather PR metadata')
+		for (const field of ['REPO_ID', 'PROJECT', 'SOURCE_BRANCH', 'TARGET_BRANCH', 'PR_TITLE', 'PR_DESCRIPTION']) {
+			assert.ok(step4.includes(field), `Step 4 must capture ${field} from the PR metadata response`)
+		}
+	})
+
+	it('Step 4 strips `refs/heads/` from branch refs', () => {
+		const step4 = commandContent.slice(commandContent.indexOf('## Step 4'), commandContent.indexOf('## Step 5'))
+		assert.ok(
+			step4.includes('refs/heads/') || step4.toLowerCase().includes('strip'),
+			'Step 4 must strip the refs/heads/ prefix when deriving SOURCE_BRANCH / TARGET_BRANCH'
+		)
+	})
+
+	it('passes the new metadata into the Fetcher prompt (Step 5)', () => {
+		const step5 = commandContent.slice(commandContent.indexOf('## Step 5'), commandContent.indexOf('## Step 6'))
+		for (const field of ['REPO_ID', 'SOURCE_BRANCH', 'TARGET_BRANCH', 'PR_TITLE', 'PR_DESCRIPTION']) {
+			assert.ok(step5.includes(field), `Step 5 Fetcher prompt must forward ${field} to the ADO Fetcher agent`)
+		}
+	})
+
+	it('Step 5 parses MODE, IS_REREVIEW, PRIOR_ITERATION_ID, SUMMARY_THREAD_ID, RAW_THREADS_JSON from the Fetcher result block', () => {
+		const step5 = commandContent.slice(commandContent.indexOf('## Step 5'), commandContent.indexOf('## Step 6'))
+		for (const field of ['MODE', 'IS_REREVIEW', 'PRIOR_ITERATION_ID', 'SUMMARY_THREAD_ID', 'RAW_THREADS_JSON']) {
+			assert.ok(step5.includes(field), `Step 5 must parse ${field} from the Fetcher's ADO_FETCHER_RESULT block`)
+		}
+	})
+
+	it('orchestrator file stays ≤ 200 lines (ADR 0013 thin-orchestrator invariant)', () => {
+		const lineCount = commandContent.split('\n').length
+		assert.ok(lineCount <= 200, `Orchestrator is ${lineCount} lines; ADR 0013 caps it at ≤ 200`)
+	})
+})
diff --git a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-build.md b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-build.md
index 917bee94..95ef3f38 100644
--- a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-build.md
+++ b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-build.md
@@ -55,5 +55,5 @@ giving the fastest possible end-to-end build time.
 ## Runs
 
 ```
-archon run .archon/workflows/build.yaml --input slug=<slug>
+archon workflow run unic-dlc-build --input slug=<slug>
 ```
diff --git a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-cleanup.md b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-cleanup.md
index 86d98e32..36497ac2 100644
--- a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-cleanup.md
+++ b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-cleanup.md
@@ -34,9 +34,10 @@ Where `<slug>` is the session identifier used throughout the plan → build →
    Each ADR is shown with Context, Decision, and Consequences. The user accepts (A),
    rejects (R), or edits (E) each candidate. Only accepted ADRs are written to `docs/adr/`.
 
-3. **`run-triage` node** — invokes the shared `.archon/workflows/triage.yaml` directly via
-   `archon run`. No triage logic is duplicated in the cleanup workflow. This produces
-   `HANDOFF.md` and updates `docs/workflow/ROADMAP.md` exactly once per cleanup run.
+3. **`run-triage` node** — invokes the shared triage workflow by name via
+   `archon workflow run unic-dlc-triage`. No triage logic is duplicated in the cleanup
+   workflow. This produces `HANDOFF.md` and updates `docs/workflow/ROADMAP.md` exactly
+   once per cleanup run.
 
 ## Prerequisites
 
@@ -56,5 +57,5 @@ Where `<slug>` is the session identifier used throughout the plan → build →
 ## Runs
 
 ```
-archon run .archon/workflows/cleanup.yaml --input slug=<slug>
+archon workflow run unic-dlc-cleanup --input slug=<slug>
 ```
diff --git a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-explore.md b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-explore.md
index 10a1c3ff..29ba263e 100644
--- a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-explore.md
+++ b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-explore.md
@@ -25,7 +25,7 @@ Runs the `explore` workflow: four parallel research agents investigate the proje
 ## Usage
 
 ```sh
-archon run .archon/workflows/explore.yaml --input slug=<slug>
+archon workflow run unic-dlc-explore --input slug=<slug>
 ```
 
 Or invoke from Claude Code:
diff --git a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-plan.md b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-plan.md
index 3c55e8d4..a659f25e 100644
--- a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-plan.md
+++ b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-plan.md
@@ -21,7 +21,7 @@ Runs the `plan` workflow: loads project context and prior research, runs an adve
 ## Usage
 
 ```sh
-archon run .archon/workflows/plan.yaml --input slug=<slug>
+archon workflow run unic-dlc-plan --input slug=<slug>
 ```
 
 Or invoke from Claude Code:
diff --git a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-qa.md b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-qa.md
index 49b62d12..f44bfbee 100644
--- a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-qa.md
+++ b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-qa.md
@@ -23,7 +23,7 @@ a UAT checklist derived from the PRD's acceptance criteria, and merges the PR on
 ## Usage
 
 ```sh
-archon run .archon/workflows/qa.yaml --input slug=<slug>
+archon workflow run unic-dlc-qa --input slug=<slug>
 ```
 
 Or invoke from Claude Code:
@@ -74,5 +74,5 @@ The `uat-gate` node pauses and asks for your decision:
 ## Runs
 
 ```
-archon run .archon/workflows/qa.yaml --input slug=<slug>
+archon workflow run unic-dlc-qa --input slug=<slug>
 ```
diff --git a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-review.md b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-review.md
index 9fe27e84..3f016f28 100644
--- a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-review.md
+++ b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-review.md
@@ -9,7 +9,7 @@ Runs the `review` workflow: analyses the current PR diff across four review aspe
 single structured comment on the PR.
 
 **No runtime dependency on pr-review-toolkit or any other plugin.**
-All review logic is self-contained in `.archon/workflows/review.yaml`.
+All review logic is self-contained in `.archon/workflows/unic-dlc-review.yaml`.
 
 ## When to use
 
@@ -36,7 +36,7 @@ Each section lists specific findings with `file:line` references, or an explicit
 ## Usage
 
 ```sh
-archon run .archon/workflows/review.yaml
+archon workflow run unic-dlc-review
 ```
 
 Or invoke from Claude Code:
diff --git a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-triage.md b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-triage.md
index e30554bd..e0eb8ad2 100644
--- a/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-triage.md
+++ b/apps/claude-code/unic-archon-dlc/.archon/commands/unic-dlc-triage.md
@@ -20,7 +20,7 @@ Runs the `triage` workflow: reads current issue states from the configured track
 ## Usage
 
 ```sh
-archon run .archon/workflows/triage.yaml
+archon workflow run unic-dlc-triage
 ```
 
 Or invoke from Claude Code:
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-build.yaml b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-build.yaml
new file mode 100644
index 00000000..37bc35fc
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-build.yaml
@@ -0,0 +1,365 @@
+name: unic-dlc-build
+description: >
+  TDD build execution wrapper: slopcheck gate → generated per-slug build workflow.
+  Each issue goes red (failing tests) before green (passing implementation).
+  Independent issues run in parallel; chained issues run in dependency order.
+
+inputs:
+  slug:
+    description: Planning session identifier — matches the slug used in /unic-dlc-plan.
+    required: true
+
+nodes:
+  - id: slopcheck
+    name: Slopcheck — Verify New Packages
+    type: bash
+    runtime: bun
+    depends_on: []
+    script: |
+      #!/usr/bin/env node
+      // @ts-check
+      import { execFileSync } from 'node:child_process'
+      import { existsSync, readFileSync } from 'node:fs'
+      import { join } from 'node:path'
+      import { classifyPackages, parseNewPackages } from './lib/slopcheck.mjs'
+
+      const projectDir = process.cwd()
+      const pkgPath = join(projectDir, 'package.json')
+
+      if (!existsSync(pkgPath)) {
+        console.log('slopcheck: no package.json found, skipping.')
+        process.exit(0)
+      }
+
+      // Get the previous package.json from HEAD (if in a git repo)
+      let prevDeps = {}
+      try {
+        const prev = execFileSync('git', ['show', 'HEAD:package.json'], { cwd: projectDir, encoding: 'utf8' })
+        const parsed = JSON.parse(prev)
+        prevDeps = { ...parsed.dependencies, ...parsed.devDependencies }
+      } catch {
+        // Not in a git repo or no HEAD — treat all current packages as new (strictest)
+        console.log('slopcheck: cannot read HEAD:package.json, treating all packages as new.')
+      }
+
+      const current = JSON.parse(readFileSync(pkgPath, 'utf8'))
+      const currentDeps = { ...current.dependencies, ...current.devDependencies }
+      const newPkgs = parseNewPackages(prevDeps, currentDeps)
+
+      if (newPkgs.length === 0) {
+        console.log('slopcheck: no new packages detected. ✓')
+        process.exit(0)
+      }
+
+      console.log(`slopcheck: found ${newPkgs.length} new package(s): ${newPkgs.join(', ')}`)
+
+      // Determine registry check strategy
+      let registryFn = null
+
+      // Strategy 1: Python slopcheck tool
+      try {
+        execFileSync('slopcheck', ['--version'], { encoding: 'utf8' })
+        registryFn = async (/** @type {string} */ name) => {
+          try {
+            execFileSync('slopcheck', ['check', name], { encoding: 'utf8' })
+            return true
+          } catch {
+            return false
+          }
+        }
+        console.log('slopcheck: using Python slopcheck tool.')
+      } catch {
+        // Python tool not on PATH, try npm registry HEAD check
+        try {
+          const { default: https } = await import('node:https')
+          registryFn = (/** @type {string} */ name) =>
+            new Promise((resolve) => {
+              const encoded = encodeURIComponent(name).replace(/^%40/, '@')
+              const req = https.request(
+                { hostname: 'registry.npmjs.org', path: `/${encoded}`, method: 'HEAD' },
+                (res) => resolve(res.statusCode === 200),
+              )
+              req.on('error', () => resolve(false))
+              req.setTimeout(5000, () => {
+                req.destroy()
+                resolve(false)
+              })
+              req.end()
+            })
+          console.log('slopcheck: using npm registry HEAD check fallback.')
+        } catch {
+          console.log('slopcheck: no registry check available — all new packages treated as [ASSUMED].')
+        }
+      }
+
+      const verdicts = await classifyPackages(newPkgs, registryFn)
+      const assumed = verdicts.filter((v) => v.assumed)
+
+      if (assumed.length === 0) {
+        console.log('slopcheck: all new packages verified in registry. ✓')
+        process.exit(0)
+      }
+
+      console.log('\nslopcheck: [ASSUMED] packages require human verification:')
+      for (const pkg of assumed) {
+        console.log(`  ⚠  ${pkg.name} — not found in registry or unverifiable`)
+        console.log(`     checkpoint:human-verify — approve or reject this package before install continues`)
+      }
+      console.log(
+        '\nslopcheck: halting build. Resolve the above packages (remove them, pin a real version,',
+      )
+      console.log('or run the build with SLOPCHECK_BYPASS=1 to accept assumed packages explicitly.')
+
+      if (!process.env.SLOPCHECK_BYPASS) {
+        process.exit(1)
+      }
+
+  - id: run-build
+    name: Execute Generated Build Workflow
+    type: prompt
+    depends_on: [slopcheck]
+    prompt: |
+      You are running the unic-archon-dlc build workflow — run-build node.
+
+      The slopcheck gate has passed. Now execute the generated build workflow for this slug.
+
+      The generated workflow is at: .archon/workflows/build-{{ inputs.slug }}.yaml
+
+      Actions:
+      1. Verify that .archon/workflows/build-{{ inputs.slug }}.yaml exists.
+         If it does not exist, print:
+           "build-{{ inputs.slug }}.yaml not found. Run /unic-dlc-plan {{ inputs.slug }} first to generate it."
+         and halt.
+
+      2. Invoke the generated workflow via Archon by name:
+         Run: archon workflow run unic-dlc-build-{{ inputs.slug }} --input slug={{ inputs.slug }}
+
+         The generated workflow contains:
+           - code-red-<id> nodes: write FAILING acceptance tests per issue
+           - code-green-<id> nodes: write minimum implementation to pass those tests
+           - Dependency edges enforce: red before green within each issue
+           - Independent issues run code-red (and code-green) in parallel
+
+      3. After the workflow completes, print:
+         "build-{{ inputs.slug }} complete. All issues have been implemented following red→green discipline."
+
+      TDD rules enforced by the generated DAG:
+        - Never write implementation before writing failing tests.
+        - code-green nodes write ONLY enough code to pass the current tests.
+        - No speculative features, no extra abstractions.
+        - If a code-green node fails (tests still failing after implementation), pause and
+          ask the user whether to retry or flag the issue for human review.
+
+  - id: verification
+    name: Verification — Test Suite, Stubs, Wiring
+    type: bash
+    runtime: bun
+    depends_on: [run-build]
+    script: |
+      #!/usr/bin/env node
+      // @ts-check
+      import { readFileSync, readdirSync, statSync } from 'node:fs'
+      import { join, extname } from 'node:path'
+      import { execFileSync } from 'node:child_process'
+      import { detectStubs } from './lib/stub-detector.mjs'
+
+      const slug = process.env.ARCHON_INPUT_SLUG
+      if (!slug) {
+        console.error('verification: ARCHON_INPUT_SLUG is not set')
+        process.exit(1)
+      }
+
+      let exitCode = 0
+
+      // 1. Run the test suite
+      console.log('\n=== verification: running test suite ===')
+      try {
+        execFileSync('pnpm', ['test'], { stdio: 'inherit' })
+        console.log('verification: tests PASSED ✓')
+      } catch {
+        console.error('verification: tests FAILED ✗')
+        exitCode = 1
+      }
+
+      // 2. Detect stub patterns in changed source files
+      console.log('\n=== verification: checking for stub patterns ===')
+      let changedFiles = []
+      try {
+        const diff = execFileSync('git', ['diff', '--name-only', 'HEAD'], { encoding: 'utf8' })
+        changedFiles = diff.split('\n').filter(Boolean).filter(f => /\.(mjs|js|ts|py|rb|go|rs)$/.test(f))
+      } catch {
+        console.log('verification: cannot read git diff, skipping stub check.')
+      }
+
+      const stubFindings = []
+      for (const file of changedFiles) {
+        try {
+          const src = readFileSync(file, 'utf8')
+          const stubs = detectStubs(src)
+          for (const s of stubs) stubFindings.push({ file, ...s })
+        } catch { /* file deleted or unreadable */ }
+      }
+
+      if (stubFindings.length > 0) {
+        console.error(`verification: found ${stubFindings.length} stub pattern(s):`)
+        for (const f of stubFindings) {
+          console.error(`  ${f.file}:${f.line}  [${f.pattern}]  ${f.text}`)
+        }
+        exitCode = 1
+      } else {
+        console.log('verification: no stub patterns detected ✓')
+      }
+
+      // 3. Check coverage threshold
+      const configPath = '.archon/unic-dlc.config.json'
+      let coverageThreshold = null
+      try {
+        const cfg = JSON.parse(readFileSync(configPath, 'utf8'))
+        coverageThreshold = cfg.coverage_threshold ?? null
+      } catch { /* config absent */ }
+
+      if (coverageThreshold !== null) {
+        console.log(`\n=== verification: checking coverage threshold (${coverageThreshold}%) ===`)
+        try {
+          execFileSync('pnpm', ['test', '--coverage'], { stdio: 'inherit' })
+        } catch {
+          console.error(`verification: coverage check failed (threshold: ${coverageThreshold}%)`)
+          exitCode = 1
+        }
+      }
+
+      process.exit(exitCode)
+
+  - id: goals-check
+    name: Goals-Check — PRD Coverage Matrix
+    type: prompt
+    depends_on: [verification]
+    prompt: |
+      You are running the unic-archon-dlc build workflow — goals-check node.
+
+      Goal: produce a coverage matrix that maps every acceptance criterion from the PRD
+      to implementation evidence (file paths and test names). Any criterion without evidence
+      is "MISSING" and fails this node.
+
+      ## Step 1 — Load the PRD and issues
+
+      Read docs/workflow/{{ inputs.slug }}/PRD.md in full.
+      Read docs/workflow/{{ inputs.slug }}/issues.json and parse the issues array.
+
+      Extract every acceptance criterion from:
+        - Each user story in the PRD's "User Stories" section (format: "As a ... I want ...")
+        - Each issue's `acceptance_criteria` array in issues.json
+
+      Deduplicate criteria that appear in both (match by meaning, not exact text).
+
+      ## Step 2 — Find implementation evidence
+
+      For each criterion, search the codebase for:
+        - Source files that implement the described behaviour (file paths with function/class names)
+        - Test files that assert this criterion passes (test names, file:line references)
+
+      Use your knowledge of the changed files from the just-completed build to focus the search.
+
+      ## Step 3 — Build the coverage matrix
+
+      Format the matrix as a Markdown table:
+
+        | Criterion | Evidence | Status |
+        |-----------|----------|--------|
+        | <criterion text> | <file:line or test name> | ✓ COVERED / ✗ MISSING |
+
+      Rules:
+        - A criterion is COVERED if at least one test asserts it AND at least one source file
+          implements it.
+        - A criterion is MISSING if either the test or the implementation is absent.
+
+      ## Step 4 — Report and fail if any MISSING
+
+      If any rows are MISSING:
+        Print: "goals-check: FAILED — N criterion/criteria without coverage."
+        List the missing criteria.
+        The workflow will return control to `verification` after the report step.
+
+      If all criteria are COVERED:
+        Print: "goals-check: PASSED — all acceptance criteria covered. ✓"
+
+      Store the matrix as `{{ matrix }}` for use in the report node.
+
+  - id: report
+    name: Report — Consolidate Build Outcomes
+    type: prompt
+    depends_on: [goals-check]
+    prompt: |
+      You are running the unic-archon-dlc build workflow — report node.
+
+      Goal: write docs/workflow/{{ inputs.slug }}/report.md consolidating all build outcomes.
+
+      ## Required sections (exactly five)
+
+      ### 1. What Was Built
+      A high-level summary (3-5 bullets) of the issues that were implemented in this build cycle.
+      Reference each issue id and title.
+
+      ### 2. Goals-Check Coverage Matrix
+      Paste the coverage matrix produced by the goals-check node verbatim.
+
+      ### 3. Test Outcomes
+      - Pass/fail counts from the test run.
+      - Coverage numbers if available.
+      - Any flaky or skipped tests.
+
+      ### 4. Decisions Made
+      Bullet list of technical decisions made during implementation that should be recorded.
+      For each decision, propose an ADR filename (docs/adr/NNNN-<slug>.md) if the decision
+      is significant enough to warrant one. Write the ADR inline for review.
+
+      ### 5. Tech Debt Flagged
+      Bullet list of items deferred during this build (stub patterns that were accepted,
+      known limitations, shortcuts taken under time pressure). These items feed into
+      the cleanup workflow (slice 13).
+
+      ## Output
+
+      Write the report to docs/workflow/{{ inputs.slug }}/report.md.
+      Print: "report written to docs/workflow/{{ inputs.slug }}/report.md"
+
+      If goals-check returned any MISSING criteria, include a note in the "What Was Built"
+      section: "⚠ goals-check found N uncovered criteria — see matrix below."
+
+  - id: build-pr-gate
+    name: Build PR Gate — Human Review
+    type: interactive
+    fresh_context: true
+    depends_on: [report]
+    prompt: |
+      You are running the unic-archon-dlc build workflow — build-pr-gate node.
+
+      The implementation, goals-check matrix, and report are ready for human review.
+
+      Actions:
+      1. Stage all changed source and test files plus the report:
+         - All files changed during the build (git status --short)
+         - docs/workflow/{{ inputs.slug }}/report.md
+         - Any ADR files proposed in the report (docs/adr/NNNN-*.md that are new)
+      2. Open a PR targeting the develop branch with:
+         - Title: "build({{ inputs.slug }}): implementation"
+         - Body:
+           ## Summary
+           - TDD implementation for `{{ inputs.slug }}` session
+           - Goals-check matrix: all criteria covered (or: N criteria missing — see report)
+           - report.md attached
+
+           ## Review Checklist
+           - [ ] All acceptance criteria have implementation evidence
+           - [ ] No TODO / FIXME stubs left in implementation code
+           - [ ] Test coverage meets configured threshold
+           - [ ] Tech debt section in report.md is accurate
+           - [ ] Proposed ADRs (if any) are ready to record
+      3. Pause and wait for human review.
+         - If the PR is **approved** and merged:
+           Print: "Build PR gate passed."
+           Exit.
+         - If the PR is **rejected** or changes are requested:
+           Print: "Build PR gate: changes requested. Returning to verification."
+           Return control to the verification node for another pass.
+           Do NOT restart the full build (only re-verify and re-report).
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-cleanup.yaml b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-cleanup.yaml
new file mode 100644
index 00000000..7cbc946e
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-cleanup.yaml
@@ -0,0 +1,158 @@
+name: unic-dlc-cleanup
+description: >
+  Post-merge cleanup: architecture review, ADR consolidation, then triage rerun.
+  Catches technical and intent drift, locks architectural decisions, and refreshes
+  project state via the shared triage workflow.
+
+inputs:
+  slug:
+    description: Planning session identifier — matches the slug from the completed build cycle.
+    required: true
+
+nodes:
+  - id: arch-review
+    name: Architecture Review — Drift Detection
+    type: prompt
+    depends_on: []
+    prompt: |
+      You are running the unic-archon-dlc cleanup workflow — arch-review node.
+
+      Goal: compare what was intended (PRD) against what was shipped (report + code) and
+      surface any drift before the project moves on.
+
+      ## Step 1 — Load context
+
+      Read the following files in full:
+        - docs/workflow/{{ inputs.slug }}/PRD.md         (intent)
+        - docs/workflow/{{ inputs.slug }}/report.md      (technical outcome)
+        - docs/workflow/{{ inputs.slug }}/issues.json    (issue breakdown)
+
+      Also read the list of changed files from the build:
+        Run: git diff --name-only HEAD~1 HEAD
+        (or adapt the range if HEAD~1 is not the merge commit)
+
+      For each changed source file, read its content.
+
+      ## Step 2 — Detect technical drift
+
+      Review the shipped code against architectural best practices:
+
+        - **Too-shallow modules**: functions or classes that are mere pass-throughs with no
+          real logic (a deep module hides complexity behind a small interface — the inverse is
+          the anti-pattern here).
+        - **Tight coupling**: direct imports of internal implementation details across module
+          boundaries; callers that know about the internals of their dependencies.
+        - **Leaky abstractions**: functions that expose internal data structures, require
+          callers to know about implementation order, or have error modes that only make sense
+          if you understand the internals.
+
+      For each issue found, note: file path, line(s), problem description, suggested fix.
+
+      ## Step 3 — Detect intent drift
+
+      Compare the PRD's user stories and acceptance criteria against the implementation:
+
+        - Delivered behaviour that diverges from what the PRD describes (not the same as tech
+          debt — this is wrong behaviour, not merely imperfect code).
+        - Acceptance criteria from the PRD that were silently dropped or narrowed during build.
+        - Features added during build that were not in the PRD (scope creep).
+
+      For each issue found, note: PRD section referenced, what was described vs what was
+      delivered, recommended action (fix now / create follow-up issue / accept as ADR).
+
+      ## Step 4 — Identify deepening opportunities
+
+      List at least one module that could be deepened to hide more complexity behind its
+      current interface — or explicitly state "No deepening opportunities identified."
+
+      ## Step 5 — Write arch-review.md
+
+      Write docs/workflow/{{ inputs.slug }}/arch-review.md with the following sections:
+
+        ## Architecture Review — {{ inputs.slug }}
+
+        ### Technical Drift
+        <findings or "No technical drift detected.">
+
+        ### Intent Drift
+        <findings or "No intent drift detected.">
+
+        ### Deepening Opportunities
+        <at least one suggestion, or "None identified.">
+
+        ### Summary
+        CLEAN | ISSUES FOUND (with count)
+
+      Print: "arch-review written to docs/workflow/{{ inputs.slug }}/arch-review.md"
+
+  - id: adr-consolidation
+    name: ADR Consolidation — Per-ADR Approval Gate
+    type: interactive
+    fresh_context: true
+    depends_on: [arch-review]
+    prompt: |
+      You are running the unic-archon-dlc cleanup workflow — adr-consolidation node.
+
+      Goal: review every architectural decision recorded during the build cycle and get
+      explicit human approval before writing any ADR to docs/adr/.
+
+      ## Step 1 — Collect ADR candidates
+
+      Gather decisions from:
+        1. `docs/workflow/{{ inputs.slug }}/report.md` — "Decisions Made" section
+        2. `docs/workflow/{{ inputs.slug }}/arch-review.md` — any decisions surfaced during review
+           (e.g. "accept as ADR" items from intent drift findings)
+        3. Any inline ADR drafts written by the build workflow nodes
+
+      For each candidate, present it in this format:
+
+        ---
+        ADR Candidate N of M
+        Proposed file: docs/adr/NNNN-<slug>.md
+        Status: [Proposed]
+        Context: <one paragraph — what situation prompted this decision>
+        Decision: <one sentence — what was decided>
+        Consequences: <bullet list — trade-offs accepted>
+        ---
+
+        Accept (A) / Reject (R) / Edit (E)?
+
+      ## Step 2 — Gate each ADR individually
+
+      Wait for the user to respond with A, R, or E for each candidate:
+        - **A (Accept)**: write the ADR to docs/adr/NNNN-<slug>.md using the next
+          available NNNN (check existing files in docs/adr/).
+          Print: "ADR written: docs/adr/NNNN-<slug>.md"
+        - **R (Reject)**: skip this candidate.
+          Print: "ADR skipped: <title>"
+        - **E (Edit)**: show the ADR draft again and allow the user to revise the
+          Context, Decision, or Consequences fields. Re-present for A/R after edits.
+
+      ## Step 3 — Summary
+
+      After all candidates are processed, print:
+
+        adr-consolidation complete.
+        Accepted: N | Rejected: M | Edited then accepted: K
+        Files written: (list of docs/adr/NNNN-*.md paths, or "none")
+
+  - id: run-triage
+    name: Run Triage — Refresh Project State
+    type: bash
+    runtime: bun
+    depends_on: [adr-consolidation]
+    script: |
+      #!/usr/bin/env node
+      // Reuse the shared triage workflow — no duplicate triage logic in cleanup.
+      // triage produces HANDOFF.md and updates ROADMAP.md exactly once per run.
+      import { execFileSync } from 'node:child_process'
+
+      try {
+        execFileSync('archon', ['workflow', 'run', 'unic-dlc-triage'], {
+          stdio: 'inherit',
+          env: { ...process.env },
+        })
+      } catch (err) {
+        console.error('run-triage: archon workflow run failed:', err.message)
+        process.exit(1)
+      }
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-explore.yaml b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-explore.yaml
new file mode 100644
index 00000000..a112c5dd
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-explore.yaml
@@ -0,0 +1,210 @@
+name: unic-dlc-explore
+description: Parallel research across four dimensions, then synthesize findings into docs/workflow/<slug>/findings.md.
+
+inputs:
+  slug:
+    description: Short identifier for this exploration (used as the findings directory name under docs/workflow/).
+    required: true
+
+nodes:
+  - id: research-stack
+    name: Research Stack
+    type: prompt
+    depends_on: []
+    prompt: |
+      You are running the unic-archon-dlc explore workflow — research-stack node.
+
+      Goal: understand the technology stack of this project.
+
+      1. Read package.json (root and any workspace members listed in pnpm-workspace.yaml).
+      2. Note the runtime, package manager, module system, and key devDependencies.
+      3. Check .nvmrc or .node-version for the pinned Node.js version.
+      4. Look for any framework, test runner, linter, or build tool references.
+      5. Output a concise Markdown summary under the heading "## Stack Findings" covering:
+         - Runtime version
+         - Package manager
+         - Module system (CJS/ESM)
+         - Key frameworks / libraries
+         - Toolchain (lint, format, type-check, test)
+
+  - id: research-features
+    name: Research Features
+    type: prompt
+    depends_on: []
+    prompt: |
+      You are running the unic-archon-dlc explore workflow — research-features node.
+
+      Goal: identify the key features and capabilities of this project.
+
+      1. Read README.md (if present) for a high-level feature list.
+      2. Scan docs/plans/ and docs/issues/ for feature specs and issue slugs.
+      3. Inspect any CHANGELOG.md files for shipped feature entries.
+      4. Look at exported function names in lib/*.mjs to infer capabilities.
+      5. Output a concise Markdown summary under the heading "## Features Findings" covering:
+         - Core capabilities already shipped
+         - In-progress or planned features (from specs/issues)
+         - Any notable user-facing commands or workflows
+
+  - id: research-architecture
+    name: Research Architecture
+    type: prompt
+    depends_on: []
+    prompt: |
+      You are running the unic-archon-dlc explore workflow — research-architecture node.
+
+      Goal: understand the structural and architectural decisions of this project.
+
+      1. Read CONTEXT.md (if present) for the domain model and layering decisions.
+      2. Read docs/adr/ entries (newest five by filename) for recorded decisions.
+      3. Examine the directory tree: lib/, test/, hooks/, .archon/, and any packages/.
+      4. Identify patterns: pure functions vs. stateful modules, DAG/workflow orchestration, plugin conventions.
+      5. Note any cross-platform constraints (Windows, macOS, Linux) visible in the code.
+      6. Output a concise Markdown summary under the heading "## Architecture Findings" covering:
+         - Directory structure and module responsibilities
+         - Key design patterns and conventions
+         - Recorded decisions from ADRs
+         - Cross-platform or portability constraints
+
+  - id: research-pitfalls
+    name: Research Pitfalls
+    type: prompt
+    depends_on: []
+    prompt: |
+      You are running the unic-archon-dlc explore workflow — research-pitfalls node.
+
+      Goal: surface known pitfalls, edge cases, and areas of technical debt.
+
+      1. Search for TODO, FIXME, HACK, and XXX comments across all source files.
+      2. Look for error-handling gaps: try/catch blocks that swallow errors, unhandled promise rejections.
+      3. Check CI configuration (.github/workflows/ or .gitlab-ci.yml) for flaky or skipped jobs.
+      4. Inspect test coverage: are there lib modules with no corresponding test file?
+      5. Read the most recent HANDOFF.md (if present) for documented blockers.
+      6. Output a concise Markdown summary under the heading "## Pitfalls Findings" covering:
+         - Known bugs or TODOs in the code
+         - Untested modules or coverage gaps
+         - CI issues or flaky behaviours
+         - Blockers from the latest HANDOFF.md
+
+  - id: synthesize
+    name: Synthesize Findings
+    type: prompt
+    depends_on: [research-stack, research-features, research-architecture, research-pitfalls]
+    prompt: |
+      You are running the unic-archon-dlc explore workflow — synthesize node.
+
+      You have four research outputs from the parallel nodes:
+        - research-stack    → "## Stack Findings"
+        - research-features → "## Features Findings"
+        - research-architecture → "## Architecture Findings"
+        - research-pitfalls → "## Pitfalls Findings"
+
+      Additionally:
+        - The user's stated intent for this exploration is captured in the `slug` workflow input.
+        - Read CONTEXT.md at the project root if it exists — it contains the canonical domain vocabulary.
+        - If this is a monorepo, also read CONTEXT-MAP.md at the repo root.
+
+      Your job:
+      1. Call initFindingsDir(projectDir, slug) to create docs/workflow/<slug>/ (idempotent).
+      2. Call writeFindingsMd(findingsDir, { stack, features, architecture, pitfalls, brief }) where:
+         - `stack`, `features`, `architecture`, `pitfalls` are the respective findings summaries
+           (strip the "## … Findings" heading and keep only the body text).
+         - `brief` is a new 3-5 sentence "Integrated Brief" that synthesises all four dimensions
+           into a single coherent narrative, highlighting the most important insight from each,
+           surfacing any cross-cutting concerns, and recommending the single highest-impact next step.
+      3. Print: "Explore complete. Findings written to docs/workflow/<slug>/findings.md"
+
+  - id: prototype
+    name: Prototype Experiments
+    type: prompt
+    depends_on: [synthesize]
+    prompt: |
+      You are running the unic-archon-dlc explore workflow — prototype node.
+
+      Goal: design and evaluate one or more experiments that test the key unknowns
+      identified in the findings, then record a verdict for each.
+
+      Inputs available:
+        - `slug` workflow input (the exploration identifier)
+        - docs/workflow/<slug>/findings.md (written by the synthesize node)
+
+      Steps:
+      1. Read docs/workflow/<slug>/findings.md.
+      2. From the Integrated Brief and Pitfalls section identify 1-3 concrete unknowns
+         worth spiking. For each unknown, design a minimal experiment:
+           - Give it a short descriptive title.
+           - Describe what you would build or measure.
+           - Reason through what outcome is most likely given the existing evidence.
+           - Assign exactly ONE verdict: VALIDATED, INVALIDATED, or PARTIAL.
+             · VALIDATED  — the approach is confirmed viable; evidence is sufficient.
+             · INVALIDATED — the approach is ruled out; a blocker was found.
+             · PARTIAL    — the approach shows promise but requires follow-up work.
+           - Write 2-4 sentences of supporting evidence or reasoning.
+      3. Call appendSpikeVerdicts(findingsDir, verdicts) from lib/spike-verdicts.mjs,
+         passing an array of { title, verdict, notes } objects.
+         This appends a "## Spike verdicts" section to findings.md.
+      4. Print: "Prototype complete. <N> verdict(s) written to docs/workflow/<slug>/findings.md"
+
+  - id: code-preserve-gate
+    name: Spike Branch Gate
+    type: interactive
+    fresh_context: true
+    depends_on: [prototype]
+    prompt: |
+      You are running the unic-archon-dlc explore workflow — code-preserve-gate node.
+
+      Goal: decide whether to preserve any spike code on a dedicated branch.
+
+      Ask the user exactly this question:
+        "Should I preserve this spike code on a branch? (yes/no)"
+
+      If the user answers YES (or y):
+        1. Derive a branch name: spike/<slug>  (e.g. spike/auth-refresh).
+        2. Run: git checkout -b spike/<slug>
+        3. Stage and commit any unstaged files related to this spike:
+             git add -A
+             git commit -m "spike(<slug>): preserve prototype code from explore workflow"
+        4. Print: "Spike code committed on branch spike/<slug>."
+
+      If the user answers NO (or n):
+        1. Leave the worktree unchanged — do NOT stage or commit anything.
+        2. Print: "Spike code discarded. Worktree left clean."
+
+  - id: create-spike-ticket
+    name: Create Spike Ticket
+    type: prompt
+    depends_on: [code-preserve-gate]
+    prompt: |
+      You are running the unic-archon-dlc explore workflow — create-spike-ticket node.
+
+      Goal: create (or update) a spike ticket in the configured tracker.
+
+      Inputs available:
+        - `slug` workflow input
+        - docs/workflow/<slug>/findings.md (contains the ## Spike verdicts section)
+
+      Steps:
+      1. Read docs/workflow/<slug>/findings.md.
+      2. Call parseSpikeVerdicts(content) from lib/spike-verdicts.mjs to extract verdicts.
+      3. Build a one-line verdict summary, e.g.:
+           "1 VALIDATED, 1 INVALIDATED, 1 PARTIAL"
+      4. Load the tracker configuration with loadConfig(projectDir) from lib/config-loader.mjs.
+         Use config.tracker (defaults to 'github' if absent).
+      5. Load labels with getDefaultLabels(tracker) from lib/labels-config.mjs.
+      6. Check whether a spike ticket for this slug already exists:
+           - GitHub:         gh issue list --label spike --search "<slug>" --json number,title
+           - local-markdown: check for docs/issues/<slug>-spike/index.md
+           - Other trackers: search by title containing slug.
+      7a. If NO existing ticket:
+           Call buildCreateCommand(tracker, title, 'spike', 'p2', labels) where
+           title = "Spike: <slug> — <one-line verdict summary>".
+           The body should include:
+             - Link to docs/workflow/<slug>/findings.md
+             - The full ## Spike verdicts section text
+           Execute or print the generated command.
+      7b. If ticket EXISTS:
+           Call buildUpdateCommand(tracker, issueId, 'needs-triage', labels)
+           and additionally update the body to append the new verdict summary.
+           Execute or print the generated command.
+      8. Print one of:
+           "Spike ticket created: <title>"
+           "Spike ticket updated: <issueId>"
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-plan.yaml b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-plan.yaml
new file mode 100644
index 00000000..5c79b155
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-plan.yaml
@@ -0,0 +1,421 @@
+name: unic-dlc-plan
+description: >
+  Adversarial spec interview → PRD synthesis → human PR gate.
+  Produces docs/workflow/<slug>/PRD.md and optional ADR entries.
+
+inputs:
+  slug:
+    description: Short identifier for this planning session (used as the workflow directory name under docs/workflow/).
+    required: true
+
+nodes:
+  - id: load-context
+    name: Load Context
+    type: prompt
+    depends_on: []
+    prompt: |
+      You are running the unic-archon-dlc plan workflow — load-context node.
+
+      Goal: build a rich shared context before the adversarial interview begins.
+
+      1. Read CONTEXT.md at the project root if it exists. Note the domain vocabulary, key entities, and any established conventions.
+      2. Read CONTEXT-MAP.md at the repo root if it exists (present in monorepos). Note which sub-project this workflow targets.
+      3. Read all ADR files in docs/adr/ (sorted by filename). For each, note the decision title and outcome.
+      4. If docs/workflow/{{ inputs.slug }}/findings.md exists, read it. Note the Integrated Brief and any surfaced pitfalls.
+      5. Output a structured context summary with the following headings:
+         ## Domain Model
+         (key entities and vocabulary from CONTEXT.md)
+         ## Established Decisions
+         (list of ADRs: filename → one-line summary)
+         ## Prior Research
+         (Integrated Brief from findings.md, or "No findings.md for this slug." if absent)
+
+  - id: specs
+    name: Adversarial Spec Interview
+    type: loop
+    depends_on: [load-context]
+    prompt: |
+      You are running the unic-archon-dlc plan workflow — specs (adversarial interview) node.
+
+      Context already loaded in load-context:
+        - Domain vocabulary and entities
+        - Established ADR decisions
+        - Prior research findings (if any)
+
+      Discuss mode: {{ workflow.discuss_mode | default('interview') }}
+        - "interview"    → ask one open-ended question at a time; probe deeply; challenge assumptions
+        - "assumptions"  → surface all implicit assumptions upfront, then confirm or refute each
+
+      Interview guidelines:
+      1. If discuss_mode is "assumptions": enumerate every assumption you can identify about the
+         problem space, proposed solution, constraints, and acceptance criteria before asking questions.
+         For each assumption, ask the user to confirm, refute, or refine it.
+         If discuss_mode is "interview": ask one focused, adversarial question at a time.
+         Do not ask multiple questions in a single turn.
+      2. For every non-obvious architectural or process decision that emerges, propose an ADR entry:
+         - Suggest a filename following the pattern docs/adr/NNNN-<short-slug>.md
+         - Draft the ADR inline (Status, Context, Decision, Consequences)
+         - Write it live to docs/adr/NNNN-<short-slug>.md if the user confirms
+         - Use the next available NNNN by checking existing files in docs/adr/
+      3. Continue until the user signals they are done (e.g. "done", "that's enough", "proceed").
+      4. Before exiting, produce a structured interview transcript summary:
+         ## Interview Summary
+         (3-5 bullet points capturing the key decisions and constraints surfaced)
+         ## ADRs Produced
+         (list of any ADR files written during this session, or "None.")
+
+  - id: to-prd
+    name: Synthesise PRD
+    type: prompt
+    depends_on: [specs]
+    prompt: |
+      You are running the unic-archon-dlc plan workflow — to-prd node.
+
+      You have:
+        - The context summary from load-context (domain model, ADRs, prior research)
+        - The interview transcript summary and ADRs produced from the specs node
+        - The workflow slug: {{ inputs.slug }}
+
+      Your job: synthesise all of the above into a structured PRD.
+
+      1. Call writePrd(projectDir, slug, sections) where sections has exactly these keys:
+         - problemStatement: 2-4 sentences describing the pain or gap being addressed
+         - solution: 2-4 sentences describing the proposed approach at a high level
+         - userStories: a bulleted list of "As a <role>, I want <capability>, so that <benefit>." stories (at least 3)
+         - implementationDecisions: bullet list of key technical choices; reference ADR filenames where applicable
+         - testingDecisions: bullet list of testing strategy choices (framework, coverage targets, edge cases to cover)
+         - outOfScope: bullet list of explicitly excluded items to prevent scope creep
+         - furtherNotes: any open questions, dependencies, or items to revisit after v1
+
+      2. The resulting file must be docs/workflow/{{ inputs.slug }}/PRD.md.
+      3. Print: "PRD written to docs/workflow/{{ inputs.slug }}/PRD.md"
+
+  - id: prd-gate
+    name: PRD Human Gate
+    type: interactive
+    fresh_context: true
+    depends_on: [to-prd]
+    prompt: |
+      You are running the unic-archon-dlc plan workflow — prd-gate (first human PR gate) node.
+
+      The PRD and any new ADRs are ready for human review.
+
+      Actions:
+      1. Run validatePrdSections(content) on the PRD.md content to confirm all 7 sections are present.
+         If any sections are missing, list them and offer to fix them before opening the PR.
+      2. Stage the following files for a new PR:
+         - docs/workflow/{{ inputs.slug }}/PRD.md
+         - Any ADR files written during the specs node (docs/adr/NNNN-*.md that are new or modified)
+      3. Open a PR targeting the develop branch with:
+         - Title: "plan({{ inputs.slug }}): PRD and ADRs"
+         - Body:
+           ## Summary
+           - PRD for `{{ inputs.slug }}` workflow session
+           - Lists any new or updated ADR files
+           - Includes all 7 required PRD sections
+
+           ## Review Checklist
+           - [ ] Problem Statement clearly describes the pain or gap
+           - [ ] Solution is specific and implementable
+           - [ ] User Stories cover the primary use cases
+           - [ ] Implementation Decisions reference relevant ADRs
+           - [ ] Testing Decisions are concrete and actionable
+           - [ ] Out of Scope prevents scope creep
+           - [ ] Further Notes captures open questions
+      4. Pause and wait for human review.
+         - If the PR is **approved** and merged: print "PRD gate passed. Proceeding." and exit.
+         - If the PR is **rejected** or changes are requested: print "PRD gate: changes requested. Returning to specs."
+           Then return control to the specs node so the interview can be resumed or refined.
+
+  - id: to-issues
+    name: Decompose PRD into Issues
+    type: prompt
+    depends_on: [prd-gate]
+    prompt: |
+      You are running the unic-archon-dlc plan workflow — to-issues node.
+
+      Goal: decompose the approved PRD into independently-deliverable, vertically-sliced issues and
+      publish them to the tracker in dependency order.
+
+      ## Step 1 — Read the PRD
+
+      Read docs/workflow/{{ inputs.slug }}/PRD.md in full.
+
+      ## Step 2 — Decompose into vertical slices
+
+      Produce a draft list of issues. Each issue MUST have all of the following fields:
+
+        - id          : short kebab-case identifier unique within this file (e.g. "issue-01")
+        - title       : one-line description of the deliverable
+        - type        : one of: feature | bug | spike | tech-debt | docs
+        - priority    : one of: p0 | p1 | p2 | p3
+        - blocked_by  : array of IDs in *this* file that must ship first (empty array if independent)
+        - acceptance_criteria : non-empty array of statements that can each be demonstrated independently
+        - summary     : one short paragraph describing the work
+
+      Rules for good vertical slices:
+        - Each slice delivers observable user value on its own (or unblocks a slice that does).
+        - Slices are as small as possible while still being independently deployable/demonstrable.
+        - No slice should duplicate work covered by another slice.
+        - blocked_by must form a DAG — no cycles.
+
+      ## Step 3 — Validation loop (iterate until approved)
+
+      Present the breakdown to the user as a numbered list in this format:
+        <number>. [<type>/<priority>] <title>
+           blocked_by: <comma-separated IDs, or "none">
+           summary: <summary>
+           acceptance_criteria:
+             - <criterion 1>
+             - …
+
+      Then ask: "Validate this breakdown, or tell me which issues to split, merge, or relabel."
+
+      Iterate — apply requested changes and re-present — until the user approves explicitly
+      (e.g. replies "approved", "looks good", "proceed").
+
+      ## Step 4 — Validate schema before publishing
+
+      Before publishing, call validateIssue(issue) on every issue in the list.
+      If any issue fails validation, list the errors and fix them before proceeding.
+      Call sortByDependency(issues) to obtain the publication order.
+      If a circular dependency is detected, surface it to the user and ask them to resolve it.
+
+      ## Step 5 — Publish to tracker in dependency order
+
+      Load the tracker configuration from .archon/unic-dlc.config.json (field: tracker).
+      For each issue in sorted order:
+        1. Call buildCreateCommand(tracker, issue.title, issue.type, issue.priority, labels) to generate
+           the CLI command.
+        2. Print the command.
+        3. Execute the command (or, if in dry-run mode, skip execution and print "[dry-run]").
+
+      ## Step 6 — Write issues.json
+
+      Call buildIssuesJson(issues) and write the result to docs/workflow/{{ inputs.slug }}/issues.json.
+      Print: "issues.json written to docs/workflow/{{ inputs.slug }}/issues.json"
+
+  - id: nyquist-map
+    name: Nyquist Map — Attach Test Commands
+    type: prompt
+    depends_on: [to-issues]
+    prompt: |
+      You are running the unic-archon-dlc plan workflow — nyquist-map node.
+
+      Goal: for every issue in the just-published list, decide the exact shell command (or flag as
+      planned) that will verify the issue is done. This ensures every issue is testable before
+      implementation begins.
+
+      ## Step 1 — Load issues
+
+      Read docs/workflow/{{ inputs.slug }}/issues.json and parse the issues array.
+
+      ## Step 2 — Attach test commands
+
+      For each issue, determine the most appropriate verification command:
+
+        - If a concrete, runnable test command exists (e.g. `pnpm test`, `pnpm --filter <name> test`,
+          a specific test file path, or an integration check script), set:
+            test_command: "<exact shell command>"
+
+        - If no test exists yet (e.g. the issue creates new untested functionality, or the test itself
+          is the deliverable of another issue), set:
+            test_command_planned: true
+
+      Guidelines:
+        - Prefer the most specific command that exercises only the acceptance criteria of that issue.
+        - A `test_command_planned: true` is not a failure — it signals that a test must be written as
+          part of implementing that issue.
+        - Do NOT set both test_command and test_command_planned on the same issue.
+        - If the issue's type is 'docs' or 'spike', set test_command_planned: true unless a linting
+          or build check is meaningful.
+
+      ## Step 3 — Update issues.json
+
+      Merge the test_command / test_command_planned fields into each issue object.
+      Call buildIssuesJson(updatedIssues) and overwrite docs/workflow/{{ inputs.slug }}/issues.json.
+      Print a summary table:
+
+        Issue ID     | Test command / planned
+        -------------|--------------------------------------
+        issue-01     | pnpm --filter my-plugin test
+        issue-02     | [planned]
+        …
+
+      Print: "nyquist-map complete. issues.json updated with test commands."
+
+  - id: plan-checker
+    name: Plan Checker — Validate Issue Plan
+    type: loop
+    depends_on: [nyquist-map]
+    prompt: |
+      You are running the unic-archon-dlc plan workflow — plan-checker node.
+
+      Goal: validate the issue plan across four dimensions before generating the build YAML.
+      Max 3 iterations. Escalate immediately to the human gate if the issue count does not
+      decrease between consecutive iterations (stall detection).
+
+      ## Step 1 — Load issues
+
+      Read docs/workflow/{{ inputs.slug }}/issues.json and parse the issues array.
+      Track: iteration number (start at 1), previous issue count (start at -1).
+
+      ## Step 2 — Stall detection
+
+      If iteration > 1 AND current issue count >= previous issue count:
+        Print: "plan-checker: stall detected (issue count did not decrease). Escalating to human gate."
+        Exit the loop and proceed to yaml-gen.
+
+      If iteration > 3:
+        Print: "plan-checker: max iterations reached. Escalating to human gate."
+        Exit the loop and proceed to yaml-gen.
+
+      ## Step 3 — Validate four dimensions
+
+      ### 3.1 Consistency
+        - Collect all `blocked_by` values across all issues.
+        - Check that every referenced ID exists in the issue list.
+        - Orphaned references: IDs in `blocked_by` that don't exist in the list.
+        - Check that every acceptance criterion from every PRD user story is covered by
+          at least one issue's acceptance_criteria.
+        - Report: list of orphaned references; list of uncovered PRD criteria.
+
+      ### 3.2 Completeness
+        - For every issue, verify all mandatory fields are present and non-empty:
+          id, title, type, priority, blocked_by, acceptance_criteria, summary.
+        - acceptance_criteria must be a non-empty array.
+        - Report: list of issues with missing or empty mandatory fields.
+
+      ### 3.3 Decision coverage
+        - Read CONTEXT.md at the project root (if present). Extract trackable decisions
+          (sentences or bullet points that describe a binding technical or process choice).
+        - Read the interview transcript summary produced by the specs node (look for
+          "## Interview Summary" in the workflow session's notes or the specs node output).
+        - For each trackable decision, check whether at least one issue references or
+          addresses it (in title, summary, or acceptance_criteria).
+        - Report: list of unaddressed decisions.
+
+      ### 3.4 Nyquist compliance
+        - Check each issue for either `test_command` (string) or `test_command_planned: true`.
+        - Issues missing both fields fail this check.
+        - Report: list of issues without test coverage mapping.
+
+      ## Step 4 — Build the validation report
+
+      Produce a structured report:
+
+        ## plan-checker Report (iteration {{ loop.iteration }})
+
+        ### Consistency
+        <orphaned references or "✓ No orphaned references">
+        <uncovered PRD criteria or "✓ All PRD criteria covered">
+
+        ### Completeness
+        <missing fields per issue or "✓ All issues have required fields">
+
+        ### Decision Coverage
+        <unaddressed decisions or "✓ All trackable decisions addressed">
+
+        ### Nyquist Compliance
+        <issues missing test mapping or "✓ All issues have test_command or test_command_planned">
+
+        ### Summary
+        <overall PASS or FAIL with counts>
+
+      Write the report to docs/workflow/{{ inputs.slug }}/plan-checker-report.md
+      (overwrite on each iteration).
+
+      ## Step 5 — Act on the report
+
+      If all four dimensions PASS:
+        Print: "plan-checker: all checks passed. Proceeding to yaml-gen."
+        Exit the loop.
+
+      Otherwise:
+        Present the report to the user.
+        Ask: "Which issues should I fix? (Or type 'accept' to proceed despite failures.)"
+        If the user types 'accept':
+          Print: "plan-checker: user accepted failures. Proceeding."
+          Exit the loop.
+        Otherwise:
+          Apply the requested fixes to issues.json.
+          Increment iteration counter and loop again.
+
+  - id: yaml-gen
+    name: YAML Gen — Build Workflow Generator
+    type: bash
+    runtime: bun
+    depends_on: [plan-checker]
+    script: |
+      #!/usr/bin/env node
+      // @ts-check
+      import { readFileSync, writeFileSync, mkdirSync } from 'node:fs'
+      import { join } from 'node:path'
+      import { buildYaml } from './lib/dag-builder.mjs'
+
+      const slug = process.env.ARCHON_INPUT_SLUG
+      if (!slug) {
+        console.error('yaml-gen: ARCHON_INPUT_SLUG is not set')
+        process.exit(1)
+      }
+
+      const issuesPath = join('docs', 'workflow', slug, 'issues.json')
+      let issues
+      try {
+        issues = JSON.parse(readFileSync(issuesPath, 'utf8'))
+      } catch (err) {
+        console.error(`yaml-gen: cannot read ${issuesPath}: ${err.message}`)
+        process.exit(1)
+      }
+
+      let yaml
+      try {
+        yaml = buildYaml(slug, issues)
+      } catch (err) {
+        console.error(`yaml-gen: ${err.message}`)
+        process.exit(1)
+      }
+
+      const outDir = join('.archon', 'workflows')
+      mkdirSync(outDir, { recursive: true })
+      const outPath = join(outDir, `build-${slug}.yaml`)
+      writeFileSync(outPath, yaml)
+      console.log(`yaml-gen: build workflow written to ${outPath}`)
+
+  - id: plan-pr-gate
+    name: Plan PR Gate — Human Review
+    type: interactive
+    fresh_context: true
+    depends_on: [yaml-gen]
+    prompt: |
+      You are running the unic-archon-dlc plan workflow — plan-pr-gate (second human PR gate) node.
+
+      The issue plan and build workflow are ready for human review.
+
+      Actions:
+      1. Stage the following files for a new PR:
+         - docs/workflow/{{ inputs.slug }}/issues.json
+         - docs/workflow/{{ inputs.slug }}/plan-checker-report.md
+         - .archon/workflows/build-{{ inputs.slug }}.yaml
+      2. Open a PR targeting the develop branch with:
+         - Title: "plan({{ inputs.slug }}): issue plan and build workflow"
+         - Body:
+           ## Summary
+           - Issue plan for `{{ inputs.slug }}` ({{ issues | length }} issues)
+           - plan-checker validation report attached
+           - Auto-generated `build-{{ inputs.slug }}.yaml` Archon workflow
+
+           ## Review Checklist
+           - [ ] All issues are independently deliverable vertical slices
+           - [ ] Dependency order is correct (no orphaned blocked_by references)
+           - [ ] Every issue has a test command or is marked test_command_planned
+           - [ ] build-{{ inputs.slug }}.yaml parallel/serial structure looks correct
+           - [ ] plan-checker report shows no critical failures (or failures are intentional)
+      3. Pause and wait for human review.
+         - If the PR is **approved** and merged:
+           Print: "Plan PR gate passed. Build workflow ready at .archon/workflows/build-{{ inputs.slug }}.yaml"
+           Exit.
+         - If the PR is **rejected** or changes are requested:
+           Print: "Plan PR gate: changes requested. Returning to plan-checker for another validation pass."
+           Return control to the plan-checker node so validation can be re-run on the updated plan.
+           Do NOT return to to-issues (no full re-decomposition).
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-qa.yaml b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-qa.yaml
new file mode 100644
index 00000000..dcf50f18
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-qa.yaml
@@ -0,0 +1,361 @@
+name: unic-dlc-qa
+description: >
+  QA workflow: e2e test execution → coverage gate → UAT checklist → merge.
+  Runs after the build workflow is complete and the build PR has been approved.
+
+inputs:
+  slug:
+    description: Planning session identifier — matches the slug used in /unic-dlc-plan and /unic-dlc-build.
+    required: true
+
+nodes:
+  - id: e2e
+    name: End-to-End Tests
+    type: bash
+    runtime: bun
+    depends_on: []
+    script: |
+      #!/usr/bin/env node
+      // @ts-check
+      import { execFileSync } from 'node:child_process'
+      import { existsSync, readFileSync } from 'node:fs'
+      import { join } from 'node:path'
+
+      const projectDir = process.cwd()
+      const configPath = join(projectDir, '.archon', 'unic-dlc.config.json')
+
+      if (!existsSync(configPath)) {
+        console.error('e2e: .archon/unic-dlc.config.json not found.')
+        console.error('Run the install hook first: archon install')
+        process.exit(1)
+      }
+
+      const config = JSON.parse(readFileSync(configPath, 'utf8'))
+      const e2eCommand = config.e2e_command ?? null
+
+      if (!e2eCommand) {
+        console.error('e2e: e2e_command is not configured.')
+        console.error('')
+        console.error('To set it, run the install hook with --reconfigure:')
+        console.error('  archon install --reconfigure')
+        console.error('and provide a value for "e2e_command" (e.g. "pnpm test:e2e" or "npx playwright test").')
+        process.exit(1)
+      }
+
+      const parts = e2eCommand.trim().split(/\s+/)
+      const cmd = parts[0]
+      const args = parts.slice(1)
+
+      console.log(`e2e: running → ${e2eCommand}`)
+      console.log('')
+
+      try {
+        execFileSync(cmd, args, { cwd: projectDir, stdio: 'inherit' })
+        console.log('')
+        console.log('e2e: PASSED ✓')
+      } catch {
+        console.error('')
+        console.error('e2e: FAILED ✗')
+        process.exit(1)
+      }
+
+  - id: coverage-gate
+    name: Coverage Gate
+    type: bash
+    runtime: bun
+    depends_on: [e2e]
+    script: |
+      #!/usr/bin/env node
+      // @ts-check
+      import { execFileSync } from 'node:child_process'
+      import { existsSync, readFileSync } from 'node:fs'
+      import { join } from 'node:path'
+
+      const projectDir = process.cwd()
+      const configPath = join(projectDir, '.archon', 'unic-dlc.config.json')
+
+      let coverageThreshold = null
+      if (existsSync(configPath)) {
+        const config = JSON.parse(readFileSync(configPath, 'utf8'))
+        coverageThreshold = config.coverage_threshold ?? null
+      }
+
+      if (coverageThreshold === null) {
+        console.log('coverage-gate: no coverage_threshold configured — skipping.')
+        console.log('To enable, add "coverage_threshold": <number> to .archon/unic-dlc.config.json.')
+        process.exit(0)
+      }
+
+      console.log(`coverage-gate: threshold is ${coverageThreshold}%`)
+      console.log('coverage-gate: running test suite with coverage...')
+      console.log('')
+
+      // Run tests with coverage; capture output to parse coverage percentage
+      let coverageOutput = ''
+      let failed = false
+      try {
+        coverageOutput = execFileSync('pnpm', ['test', '--coverage'], {
+          cwd: projectDir,
+          encoding: 'utf8',
+          stdio: ['inherit', 'pipe', 'pipe'],
+        })
+        process.stdout.write(coverageOutput)
+      } catch (/** @type {any} */ err) {
+        if (err.stdout) process.stdout.write(err.stdout)
+        if (err.stderr) process.stderr.write(err.stderr)
+        coverageOutput = (err.stdout ?? '') + (err.stderr ?? '')
+        failed = true
+      }
+
+      // Attempt to extract a coverage percentage from the output.
+      // Handles common formats: "All files | 87.5 |", "Stmts: 87.5%", "87.5% statements"
+      const percentMatch =
+        coverageOutput.match(/All files\s*\|\s*([\d.]+)/) ??
+        coverageOutput.match(/Stmts\s*[:\|]\s*([\d.]+)/) ??
+        coverageOutput.match(/([\d.]+)%\s+statements/i)
+
+      if (percentMatch) {
+        const actual = parseFloat(percentMatch[1])
+        if (actual < coverageThreshold) {
+          console.error(`\ncoverage-gate: FAILED (${actual}% < threshold ${coverageThreshold}%)`)
+          process.exit(1)
+        }
+        console.log(`\ncoverage-gate: PASSED (${actual}%)`)
+      } else if (failed) {
+        // Could not parse percentage but the command failed
+        console.error(`\ncoverage-gate: FAILED (test suite exited non-zero; threshold ${coverageThreshold}%)`)
+        process.exit(1)
+      } else {
+        // Command passed but no parseable percentage — treat as passed with a warning
+        console.log(`\ncoverage-gate: PASSED (coverage % could not be parsed; threshold ${coverageThreshold}%)`)
+      }
+
+  - id: uat-gate
+    name: UAT Gate — Acceptance Checklist
+    type: interactive
+    fresh_context: true
+    depends_on: [coverage-gate]
+    prompt: |
+      You are running the unic-archon-dlc QA workflow — uat-gate node.
+
+      Goal: walk through the UAT checklist derived from the PRD's acceptance criteria alongside
+      the e2e results, then get human sign-off before the merge step proceeds.
+
+      ## Step 1 — Load artefacts
+
+      1. Read `docs/workflow/{{ inputs.slug }}/PRD.md`.
+         Extract all acceptance criteria. They are typically found in:
+           - Each "User Story" block as sub-bullet "Acceptance criteria:" lists.
+           - A top-level "## Acceptance Criteria" section if present.
+         Number each criterion sequentially: AC-1, AC-2, …
+
+      2. Check whether `docs/workflow/{{ inputs.slug }}/report.md` exists and read it if so.
+         The report contains the goals-check coverage matrix and test outcomes from the build phase.
+
+      ## Step 2 — Present the UAT checklist
+
+      Print the following header:
+
+        === UAT Checklist for {{ inputs.slug }} ===
+
+      Then for each acceptance criterion:
+
+        AC-N: <criterion text>
+          Evidence: <matching row from coverage matrix, or "not recorded">
+          [ ] Verified by human
+
+      Follow the list with:
+
+        e2e result: PASSED  (or FAILED — see e2e node output above)
+        Coverage:   PASSED  (or SKIPPED / FAILED — see coverage-gate output above)
+
+      ## Step 3 — Ask for sign-off
+
+      Ask the user:
+        "Please review each item above and confirm:
+         1. All acceptance criteria are verified (tick each [ ] mentally or note any failures).
+         2. You approve this build for merge.
+
+         Type APPROVE to continue to the merge step.
+         Type REJECT <AC-N> to flag a failing checklist item — I will return control to whoever
+         owns that item for remediation."
+
+      ## Step 4 — Act on the response
+
+      If the user types APPROVE (case-insensitive):
+        Print: "uat-gate: APPROVED. Proceeding to merge."
+
+      If the user types REJECT <AC-N> (or REJECT with any text):
+        Print: "uat-gate: REJECTED — {{ user response }}."
+        Print: "Returning control. Address the failing criterion and re-run /unic-dlc-qa {{ inputs.slug }} to retry."
+        Halt the workflow (do not proceed to the merge node).
+
+  - id: verify-pr-base
+    name: Verify PR Base Branch
+    type: bash
+    runtime: bun
+    depends_on: [uat-gate]
+    script: |
+      #!/usr/bin/env node
+      // @ts-check
+      import { readFileSync } from 'node:fs'
+      import { execFileSync } from 'node:child_process'
+
+      let config = {}
+      try {
+        config = JSON.parse(readFileSync('.archon/unic-dlc.config.json', 'utf8'))
+      } catch {
+        console.error('verify-pr-base: cannot read .archon/unic-dlc.config.json')
+        process.exit(1)
+      }
+
+      const tracker = config.tracker ?? 'github'
+      const branching = config.branching ?? 'gitflow'
+      const expectedBase = branching === 'gitflow' ? 'develop' : 'main'
+
+      if (tracker === 'github') {
+        let actualBase
+        try {
+          const out = execFileSync('gh', ['pr', 'view', '--json', 'baseRefName'], { encoding: 'utf8' })
+          actualBase = JSON.parse(out).baseRefName
+        } catch (err) {
+          console.error(`verify-pr-base: cannot query PR base — ${err.message}`)
+          process.exit(1)
+        }
+        if (actualBase !== expectedBase) {
+          console.error(
+            `verify-pr-base: PR targets "${actualBase}" but expected "${expectedBase}" (branching: ${branching}).`,
+          )
+          console.error('  Retarget the PR before merging.')
+          process.exit(1)
+        }
+        console.log(`verify-pr-base: PR base is "${actualBase}" ✓`)
+      } else if (tracker === 'ado') {
+        let actualBase
+        try {
+          const out = execFileSync('az', ['repos', 'pr', 'show', '--output', 'json'], { encoding: 'utf8' })
+          actualBase = JSON.parse(out).targetRefName?.replace('refs/heads/', '')
+        } catch (err) {
+          console.error(`verify-pr-base: cannot query ADO PR — ${err.message}`)
+          process.exit(1)
+        }
+        if (actualBase !== expectedBase) {
+          console.error(
+            `verify-pr-base: PR targets "${actualBase}" but expected "${expectedBase}" (branching: ${branching}).`,
+          )
+          process.exit(1)
+        }
+        console.log(`verify-pr-base: PR base is "${actualBase}" ✓`)
+      } else {
+        // Jira / local-markdown — no CLI to query PR base
+        console.log(
+          `verify-pr-base: tracker "${tracker}" does not support PR base verification. Skipping (warning only).`,
+        )
+      }
+
+  - id: merge
+    name: Merge PR
+    type: bash
+    runtime: bun
+    depends_on: [verify-pr-base]
+    script: |
+      #!/usr/bin/env node
+      // @ts-check
+      import { execFileSync } from 'node:child_process'
+      import { existsSync, readFileSync } from 'node:fs'
+      import { join } from 'node:path'
+
+      const projectDir = process.cwd()
+      const configPath = join(projectDir, '.archon', 'unic-dlc.config.json')
+
+      if (!existsSync(configPath)) {
+        console.error('merge: .archon/unic-dlc.config.json not found.')
+        process.exit(1)
+      }
+
+      const config = JSON.parse(readFileSync(configPath, 'utf8'))
+      const tracker = config.tracker ?? 'github'
+      const branching = config.branching ?? 'gitflow'
+      const prStrategy = config.pr_strategy ?? 'squash'
+      const slug = process.env.ARCHON_INPUT_SLUG ?? ''
+
+      console.log(`merge: tracker=${tracker}, branching=${branching}, pr_strategy=${prStrategy}`)
+      console.log('')
+
+      // Merge via the configured tracker
+      switch (tracker) {
+        case 'github': {
+          const mergeFlag = prStrategy === 'merge' ? '--merge' : '--squash'
+          console.log(`merge: running gh pr merge ${mergeFlag}`)
+          try {
+            execFileSync('gh', ['pr', 'merge', mergeFlag], { cwd: projectDir, stdio: 'inherit' })
+            console.log('merge: PR merged successfully via GitHub CLI.')
+          } catch {
+            console.error('merge: gh pr merge failed. Ensure you are on a feature branch with an open PR.')
+            process.exit(1)
+          }
+          break
+        }
+
+        case 'ado': {
+          console.log('merge: running az repos pr update --status completed')
+          try {
+            execFileSync('az', ['repos', 'pr', 'update', '--status', 'completed'], {
+              cwd: projectDir,
+              stdio: 'inherit',
+            })
+            console.log('merge: PR completed successfully via Azure DevOps CLI.')
+          } catch {
+            console.error('merge: az repos pr update failed. Ensure the Azure DevOps PR is open and the CLI is authenticated.')
+            process.exit(1)
+          }
+          break
+        }
+
+        case 'jira':
+          console.log('Merge via Jira: transition the linked PR to Done manually.')
+          console.log('  1. Open the Jira issue linked to this feature.')
+          console.log('  2. Transition it to "Done" (or "In Review → Done" depending on your board).')
+          console.log('  3. Merge the corresponding PR in your VCS provider.')
+          break
+
+        case 'local-markdown':
+          console.log('Merge: update Status in issues to resolved.')
+          console.log(`  Edit docs/issues/<slug>/index.md and set: Status: resolved`)
+          console.log('  Then commit and push the change manually.')
+          break
+
+        default:
+          console.log(`merge: unknown tracker "${tracker}". Merge the PR manually in your tracker.`)
+          break
+      }
+
+      // Delete feature branch after merge when using Gitflow
+      if (branching === 'gitflow' && (tracker === 'github' || tracker === 'ado')) {
+        console.log('')
+        console.log('merge: Gitflow — deleting feature branch...')
+        try {
+          // Determine current branch name
+          const currentBranch = execFileSync('git', ['rev-parse', '--abbrev-ref', 'HEAD'], {
+            cwd: projectDir,
+            encoding: 'utf8',
+          }).trim()
+
+          if (currentBranch.startsWith('feature/')) {
+            // Switch to develop before deleting
+            execFileSync('git', ['checkout', 'develop'], { cwd: projectDir, stdio: 'inherit' })
+            execFileSync('git', ['branch', '-d', currentBranch], { cwd: projectDir, stdio: 'inherit' })
+            console.log(`merge: feature branch "${currentBranch}" deleted. ✓`)
+          } else {
+            console.log(`merge: current branch "${currentBranch}" is not a feature branch — skipping delete.`)
+          }
+        } catch {
+          console.log('merge: could not delete feature branch automatically — delete it manually if needed.')
+        }
+      } else if (branching !== 'gitflow') {
+        console.log('')
+        console.log('merge: github-flow — leaving branch cleanup to GitHub (auto-delete on merge).')
+      }
+
+      console.log('')
+      console.log(`merge: QA workflow for "${slug}" complete. ✓`)
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-review.yaml b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-review.yaml
new file mode 100644
index 00000000..b085baa3
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-review.yaml
@@ -0,0 +1,239 @@
+name: unic-dlc-review
+description: >
+  Self-contained code review across four aspects — code quality, test coverage adequacy,
+  silent failure patterns, and type design quality. Posts (or updates) a single structured
+  comment on the current PR via the configured tracker adapter.
+  No runtime dependency on pr-review-toolkit or any other plugin.
+
+nodes:
+  - id: code-review
+    name: Code Review — Four Aspects
+    type: prompt
+    depends_on: []
+    prompt: |
+      You are running the unic-archon-dlc review workflow — code-review node.
+
+      ## Setup
+
+      1. Read `.archon/unic-dlc.config.json` to determine:
+         - `tracker` (github | ado | jira | local-markdown; default: github)
+         - `branching` strategy if present
+
+      2. Read project conventions from whichever file exists (in order):
+         - `CLAUDE.md`
+         - `AGENTS.md`
+         - `.claude/CLAUDE.md`
+         Note all style rules, naming conventions, tab/space preferences, and
+         explicit "do not" rules.
+
+      3. Identify the current open PR:
+         - github: `gh pr view --json number,title,body,headRefName`
+         - ado:    `az repos pr list --status active --output json | head -1`
+         - local-markdown: use the most recent `docs/workflow/*/report.md` as the review target
+         Store `PR_ID` (issue/PR number or identifier) and `PR_TITLE`.
+
+      4. Compute the diff to review:
+         - github / ado / local-markdown:
+           `git diff origin/$(git rev-parse --abbrev-ref --symbolic-full-name @{u} 2>/dev/null | sed 's|.*/||' || echo main)...HEAD`
+         - If the above fails, fall back to: `git diff HEAD~1...HEAD`
+         Store the result as `DIFF_TEXT`.
+
+      5. List all source files touched in the diff:
+         `git diff --name-only origin/$(git symbolic-ref --short HEAD 2>/dev/null)...HEAD 2>/dev/null || git diff --name-only HEAD~1`
+         Store as `CHANGED_FILES` (newline-separated list).
+
+      6. Read each file in `CHANGED_FILES` in full (skip deleted files).
+
+      ---
+
+      ## Aspect 1 — Code Quality
+
+      Analyse the diff and changed file contents for:
+      - **Readability**: Is each function easy to understand in isolation?
+      - **Naming**: Do variable, function, and module names communicate intent without
+        needing comments?
+      - **Function length**: Are any functions longer than ~40 lines? Could they be split?
+      - **Project-convention adherence**: Check against the rules you read from CLAUDE.md /
+        AGENTS.md. Examples include tab vs. space indentation, quote style, semicolons,
+        trailing commas, explicit returns, early returns over nested conditionals.
+
+      For each problem found, record a finding in this exact format:
+        - `file:line — [code-quality] <one-line description>`
+
+      If no problems are found, record:
+        - `No findings.`
+
+      Store all findings under the heading `### Code Quality`.
+
+      ---
+
+      ## Aspect 2 — Test Coverage Adequacy
+
+      Locate test files in `CHANGED_FILES` or in `test/`, `tests/`, `__tests__/`,
+      `*.test.*`, `*.spec.*` directories. For each changed source module, check whether:
+
+      - Tests exercise the module through its **public interface** (exported functions),
+        not internal helpers.
+      - Assertions verify **outputs and side-effects**, not internal state or call counts
+        on internal collaborators.
+      - No internal collaborators are mocked — only true external boundaries (network,
+        disk, third-party CLIs) should be mocked.
+      - Every exported function has at least one happy-path test and at least one
+        edge-case or error-path test.
+
+      For each gap found, record a finding:
+        - `file:line — [test-coverage] <one-line description>`
+
+      If no gaps are found, record:
+        - `No findings.`
+
+      Store all findings under the heading `### Test Coverage Adequacy`.
+
+      ---
+
+      ## Aspect 3 — Silent Failure Patterns
+
+      Scan the diff for patterns that swallow errors and hide bugs:
+
+      - **Empty catch blocks**: `catch { }` or `catch (e) { /* ignored */ }`
+      - **Swallowed exceptions**: `catch` that logs but does not rethrow, return a default,
+        or propagate the error to the caller.
+      - **Silent fallbacks**: `?? defaultValue` or `|| fallback` on values that should
+        never be absent — where the fallback masks a missing configuration or corrupt state.
+      - **Unhandled promise rejections**: `.then(…)` without `.catch(…)` or `await` outside
+        try/catch in async functions.
+      - **Ignored return values**: calling a function whose return value signals an error
+        or state change, then discarding it.
+
+      For each pattern found, record a finding:
+        - `file:line — [silent-failure] <one-line description>`
+
+      If no patterns are found, record:
+        - `No findings.`
+
+      Store all findings under the heading `### Silent Failure Patterns`.
+
+      ---
+
+      ## Aspect 4 — Type Design Quality
+
+      Analyse type annotations, JSDoc types, or TypeScript types introduced or changed
+      in the diff:
+
+      - **Encapsulation**: Are types narrow enough to prevent invalid usage? Or are they
+        overly broad (e.g. `string` where a union literal would be more precise)?
+      - **Invariants in types**: Do types encode business rules? E.g. a `NonEmptyArray<T>`
+        instead of `T[]` where empty is invalid.
+      - **Illegal-state-unrepresentable**: Are there mutually exclusive fields that could
+        be unified into a discriminated union? Are there optional fields that should be
+        required, or vice-versa?
+      - **JSDoc completeness** (for `.mjs` / `// @ts-check` files): Are all exported
+        function parameters and return types annotated?
+
+      For each weakness found, record a finding:
+        - `file:line — [type-design] <one-line description>`
+
+      If no weaknesses are found, record:
+        - `No findings.`
+
+      Store all findings under the heading `### Type Design Quality`.
+
+      ---
+
+      ## Compose the review comment
+
+      Assemble a Markdown comment with this exact structure:
+
+      ```
+      <!-- unic-dlc-review -->
+      ## unic-dlc-review
+
+      > PR: <PR_TITLE>
+
+      ### Code Quality
+      <findings or "No findings.">
+
+      ### Test Coverage Adequacy
+      <findings or "No findings.">
+
+      ### Silent Failure Patterns
+      <findings or "No findings.">
+
+      ### Type Design Quality
+      <findings or "No findings.">
+
+      ---
+      *Generated by unic-archon-dlc · re-run `/unic-dlc-review` to update this comment.*
+      ```
+
+      The `<!-- unic-dlc-review -->` HTML comment at the very first line is the sentinel
+      used for update detection. Do not remove or alter it.
+
+      Store the assembled Markdown as `REVIEW_BODY`.
+
+      ---
+
+      ## Post or update the comment
+
+      Search for an existing review comment on this PR:
+
+      **github:**
+      ```
+      gh pr view "$PR_ID" --json comments --jq '.comments[] | select(.body | startswith("<!-- unic-dlc-review -->")) | .databaseId'
+      ```
+      If a comment ID is found (call it `EXISTING_ID`):
+        Update: `gh api repos/{owner}/{repo}/issues/comments/$EXISTING_ID -X PATCH -f body="$REVIEW_BODY"`
+        Print:  "review: updated existing comment $EXISTING_ID on PR #$PR_ID"
+      If no comment exists:
+        Create: `gh pr comment "$PR_ID" --body "$REVIEW_BODY"`
+        Print:  "review: posted new comment on PR #$PR_ID"
+
+      **ado:**
+      Fetch threads: `az repos pr thread list --id "$PR_ID" --output json`
+      Search for a thread whose first comment starts with `<!-- unic-dlc-review -->`.
+      If found (`THREAD_ID`, `COMMENT_ID`):
+        Update: `az repos pr thread comment update --id "$PR_ID" --thread-id "$THREAD_ID" --comment-id "$COMMENT_ID" --content "$REVIEW_BODY"`
+        Print:  "review: updated existing thread $THREAD_ID on PR $PR_ID"
+      If not found:
+        Create: `az repos pr thread create --id "$PR_ID" --comment "$REVIEW_BODY"`
+        Print:  "review: posted new thread on PR $PR_ID"
+
+      **local-markdown:**
+      Look for `docs/workflow/<slug>/review-comment.md` (where slug is the most recent
+      workflow directory under docs/workflow/).
+      If it exists:
+        Overwrite the file with `REVIEW_BODY`.
+        Print: "review: updated docs/workflow/<slug>/review-comment.md"
+      If it does not exist:
+        Write `REVIEW_BODY` to `docs/workflow/<slug>/review-comment.md`.
+        Print: "review: created docs/workflow/<slug>/review-comment.md"
+
+      **jira:**
+      Search for a comment whose body starts with `<!-- unic-dlc-review -->`:
+        `jira issue comment list "$PR_ID" --output json`
+      If found (`COMMENT_ID`):
+        `jira issue comment update "$PR_ID" "$COMMENT_ID" --body "$REVIEW_BODY"`
+        Print: "review: updated existing comment $COMMENT_ID on issue $PR_ID"
+      If not found:
+        `jira issue comment add "$PR_ID" --body "$REVIEW_BODY"`
+        Print: "review: posted new comment on issue $PR_ID"
+
+      ---
+
+      ## Summary
+
+      After posting or updating, print a human-readable summary:
+
+      ```
+      unic-dlc-review complete.
+
+      Aspects reviewed: Code Quality · Test Coverage Adequacy · Silent Failure Patterns · Type Design Quality
+      PR: <PR_TITLE> (<PR_ID>)
+      Action: <created | updated> review comment
+
+      Findings:
+        Code Quality:           <N finding(s) | No findings>
+        Test Coverage Adequacy: <N finding(s) | No findings>
+        Silent Failure Patterns: <N finding(s) | No findings>
+        Type Design Quality:    <N finding(s) | No findings>
+      ```
diff --git a/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-triage.yaml b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-triage.yaml
new file mode 100644
index 00000000..6134ee6d
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/.archon/workflows/unic-dlc-triage.yaml
@@ -0,0 +1,47 @@
+name: unic-dlc-triage
+description: Read current issue states and produce a HANDOFF.md status snapshot.
+
+nodes:
+  - id: read-state
+    name: Read Issue States
+    type: prompt
+    depends_on: []
+    prompt: |
+      You are running the unic-archon-dlc triage workflow.
+
+      1. Read .archon/unic-dlc.config.json to determine the configured tracker and branching strategy.
+      2. Read docs/workflow/ROADMAP.md if it exists; note the last-updated phase.
+      3. List all issues from the configured tracker using the appropriate CLI:
+         - github: `gh issue list --json number,title,labels,state`
+         - ado:    `az boards work-item list --query "[*].{id:id,title:fields.\"System.Title\",state:fields.\"System.State\"}"`
+         - jira:   `jira issue list --output json`
+         - local-markdown: Read docs/issues/**/index.md and parse Status: lines
+      4. Group issues by their canonical state label.
+      5. Identify blockers: issues whose blocked_by references are still open.
+      6. List the five most recent ADR files in docs/adr/ (by filename order).
+      7. Determine the current lifecycle phase from ROADMAP.md or "unknown" if absent.
+      8. Output a JSON object with shape:
+         {
+           "phase": "<phase>",
+           "openIssues": { "<state>": ["<title>", ...] },
+           "blockers": ["<blocker description>", ...],
+           "recentDecisions": ["<adr-filename>", ...]
+         }
+
+  - id: produce-handoff
+    name: Produce HANDOFF.md
+    type: prompt
+    depends_on: [read-state]
+    prompt: |
+      Using the JSON snapshot from read-state:
+
+      1. Write HANDOFF.md at the repo root with exactly four sections:
+         ## Current Phase — the phase value
+         ## Open Issues by State — issues grouped by state label
+         ## Blockers — open issues whose blockers are still unresolved
+         ## Recent Decisions — ADR filenames from docs/adr/
+
+      2. Call updateRoadmap() to refresh docs/workflow/ROADMAP.md with the current phase.
+         The ROADMAP.md uses marker-delimited auto-generated regions — do not overwrite human content.
+
+      3. Print a short summary: "Triage complete. HANDOFF.md written. ROADMAP.md updated."
diff --git a/apps/claude-code/unic-archon-dlc/.claude-plugin/marketplace.json b/apps/claude-code/unic-archon-dlc/.claude-plugin/marketplace.json
index 503f88c4..9968a423 100644
--- a/apps/claude-code/unic-archon-dlc/.claude-plugin/marketplace.json
+++ b/apps/claude-code/unic-archon-dlc/.claude-plugin/marketplace.json
@@ -21,7 +21,7 @@
       "name": "unic-archon-dlc",
       "source": "./",
       "tags": ["productivity", "workflow", "ai-development"],
-      "version": "0.1.0"
+      "version": "0.1.2"
     }
   ]
 }
diff --git a/apps/claude-code/unic-archon-dlc/.claude-plugin/plugin.json b/apps/claude-code/unic-archon-dlc/.claude-plugin/plugin.json
index b17fec60..6baec753 100644
--- a/apps/claude-code/unic-archon-dlc/.claude-plugin/plugin.json
+++ b/apps/claude-code/unic-archon-dlc/.claude-plugin/plugin.json
@@ -1,6 +1,6 @@
 {
   "name": "unic-archon-dlc",
-  "version": "0.1.0",
+  "version": "0.1.2",
   "description": "A complete Archon-powered AI development lifecycle as an installable DLC pack — explore, plan, build, qa, cleanup, and triage workflows.",
   "author": {
     "name": "Unic AG",
diff --git a/apps/claude-code/unic-archon-dlc/AGENTS.md b/apps/claude-code/unic-archon-dlc/AGENTS.md
new file mode 100644
index 00000000..16033b04
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/AGENTS.md
@@ -0,0 +1,76 @@
+# AGENTS.md — unic-archon-dlc
+
+Guidance for any AI agent working inside this Plugin directory. `CLAUDE.md` in this directory is a symlink to this file.
+
+## What this Plugin is
+
+`unic-archon-dlc` is a Claude Code Plugin in the [`unic-agents-plugins`](../../../AGENTS.md) monorepo. It scaffolds an Archon-powered AI development lifecycle — six workflow DAGs (`explore`, `plan`, `build`, `qa`, `cleanup`, `triage`) plus agent-skill docs — into any Consumer project. Configuration is performed once via the `/unic-archon-dlc:setup` slash command. See [`CONTEXT.md`](CONTEXT.md) for the domain vocabulary (Session, Slug, PRD, Findings, Issues JSON, Nyquist map, yaml-gen, Setup).
+
+## Where to start
+
+Root docs (monorepo-wide conventions live here — pnpm scripts, Gitflow, SemVer, Conventional Commits, code conventions, LICENSE policy, cross-platform requirement):
+
+- [Root AGENTS.md](../../../AGENTS.md) — source of truth for cross-cutting rules
+- [Root CONTEXT.md](../../../CONTEXT.md) — monorepo-wide vocabulary (Plugin, Workspace Package, Release, Feature, Consumer)
+- [Root CONTEXT-MAP.md](../../../CONTEXT-MAP.md) — index of all bounded contexts in the repo
+- [Root docs/adr/](../../../docs/adr/) — monorepo-wide architecture decisions
+- [Root docs/process/](../../../docs/process/) — process and workflow guides
+
+This Plugin's own decisions:
+
+- [Plugin docs/adr/](docs/adr/) — Plugin-specific architecture decisions
+
+## Commands
+
+Plugin-specific pnpm scripts (run from this directory or with `pnpm --filter unic-archon-dlc <script>` from the repo root):
+
+```sh
+pnpm test                 # run node:test suite for this Plugin
+pnpm typecheck            # tsc --checkJs over this Plugin's .mjs sources
+pnpm bump <patch|minor|major>   # bump plugin.json version + promote CHANGELOG
+pnpm sync-version         # mirror plugin.json version into marketplace.json + package.json
+pnpm tag                  # create the unic-archon-dlc@<version> git tag locally
+pnpm verify:changelog     # check CHANGELOG entry for the current version
+```
+
+Monorepo-wide commands (`pnpm install`, `pnpm check`, `pnpm format`, `pnpm ci:check`) are documented in the [root AGENTS.md](../../../AGENTS.md).
+
+## Layout
+
+```tree
+.archon/                  # Archon assets installed into Consumer projects — workflow YAMLs and command stubs
+.claude-plugin/           # Plugin manifest (plugin.json) and marketplace listing
+commands/                 # Claude Code slash command definitions — only setup.md today
+lib/                      # Pure helper modules consumed by the setup command (ESM, // @ts-check)
+test/                     # node:test suites covering lib/ and command behaviour
+CONTEXT.md                # Domain vocabulary for the Archon-powered AI development lifecycle
+docs/                     # Plugin-specific documentation
+  adr/                    # Plugin Architecture Decision Records
+```
+
+## Plugin doctrines
+
+Load-bearing invariants. These either originate in a Plugin ADR or are policy decisions that are not obvious from the code.
+
+- **Setup is the sole entry point and is idempotent.** The `/unic-archon-dlc:setup` slash command is the one and only user-facing surface. Re-running it on a fully configured Consumer prints the current config; on a partial config it asks only for missing fields; on a fresh project it prompts for everything. See [ADR-0001](docs/adr/0001-setup-as-slash-command.md).
+- **Session is scoped by Slug.** Every Session artefact (Findings, PRD, Issues JSON, `build-<slug>.yaml`) is keyed by a single Slug. No cross-Session bleed. See [CONTEXT.md](CONTEXT.md).
+- **HANDOFF.md and ROADMAP.md are written exclusively by the triage workflow.** No other node, command, or workflow may write either file. Human-written content outside the `<!-- unic-archon-dlc:begin/end -->` markers in `ROADMAP.md` is never overwritten.
+- **The `## Agent skills` block in a Consumer's `CLAUDE.md` is auto-managed.** Setup writes content between `<!-- unic-archon-dlc:begin -->` and `<!-- unic-archon-dlc:end -->` markers. Everything outside the markers is preserved verbatim across re-runs.
+- **Slopcheck before build.** Every new package referenced in `package.json` is verified against the npm registry before any `code-red`/`code-green` node runs. Packages that fail are flagged `[ASSUMED]` and require explicit human approval.
+- **Nyquist map gates yaml-gen.** Every issue in Issues JSON must carry a `test_command` before the `yaml-gen` node generates `.archon/workflows/build-<slug>.yaml`.
+- **Dogfooding note.** This monorepo has had Setup run against it; the generated artefacts live under [`docs/agents/`](../../../docs/agents/) at the repo root and are managed by the marker-delimited block in the root [`AGENTS.md`](../../../AGENTS.md). Those files describe the target workflow and should be treated as the canonical agent guidance for this repo, not as a current-practice snapshot.
+
+## External dependencies
+
+- **Archon workflow engine, version ≥ 0.10**, present in the Consumer project. Setup verifies this before writing any artefacts.
+
+## Do not add
+
+- **Parallel-runner support before the linear path is operational.** The current happy path runs one Session at a time. Until that is rock-solid in real Consumers, do not add parallel-Session orchestration.
+- **Per-plugin variants of the Slug scheme.** Slug is monorepo-wide vocabulary; do not introduce a Plugin-specific Slug format.
+- **Consumer-side opt-out flags for individual workflow phases.** Until a real Consumer asks for it with a concrete use case, all six phases (`explore`, `plan`, `build`, `qa`, `cleanup`, `triage`) ship as one bundle.
+- **Direct edits to `HANDOFF.md` / `ROADMAP.md` from any non-triage node.** See doctrines above.
+
+## Plugin ADRs
+
+Plugin-specific architecture decisions live in [docs/adr/](docs/adr/).
diff --git a/apps/claude-code/unic-archon-dlc/CHANGELOG.md b/apps/claude-code/unic-archon-dlc/CHANGELOG.md
index 82bb0c61..3e626531 100644
--- a/apps/claude-code/unic-archon-dlc/CHANGELOG.md
+++ b/apps/claude-code/unic-archon-dlc/CHANGELOG.md
@@ -5,13 +5,39 @@
 ### Breaking
 - (none)
 
+### Added
+- (none)
+
+### Fixed
+- (none)
+
+## [0.1.2] — 2026-05-23
+
+### Breaking
+- (none)
+
+### Added
+- Updated `buildDomainDoc` multi-context branch in `lib/agent-docs-writer.mjs` so the generated `docs/agents/domain.md` notes that each context may keep its own `docs/adr/` for context-scoped decisions, and branches the trailing "How agents use this" paragraph by `isMulti` so the multi-context form points readers via `CONTEXT-MAP.md` and acknowledges both root and context-scoped `docs/adr/`. The wording is portable — no hardcoded path leaks into Consumer output. A node:test assertion in `test/install-agent-docs.test.mjs` guards the phrases in multi-context mode and confirms they are absent in single-context mode.
+
+### Fixed
+- (none)
+
+## [0.1.1] — 2026-05-23
+
 ### Added
 - Added `/unic-archon-dlc:setup` slash command for conversational plugin configuration
+- New `lib/dogfood-banner.mjs` module: exports `AGENT_DOC_BANNER`, `SKILLS_BLOCK_BANNER`, and `prependBanner()` — all banner strings in one place.
+- Every `docs/agents/*.md` file generated by `agent-docs-writer.mjs` now begins with `AGENT_DOC_BANNER`, signalling it is auto-generated and explaining how to regenerate it.
+- The `<!-- unic-archon-dlc:begin/end -->` block in `CLAUDE.md` now includes `SKILLS_BLOCK_BANNER` as its first line, making the auto-managed region visible in plain text (not only via HTML-comment markers).
+- Dogfood state in this repo updated: `docs/agents/*.md` and the `AGENTS.md` block now carry the banner.
 
 ### Removed
 - Removed `hooks/install.mjs` and `hooks` field from `plugin.json`
 
 ### Fixed
+- Build `run-build` node prompt now invokes the generated per-slug workflow by name (`archon workflow run unic-dlc-build-<slug>`) instead of the no-longer-supported `archon run <path>`
+- Cleanup workflow `run-triage` error message now says `archon workflow run failed` (matches the actual command); cleanup command doc references the by-name invocation instead of `archon run`
+- `docs/agents/workflow.md` (and the `agent-docs-writer.mjs` generator) now list all seven workflow DAGs — the missing `review` phase has been added alongside the six lifecycle phases. The `review` row's artifact column now covers both PR-comment trackers (github/ado/jira) and the `local-markdown` tracker (which writes `docs/workflow/<slug>/review-comment.md`); the install-agent-docs test anchors on the unique `/unic-dlc-review` command string so the row can't silently regress
 - Fixed stale reference in `CONTEXT.md`: Relationships section now credits the **Setup** slash command (not the deleted install hook) for writing config/docs into the target project
 - Fixed `repo_layout` default and valid-values columns in `README.md` configuration reference table to use `single-context` (as produced by `detectRepoLayout()`) instead of `single`
 - Fixed shell injection in `setup` command Step 5: `{ANSWERS_JSON}` is now substituted directly inside the `<<'EOJS'` heredoc instead of being assigned to a shell variable, so single quotes in e2e commands (e.g. `pnpm test --grep 'smoke'`) no longer break the assignment
@@ -22,6 +48,8 @@
 - Fixed silent discard of corrupt config in Step 2: an invalid-JSON config file now surfaces an `error` field in the output and stops setup with an actionable message, instead of silently mapping the `SyntaxError` to `STATE = 'fresh'` and overwriting the user's config
 - `runInstall`: corrupt config files (invalid JSON) now return a `stage: 'config'` error with an actionable message instead of silently discarding the existing config and overwriting it; file read errors (e.g. `EACCES`) are also surfaced as early returns
 - Added missing test for `stage: 'claude-md'` failure branch, test for corrupt-config parse error; removed always-passing placeholder test
+- Corrected dogfood banner regenerate hint from `/unic-archon-dlc-setup` to `/unic-archon-dlc:setup` (the actual slash-command name uses a colon, not a dash). Tightened `dogfood-banner.test.mjs` to assert the exact command string. Regenerated `docs/agents/*.md` carry the corrected banner.
+- `AGENT_DOC_BANNER` no longer references a non-existent "setup-runner"; it now points to the real entry point `runInstall()` in `lib/install-runner.mjs`. `SKILLS_BLOCK_BANNER` now names the slash command (`/unic-archon-dlc:setup`) explicitly so readers who land inside the marker block via search have unambiguous regenerate instructions. Regenerated `docs/agents/*.md` and the `AGENTS.md` block carry the updated wording; PRD canonical wording updated to match.
 
 ## [0.1.0] — 2026-05-15
 
diff --git a/apps/claude-code/unic-archon-dlc/CLAUDE.md b/apps/claude-code/unic-archon-dlc/CLAUDE.md
new file mode 120000
index 00000000..47dc3e3d
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/CLAUDE.md
@@ -0,0 +1 @@
+AGENTS.md
\ No newline at end of file
diff --git a/apps/claude-code/unic-archon-dlc/commands/setup.md b/apps/claude-code/unic-archon-dlc/commands/setup.md
index 17e84303..2be526fa 100644
--- a/apps/claude-code/unic-archon-dlc/commands/setup.md
+++ b/apps/claude-code/unic-archon-dlc/commands/setup.md
@@ -173,6 +173,8 @@ Parse the JSON output:
   ```
   unic-archon-dlc configured.
     config:    {result.configPath}
+    workflows: {result.workflowsCopied} workflow(s) copied
+    commands:  {result.commandsCopied} command(s) copied
     docs:      written
     CLAUDE.md: updated
   ```
diff --git a/apps/claude-code/unic-archon-dlc/docs/adr/0002-per-plugin-ralph-loops.md b/apps/claude-code/unic-archon-dlc/docs/adr/0002-per-plugin-ralph-loops.md
new file mode 100644
index 00000000..fd72d324
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/docs/adr/0002-per-plugin-ralph-loops.md
@@ -0,0 +1,24 @@
+# 0002. Each plugin has its own Ralph loop with its own ralph.yml and PROMPT.md
+
+**Status:** Superseded by [ADR-0009](0009-retire-ralph-adopt-archon-runner.md)
+
+Renumbered from monorepo-root ADR-0020 (2026-05).
+
+## Context
+
+Ralph Orchestrator implements specs one at a time in a loop. A single monorepo-wide loop would mix plugin-specific specs with workspace specs, making iteration focus unclear.
+
+## Decision
+
+Two levels of Ralph loops exist:
+
+1. **Monorepo root loop** (`pnpm ralph` at repo root): implements `docs/plans/` specs (workspace structure, shared packages, CI).
+2. **Per-plugin loop** (`pnpm --filter <name> ralph`): implements `apps/claude-code/<plugin>/docs/plans/` specs.
+
+Each level has its own `ralph.yml` and `PROMPT.md`.
+
+## Consequences
+
+- Adding a new plugin requires creating its own `ralph.yml` and `PROMPT.md`.
+- Running the root loop does not touch plugin-specific specs and vice versa.
+- Status markers (`**Status: done**`) are scoped per loop level.
diff --git a/apps/claude-code/unic-archon-dlc/docs/adr/0003-spec-template-format.md b/apps/claude-code/unic-archon-dlc/docs/adr/0003-spec-template-format.md
new file mode 100644
index 00000000..17e8eda7
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/docs/adr/0003-spec-template-format.md
@@ -0,0 +1,27 @@
+# 0003. Spec template format for Ralph-executable specs
+
+**Status:** Accepted (2025-04)
+
+Renumbered from monorepo-root ADR-0023 (2026-05).
+
+## Context
+
+Ralph Orchestrator needs specs to be unambiguous enough to execute deterministically. An ad hoc format leads to missing context, Ralph pausing for clarification, or incorrect implementations.
+
+## Decision
+
+All specs follow `docs/process/spec-template.md`, which mandates:
+
+- A header block: `Priority`, `Effort`, `Version impact`, `Depends on`, `Touches`
+- `## Current behaviour` with exact before-state snapshots (code or CLI output)
+- `## Target behaviour` with exact after-state
+- `## Implementation steps` with explicit before→after diffs or full file content
+- `## Verification` with shell commands and their expected output
+- `## Acceptance criteria` as checkboxes
+- `## Out of scope` bounding the change
+
+## Consequences
+
+- Specs are longer to write but faster for Ralph to execute without back-and-forth.
+- The "Out of scope" section prevents scope creep during automated implementation.
+- Specs that deviate from the template may cause Ralph to pause or implement incorrectly.
diff --git a/apps/claude-code/unic-archon-dlc/docs/adr/0004-ralph-atomic-iteration.md b/apps/claude-code/unic-archon-dlc/docs/adr/0004-ralph-atomic-iteration.md
new file mode 100644
index 00000000..fed3e85b
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/docs/adr/0004-ralph-atomic-iteration.md
@@ -0,0 +1,19 @@
+# 0004. Ralph implements one spec per iteration, then commits and stops
+
+**Status:** Superseded by [ADR-0009](0009-retire-ralph-adopt-archon-runner.md)
+
+Renumbered from monorepo-root ADR-0024 (2026-05).
+
+## Context
+
+Batching multiple specs into one Ralph iteration makes commits hard to review, complicates revert, and makes it harder to resume after a failure.
+
+## Decision
+
+Ralph's loop implements exactly one spec per run, creates one commit, and stops. Loop resumption reads `**Status: done**` markers in spec files to determine the next spec to implement. The loop never modifies `PROMPT.md` or `ralph.yml` to batch specs.
+
+## Consequences
+
+- Each PR from a Ralph run contains one logical unit of work.
+- The commit message references the spec number and title.
+- Resuming after a failed Ralph run is mechanical: remove the `**Status: done**` marker from the partially-implemented spec and restart.
diff --git a/apps/claude-code/unic-archon-dlc/docs/adr/0005-tdd-dispatch-by-version-impact.md b/apps/claude-code/unic-archon-dlc/docs/adr/0005-tdd-dispatch-by-version-impact.md
new file mode 100644
index 00000000..e9fd04cd
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/docs/adr/0005-tdd-dispatch-by-version-impact.md
@@ -0,0 +1,32 @@
+# 0005. Ralph uses `/tdd` for behavioral specs, direct implementation for structural ones, dispatched by `**Version impact:**`
+
+**Status:** Accepted (2026-05)
+
+Renumbered from monorepo-root ADR-0026 (2026-05).
+
+## Context
+
+The Ralph loop previously told the agent to "follow the Implementation steps exactly" for every spec. This produced horizontal slicing (building layer by layer as prescribed) rather than vertical slicing (a thin end-to-end slice that proves value immediately). It also left no explicit feedback loop before committing — the agent could implement and commit without ever running tests.
+
+Adding `/tdd` unconditionally would be awkward for structural specs (config files, release tooling, docs) where there is no meaningful unit of behavior to test first. A test asserting that `plugin.json` contains a field adds noise with no signal.
+
+## Decision
+
+`PROMPT.md` dispatches on the spec's `**Version impact:**` field:
+
+- **`none`** — implement directly; follow the Implementation steps as a recipe. No `/tdd`.
+- **`patch` / `minor` / `major`** — invoke `/tdd`; treat the Implementation steps as guidance (key files, rough order), not a recipe. The spec's Acceptance criteria are the target; the red-green-refactor cycle drives the path.
+
+`**Version impact:**` was chosen as the dispatch signal because it is already present in every spec, it reliably separates behavioral changes (which touch runtime code) from structural ones (which touch config, docs, or tooling), and it requires no new metadata.
+
+## Considered options
+
+- **Effort-gated dispatch** — switch to `/tdd` after N failed attempts. Rejected: this rewards thrashing rather than preventing it, and gives the agent no clear signal upfront.
+- **Universal `/tdd`** — apply to all specs including `none`. Rejected: a test asserting a config field exists adds noise and would fight the prescriptive implementation steps that structural specs genuinely need.
+- **New metadata field in spec frontmatter** — e.g. `**Implementation style: tdd|direct**`. Rejected: `**Version impact:**` already encodes the same distinction; a new field would duplicate information and drift.
+
+## Consequences
+
+- Specs with `**Version impact: none**` remain recipe-style; their Implementation steps should stay prescriptive.
+- Specs with a semver impact should be written acceptance-criteria-first going forward; detailed step-by-step instructions are demoted to guidance.
+- The always-on `pnpm check` block in Step 4 serves as the post-`/tdd` hygiene gate, catching formatting regressions the TDD loop does not cover.
diff --git a/apps/claude-code/unic-archon-dlc/docs/adr/0006-feature-runner-context-bundle.md b/apps/claude-code/unic-archon-dlc/docs/adr/0006-feature-runner-context-bundle.md
new file mode 100644
index 00000000..6cd14d81
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/docs/adr/0006-feature-runner-context-bundle.md
@@ -0,0 +1,45 @@
+# 0006. Feature Runner injects a scoped context bundle into every `/tdd` sub-agent invocation
+
+**Status:** Superseded by [ADR-0010](0010-retire-implement-feature-skill.md) (2026-05)
+
+Renumbered from monorepo-root ADR-0027 (2026-05).
+
+## Context
+
+The Feature Runner invokes `/tdd` as a non-interactive sub-agent (via the Agent tool) for each issue in a feature. The first draft of the Feature Runner PRD specified "the issue file content as context" as the sole input to each `/tdd` invocation.
+
+This is insufficient. `/tdd` is designed to be interactive: its planning phase asks the user to confirm interface changes and approve which behaviours to test before writing any code. In AFK mode there is no user to ask. Without additional context, `/tdd` reasons from a single vertical slice with no access to the "why" behind the feature, the architectural constraints that apply, or the vocabulary the codebase uses. This creates a risk of a correct-but-wrong implementation — code that satisfies the issue's literal description but diverges from the intent established during the grilling and PRD sessions.
+
+Matt Pocock's reference AFK loop (`afk.sh`) injects all issue files plus recent commits as a single prompt string before invoking `/tdd`, establishing the precedent that the agent needs broader context than a single work item.
+
+## Decision
+
+The Feature Runner assembles a **context bundle** for each `/tdd` sub-agent invocation:
+
+1. **Issue file** — `## What to build` and `## Acceptance criteria` serve as the pre-answered planning conversation (see ADR-0008).
+2. **PRD** — read from the feature's `docs/issues/<slug>/PRD.md` (the slug is the feature directory name; the PRD lives at a fixed path within it). Carries the shared vision from the grilling session and the "why" behind the feature. Without it, the agent lacks the context needed to judge correctness beyond the literal issue description.
+3. **Sibling issue files** — all other `NN-*.md` files in the feature directory. Provides dependency awareness and a "what is already resolved" signal without requiring the runner to summarise prior work.
+4. **Scoped CONTEXT.md** — the domain glossary for the feature's domain (see scoping rule below). Ensures test names and interface vocabulary match the project's language.
+5. **Scoped ADRs** — the architectural decisions constraining the implementation (see scoping rule below).
+6. **Recent commits** — the last 5 git commits. The grilling and PRD process typically produces changes to CONTEXT.md and ADRs that land in commits before the Feature Runner runs. These commits carry the ideation trail that informed the PRD.
+
+### ADR scoping rule
+
+ADRs and CONTEXT.md are scoped to the domain of the feature, not the monorepo:
+
+- **Plugin feature** (PRD references paths under `apps/claude-code/<plugin>/`) → inject that plugin's `docs/adr/` and `CONTEXT.md`.
+- **Repo/tooling feature** (PRD references paths outside `apps/`, e.g. `.claude/`, `docs/`, `packages/`) → inject the root `docs/adr/` and root `CONTEXT.md`.
+
+Scope is inferred by scanning the PRD for `apps/claude-code/<plugin>` path references. Root ADRs cover versioning, tagging, and CI tooling — they are noise for plugin implementation work and must not be injected into plugin feature runs.
+
+## Considered options
+
+- **Lazy discovery** — let `/tdd` explore the codebase and find ADRs and CONTEXT.md on its own. Rejected: `/tdd` does instruct the agent to use the domain glossary and respect ADRs, but in non-interactive sub-agent mode this exploration is unreliable. Injection is guaranteed; discovery is not.
+- **Issue file only** — the minimal approach from the first PRD draft. Rejected: `/tdd` loses the PRD's "why", the sibling issues' dependency signal, and the architectural constraints, all of which are needed to produce implementations that match the grilled intent.
+- **Inject everything** — all ADRs from all directories. Rejected: root ADRs are irrelevant to plugin work and add context noise without signal.
+
+## Consequences
+
+- The Feature Runner skill reads the PRD from the fixed slug-derived path `docs/issues/<slug>/PRD.md`. The `## Parent` link on issue files is informational for human readers; the runner does not parse it. Features without a `PRD.md` at that path cannot be run by the Feature Runner without manual intervention.
+- The Feature Runner skill must scan the PRD for `apps/claude-code/<plugin>` references to determine which CONTEXT.md and ADR directory to inject.
+- The context bundle grows with the number of sibling issues; for features with many issues, later invocations carry more sibling context than earlier ones. This is acceptable — it mirrors the growing "what is done" signal available in real commits.
diff --git a/apps/claude-code/unic-archon-dlc/docs/adr/0007-blocked-by-canonical-sequencing.md b/apps/claude-code/unic-archon-dlc/docs/adr/0007-blocked-by-canonical-sequencing.md
new file mode 100644
index 00000000..fc774dd7
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/docs/adr/0007-blocked-by-canonical-sequencing.md
@@ -0,0 +1,34 @@
+# 0007. `## Blocked by` is the canonical sequencing signal for Feature Runner issue execution
+
+**Status:** Accepted (2026-05)
+
+Renumbered from monorepo-root ADR-0028 (2026-05).
+
+## Context
+
+Issues in `docs/issues/<slug>/` are named with a numeric prefix (`NN-*.md`) produced by the `to-issues` skill, which publishes issues in dependency order so blockers get lower numbers. This makes numerical filename order a reliable proxy for execution order in practice.
+
+However, `to-issues` also records explicit dependency information in each issue's `## Blocked by` field. The numeric prefix is a UX convenience — it makes the dependency graph human-readable at a glance in a file browser. It is not an execution contract. The `to-issues` skill's "Blocked by" field is the canonical representation of the dependency graph: it can express non-linear dependencies that numerical order cannot (e.g. issue 03 blocking issue 02 after a user reorders slices during review).
+
+Treating numerical order as the execution contract would make the Feature Runner silently incorrect whenever `## Blocked by` and filename order diverge — a failure mode that would be invisible until a downstream issue ran on a broken foundation.
+
+## Decision
+
+The Feature Runner builds a **topological order** from `## Blocked by` references before executing any issue. Numerical filename order is used only as a tiebreaker when two issues have no dependency relationship between them.
+
+If `## Blocked by` references conflict with numerical filename order (i.e. a lower-numbered issue declares a blocker that is a higher-numbered issue), the Feature Runner halts with an error and surfaces the conflict to the user rather than proceeding in the wrong order. Silent execution on a potentially wrong order is not acceptable.
+
+Issues with `## Blocked by: None` (or equivalent) have no predecessors and may be placed anywhere in the topological order consistent with their number.
+
+## Considered options
+
+- **Numerical order only** — simpler to implement; no graph parsing required. Rejected: not an execution contract; silently wrong when user reorders slices or when `to-issues` produces a non-linear dependency graph.
+- **`## Blocked by` order, silent fallback to numerical on conflict** — avoids halting. Rejected: a conflict between the two signals indicates a malformed feature (either the issue was hand-edited or `to-issues` produced unexpected output); proceeding silently would compound the error.
+- **`## Blocked by` order, halt on conflict** — chosen. Forces the human to resolve ambiguity before the autonomous run begins, preventing downstream issues from inheriting a broken foundation.
+
+## Consequences
+
+- The Feature Runner skill must parse `## Blocked by` fields and construct a dependency graph before beginning execution.
+- Features where `## Blocked by` and numerical order disagree will not run until the conflict is resolved by the developer.
+- The `to-issues` skill's practice of publishing issues in dependency order (blockers first) remains a useful convention that keeps numerical order and the dependency graph aligned in the common case.
+- The dependency graph also reveals which issues are parallelisable (those with `## Blocked by: None` and no dependents). The Feature Runner serialises all execution regardless — see the Feature Runner PRD Out of Scope for the rationale.
diff --git a/apps/claude-code/unic-archon-dlc/docs/adr/0008-feature-runner-afk-invocation.md b/apps/claude-code/unic-archon-dlc/docs/adr/0008-feature-runner-afk-invocation.md
new file mode 100644
index 00000000..2d0cfcc1
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/docs/adr/0008-feature-runner-afk-invocation.md
@@ -0,0 +1,38 @@
+# 0008. Feature Runner invokes `/tdd` non-interactively; issue acceptance criteria replace the planning phase
+
+**Status:** Superseded by [ADR-0010](0010-retire-implement-feature-skill.md) (2026-05)
+
+Renumbered from monorepo-root ADR-0029 (2026-05).
+
+## Context
+
+`/tdd`'s planning phase is interactive: before writing any code it asks the user to confirm interface changes, prioritise which behaviours to test, and approve the plan. This is by design — it prevents the agent from outrunning its headlights on ambiguous requirements.
+
+The Feature Runner's core use case is autonomous, overnight execution (composable with `/loop`). There is no user present to answer planning questions. Invoking `/tdd` as a sub-agent via the Agent tool means there is no TTY — interactive prompts cannot be issued and the planning phase cannot execute as designed.
+
+Matt Pocock's reference AFK loop (`afk.sh`) resolves this by running Claude with `--print` (non-interactive mode) and injecting issue files as the implicit plan. The Agent tool in Claude Code is the equivalent mechanism: sub-agents run without a TTY and must infer their plan from the provided context.
+
+## Decision
+
+The Feature Runner invokes `/tdd` via the Agent tool (non-interactive). The issue's `## Acceptance criteria` section serves as the pre-answered planning conversation:
+
+- `## What to build` answers "what interface changes are needed"
+- `## Acceptance criteria` answers "which behaviours to test" and "what does done look like"
+
+Because issues are produced by `to-issues` (which slices the PRD vertically and quizzes the user on the breakdown) and then reviewed by the user before reaching `ready-for-agent`, the acceptance criteria represent a human-approved definition of done. The interactive planning phase is not bypassed in substance — it was completed during the grilling and issue-writing pipeline; the Feature Runner simply does not repeat it at runtime.
+
+`/tdd` is not modified. No AFK flag or non-interactive variant is introduced.
+
+## Considered options
+
+- **Fork `/tdd` into a non-interactive variant** (`/tdd-afk` or similar) — would allow explicit suppression of planning prompts. Rejected: creates a maintenance burden; the two variants would diverge over time; the Agent tool already provides non-interactive execution without any skill changes.
+- **Add an AFK flag to `/tdd`** — e.g. a frontmatter option or a prompt prefix that skips planning. Rejected: couples the `/tdd` skill to the Feature Runner's invocation model; `to-issues` already produces the information that the planning phase would gather.
+- **Non-interactive Agent tool invocation with injected context bundle** — chosen. No skill modifications required; the planning information is supplied via the context bundle (ADR-0006) rather than elicited at runtime.
+
+## Consequences
+
+- Issues that reach the Feature Runner must have specific, human-reviewed `## Acceptance criteria`. Vague criteria (e.g. "the feature works correctly") remove the planning substitute and leave `/tdd` without a concrete definition of done.
+- The `to-issues` + `/triage` → `ready-for-agent` pipeline is load-bearing: it is the point at which the planning conversation occurs. The Feature Runner depends on that pipeline having been followed correctly.
+- `/tdd` remains unchanged and continues to work interactively when invoked directly by the user.
+- When the queue is empty (no `ready-for-agent` features remain), the Feature Runner outputs `LOOP_COMPLETE` as its stop signal. This is what makes `/loop /implement-feature` composable for overnight draining: the `/loop` skill catches `LOOP_COMPLETE` and terminates the loop cleanly rather than spinning on an empty queue. This mirrors the Spec Runner's `completion_promise: LOOP_COMPLETE` in `ralph.yml` and Matt Pocock's `<promise>NO MORE TASKS</promise>` in `afk.sh`.
+- The Feature Runner's prompt template explicitly instructs the sub-agent to invoke the `tdd` skill via the Skill tool, rather than relying on auto-discovery via trigger phrases. This makes the procedural-guidance load deterministic across Claude Code surfaces.
diff --git a/apps/claude-code/unic-archon-dlc/docs/adr/0009-retire-ralph-adopt-archon-runner.md b/apps/claude-code/unic-archon-dlc/docs/adr/0009-retire-ralph-adopt-archon-runner.md
new file mode 100644
index 00000000..553486f6
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/docs/adr/0009-retire-ralph-adopt-archon-runner.md
@@ -0,0 +1,30 @@
+# 0009. Retire ralph-orchestrator; adopt unic-archon-dlc as the Feature Runner
+
+**Status:** Accepted (2026-05)
+
+Renumbered from monorepo-root ADR-0030 (2026-05).
+
+## Context
+
+`ralph-orchestrator` (`@ralph-orchestrator/ralph-cli`) was used as the Spec Runner for all `docs/plans/` specs at both the monorepo root and per-plugin level (see [ADR-0002](0002-per-plugin-ralph-loops.md), [ADR-0004](0004-ralph-atomic-iteration.md)). By 2026-05 all 18 monorepo-level specs (00–17) and all prior plugin specs were marked done. The `docs/plans/` format was a bootstrap mechanism, not a permanent intake path.
+
+Two factors drove the retirement:
+
+1. **Windows incompatibility** — `ralph-cli` has no Windows support. CI runs on Windows (see [ADR-0014](../../../../../docs/adr/0014-ci-matrix-three-os-two-node.md)); contributors on Windows could not run the spec loop locally.
+2. **Strategic direction** — `unic-archon-dlc` is being developed as an Archon-powered AI development lifecycle harness that this monorepo both ships and consumes. Its `unic-dlc-build` workflow is the intended long-term runner; converging on one tool reduces the harness surface.
+
+## Decision
+
+- Remove `@ralph-orchestrator/ralph-cli` from `package.json` and `pnpm-workspace.yaml`.
+- Delete all `ralph.yml`, `PROMPT.md`, and `.ralph/` artifacts from the repo root and all plugin directories.
+- Remove the `pnpm ralph` and per-plugin `ralph` scripts.
+- Retire the `docs/plans/` spec format; specs 12–16 for `pr-review` are the last batch.
+- Remove **Spec** and **Spec Runner** from `CONTEXT.md`; the **Feature Runner** (backed by `unic-dlc-build`) is the sole runner going forward.
+- In the interim, individual specs are implemented manually via `/tdd`.
+
+## Consequences
+
+- No CLI wrapper exists for running specs during the transition; developers invoke `/tdd` directly.
+- Future work items enter through the issue tracker as Features, not as `docs/plans/` spec files.
+- `unic-dlc-build` must be production-ready before fully automated AFK runs resume.
+- ADR-0002 and ADR-0004 are superseded by this decision.
diff --git a/apps/claude-code/unic-archon-dlc/docs/adr/0010-retire-implement-feature-skill.md b/apps/claude-code/unic-archon-dlc/docs/adr/0010-retire-implement-feature-skill.md
new file mode 100644
index 00000000..b83c427b
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/docs/adr/0010-retire-implement-feature-skill.md
@@ -0,0 +1,29 @@
+# 0010. Retire the `/implement-feature` skill; Feature Runner backed solely by `unic-dlc-build`
+
+**Status:** Accepted (2026-05)
+
+Renumbered from monorepo-root ADR-0031 (2026-05).
+
+## Context
+
+[ADR-0009](0009-retire-ralph-adopt-archon-runner.md) retired `ralph-orchestrator` and declared `unic-dlc-build` (shipped by `unic-archon-dlc`) the long-term Feature Runner. As an interim measure while `unic-dlc-build` matured, a `/implement-feature` Claude Code skill was added at `.claude/skills/implement-feature/`. It assembled a context bundle ([ADR-0006](0006-feature-runner-context-bundle.md)) and invoked `/tdd` as a non-interactive sub-agent for each `ready-for-agent` issue ([ADR-0008](0008-feature-runner-afk-invocation.md)).
+
+In practice the skill was not adopted:
+
+1. **Single-purpose duplication** — it replicated, in skill form, exactly what `unic-dlc-build` is built to do. Maintaining two runners doubled the surface for prompt drift and made it unclear which one was canonical.
+2. **Agent-tool invocation is fragile** — driving `/tdd` through `Agent` with `subagent_type: general-purpose` cannot match the affordances of a dedicated harness (worktree management, retry semantics, observability). Treating the interim as a real path delayed the move to `unic-dlc-build`.
+3. **Capture path conflict** — the companion `/inbox` slash command duplicated `triage`'s capture entry point. Ideas now go straight to the issue tracker (or to `triage`), so the inbox-then-grill staging area is no longer needed.
+
+## Decision
+
+- Remove `.claude/skills/implement-feature/` (the skill directory, references, and runner output formats).
+- Remove `.claude/commands/inbox.md` and `docs/inbox/README.md`. Existing files under `docs/inbox/` remain as historical artifacts; no new ones are captured there.
+- Designate `unic-dlc-build` (via `unic-archon-dlc`) as the **sole** Feature Runner implementation going forward. Until `unic-dlc-build` is wired into this repo, individual issues are implemented manually via `/tdd`, as already stated in ADR-0009.
+- [ADR-0006](0006-feature-runner-context-bundle.md) and [ADR-0008](0008-feature-runner-afk-invocation.md) are superseded by this decision. Their context-bundle and AFK-invocation choices described `/implement-feature`'s internals, not portable design constraints; `unic-dlc-build` is free to make its own decisions about both.
+
+## Consequences
+
+- `AGENTS.md`, `CONTRIBUTING.md`, `docs/process/ai-development.md`, `docs/process/development-workflow.md`, and `docs/agents/feature-runner.md` are updated to remove references to `/implement-feature` and `/inbox` and to describe the current state (manual `/tdd`, `unic-dlc-build` as the runner).
+- The Feature Runner concept (as defined in root `CONTEXT.md`) is unchanged — it remains "the skill that implements a Feature's issues end-to-end in one worktree, branch, and pull request. Backed by `unic-dlc-build`." Only the interim Claude Code skill implementation is removed.
+- [ADR-0007](0007-blocked-by-canonical-sequencing.md) is **not** superseded — it defines the canonical sequencing signal (`## Blocked by`) for any Feature Runner implementation, including `unic-dlc-build`.
+- Captured ideas under `docs/inbox/*.md` are not migrated automatically. They graduate to issues only when someone triages them.
diff --git a/apps/claude-code/unic-archon-dlc/docs/adr/README.md b/apps/claude-code/unic-archon-dlc/docs/adr/README.md
index 90224fcf..745e2429 100644
--- a/apps/claude-code/unic-archon-dlc/docs/adr/README.md
+++ b/apps/claude-code/unic-archon-dlc/docs/adr/README.md
@@ -4,6 +4,15 @@ Plugin-scoped architectural decisions. Repo-wide decisions live in `docs/adr/` a
 
 ## Index
 
-| ID   | Title                                                           | Status   |
-| ---- | --------------------------------------------------------------- | -------- |
-| 0001 | Setup is a slash command delegating to `lib/install-runner.mjs` | Accepted |
+| ID   | Title                                                                                    | Status                 |
+| ---- | ---------------------------------------------------------------------------------------- | ---------------------- |
+| 0001 | Setup is a slash command delegating to `lib/install-runner.mjs`                          | Accepted               |
+| 0002 | Each plugin has its own Ralph loop with its own ralph.yml and PROMPT.md                  | Superseded by ADR-0009 |
+| 0003 | Spec template format for Ralph-executable specs                                          | Accepted               |
+| 0004 | Ralph implements one spec per iteration, then commits and stops                          | Superseded by ADR-0009 |
+| 0005 | `/tdd` for behavioral specs, direct for structural ones, dispatched by `Version impact:` | Accepted               |
+| 0006 | Feature Runner injects a scoped context bundle into every `/tdd` sub-agent invocation    | Superseded by ADR-0010 |
+| 0007 | `## Blocked by` is the canonical sequencing signal for Feature Runner issue execution    | Accepted               |
+| 0008 | Feature Runner invokes `/tdd` non-interactively; acceptance criteria replace planning    | Superseded by ADR-0010 |
+| 0009 | Retire ralph-orchestrator; adopt unic-archon-dlc as the Feature Runner                   | Accepted               |
+| 0010 | Retire the `/implement-feature` skill; Feature Runner backed solely by `unic-dlc-build`  | Accepted               |
diff --git a/apps/claude-code/unic-archon-dlc/docs/plans/00-setup-slash-command.md b/apps/claude-code/unic-archon-dlc/docs/plans/00-setup-slash-command.md
deleted file mode 100644
index 2cce12e5..00000000
--- a/apps/claude-code/unic-archon-dlc/docs/plans/00-setup-slash-command.md
+++ /dev/null
@@ -1,97 +0,0 @@
-# 00. Setup slash command
-
-**Status: done**
-
-- Priority: P1
-- Effort: M
-- Version impact: minor (replaces the manual install path with a first-class slash command)
-- Depends on: —
-- Touches: `commands/setup.md` (new), `lib/install-runner.mjs` (new), `lib/archon-check.mjs` (new), `hooks/install.mjs` (deleted), `.claude-plugin/plugin.json`, `README.md`, `CHANGELOG.md`, `docs/adr/0001-setup-as-slash-command.md`, `docs/adr/README.md` (new), `docs/plans/README.md` (new), `CONTEXT.md` (already updated)
-
-## Context
-
-The plugin advertises a slash-command-first install flow (`README.md:135` shows `/unic-dlc-install`) but no slash commands exist at the plugin root yet. The current entry point is `hooks/install.mjs`, a readline script declared under `"hooks": { "install": ... }` in `plugin.json` — but Claude Code does not auto-execute install hooks, so users must run `node ${CLAUDE_PLUGIN_ROOT}/hooks/install.mjs` from a terminal. That path is undocumented in the README and not the experience users expect after installing a Claude Code plugin.
-
-This spec ships the first real slash command for unic-archon-dlc and the precedent for the other six workflow commands when they migrate from `.archon/commands/` (Archon workflow templates) to actual Claude Code slash commands. The architectural shape — slash command delegates to a pure lib module, no terminal CLI wrapper — is locked by [ADR-0001](../adr/0001-setup-as-slash-command.md).
-
-## Current behaviour
-
-- `hooks/install.mjs` exists as an interactive readline script with four behaviours: fresh-install (ask all), partial-fill (ask missing only), already-configured (print and exit), and `--reconfigure` (re-prompt all).
-- The script runs `checkArchon()` first (preflight + known-incompatible-versions warning) and exits with code 1 on missing binary.
-- It calls `exploreProject()` to auto-detect git remote, then `loadConfig()` for any existing config, then prompts with `node:readline/promises`.
-- After collection it writes `.archon/unic-dlc.config.json`, calls `writeAgentDocs()` to generate `docs/agents/*.md`, and calls `updateAgentSkillsBlock()` to merge a marker-delimited `## Agent skills` block into `CLAUDE.md`.
-- `plugin.json` declares `"hooks": { "install": "hooks/install.mjs" }`. Claude Code does not execute this hook on plugin install.
-- No `commands/` directory exists at the plugin root; no slash command is registered for the plugin.
-- The README's quick-start advertises `/unic-dlc-install` — this command does not exist.
-
-## Target behaviour
-
-- A `commands/setup.md` slash command exists and is invoked as `/unic-archon-dlc:setup`.
-- The command conducts the configuration conversation (tracker / branching / e2e command); it does not write files directly.
-- Filesystem writes are delegated to `lib/install-runner.mjs`, exporting `runInstall(projectDir, partialAnswers)`. The function merges `{ ...defaults, ...existing, ...partialAnswers }` (matching `hooks/install.mjs:146`), enforces the mandatory-fields invariant after merge (throws a typed error if absent), and writes config + agent docs + CLAUDE.md block in the same order and with the same per-step error semantics as the existing hook.
-- Archon preflight is performed by `lib/archon-check.mjs`, exporting a pure `checkArchon()` returning `{ ok: true, version } | { ok: false, code: 'enoent' | 'incompatible' | 'other', message }`. No `process.exit`. The slash command runs this first and bails with the friendly message on `ok: false`.
-- Idempotency follows the hybrid model from the design grilling: fresh→ask-all; partial→ask-only-missing; full + no args→print summary and exit; `setup reconfigure`→re-prompt-all; `setup <free-form intent>`→targeted tweak via conversational interpretation.
-- `hooks/install.mjs` is deleted; the `"hooks"` field is removed from `plugin.json`.
-- `README.md` quick-start references `/unic-archon-dlc:setup`, not `/unic-dlc-install`.
-- `CHANGELOG.md` has an `Added` / `Removed` entry under `[Unreleased]`.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `commands/setup.md` | New. Thin orchestrator markdown: argument parse, archon preflight, config discovery, conversational prompting, `runInstall` invocation, friendly result rendering. |
-| `lib/install-runner.mjs` | New. Extract the post-prompt body of `hooks/install.mjs:128–187` into `runInstall(projectDir, partialAnswers)`. Pure with respect to user input; performs I/O on disk only. Includes `repo_layout` detection (from `detectRepoLayout`) and the default labels lookup. |
-| `lib/archon-check.mjs` | New. Extract `hooks/install.mjs:44–67` into a result-returning function. Move `INCOMPATIBLE_ARCHON_VERSIONS` into the module. |
-| `test/install-runner.test.mjs` | New. Covers fresh / partial / full / reconfigure paths; mandatory-field enforcement; merge precedence. |
-| `test/archon-check.test.mjs` | New. Covers `enoent`, version OK, version-incompatible, generic error. |
-| `hooks/install.mjs` | Delete. |
-| `.claude-plugin/plugin.json` | Remove the `"hooks"` field. |
-| `README.md` | Replace `/unic-dlc-install` (and its surrounding paragraph at lines 131–140) with `/unic-archon-dlc:setup`. Drop mentions of "install hook"; describe the slash command. |
-| `CHANGELOG.md` | `Added: /unic-archon-dlc:setup slash command.` `Removed: hooks/install.mjs and "hooks" field from plugin.json.` |
-| `docs/adr/README.md` | New. Index file listing ADR-0001. |
-| `docs/plans/README.md` | New. Index file listing this spec. |
-| `CONTEXT.md` | Already updated with Setup, Claude Code slash command, and Archon workflow command template entries. |
-
-## Implementation steps
-
-1. **Extract `lib/archon-check.mjs`.** Copy the body of `checkArchon()`, replace `process.exit(1)` with a returned result object, move `INCOMPATIBLE_ARCHON_VERSIONS` into the module. Add `test/archon-check.test.mjs` covering the four result branches via a stubbable `execFileSync`.
-2. **Extract `lib/install-runner.mjs`.** Move the merge logic (`hooks/install.mjs:135–187`) into `runInstall(projectDir, partialAnswers)`. The function does *not* prompt and does *not* call `checkArchon` — both are the slash command's responsibility. It reads existing config via `loadConfig`, performs the merge, validates mandatory fields, writes config, calls `writeAgentDocs`, calls `updateAgentSkillsBlock`. Returns `{ ok: true, configPath, wroteDocs, wroteClaudeMd } | { ok: false, stage, message }`. Add `test/install-runner.test.mjs`.
-3. **Write `commands/setup.md`.** The markdown should: parse `$ARGUMENTS` for `reconfigure` or free-form intent; run `checkArchon` and bail with the friendly message on failure; call `exploreProject` and `loadConfig` to discover state; branch on (existing-config × arguments) per the hybrid idempotency table; collect any needed answers via conversational prompts; call `runInstall(cwd, answers)`; print the success summary or the staged-error message. Reference ADR-0001 in a one-line note at the top.
-4. **Delete `hooks/install.mjs`.** Remove the file. There is no migration shim — the readline path was never documented in the README.
-5. **Remove `"hooks"` from `.claude-plugin/plugin.json`.**
-6. **Update `README.md`.** Replace the quick-start `/unic-dlc-install` block with a `/unic-archon-dlc:setup` block. Remove references to "install hook" elsewhere in the file.
-7. **Add `CHANGELOG.md` entries** under `[Unreleased]`.
-8. **Create the two index files** (`docs/adr/README.md`, `docs/plans/README.md`) so the directories have discoverable contents.
-
-## Verification
-
-- `pnpm --filter unic-archon-dlc test` passes (covers both new lib modules).
-- `pnpm typecheck` passes (no JSDoc regressions).
-- `pnpm ci:check` passes (Biome + Prettier).
-- Manual smoke test in a clean target project:
-  - `/unic-archon-dlc:setup` on a fresh project → asks three questions → `.archon/unic-dlc.config.json`, `docs/agents/*.md`, and `CLAUDE.md` `## Agent skills` block all present.
-  - `/unic-archon-dlc:setup` re-run with full config → prints current values, exits without writing.
-  - `/unic-archon-dlc:setup reconfigure` → re-prompts all three questions, overwrites config.
-  - `/unic-archon-dlc:setup` with `archon` removed from PATH → friendly "archon not on PATH" message, no partial writes.
-  - `/unic-archon-dlc:setup change branching to github-flow` (full config existing) → updates one field, leaves others, writes config.
-
-## Acceptance criteria
-
-- `commands/setup.md` exists and is the only documented entry point for configuring the plugin.
-- `hooks/install.mjs` no longer exists.
-- `plugin.json` has no `hooks` field.
-- `lib/install-runner.mjs` and `lib/archon-check.mjs` exist with tests, both pass.
-- The merge precedence in `runInstall` is `defaults < existing < partialAnswers`, matching the deleted hook byte-for-byte for that block.
-- `runInstall` returns `{ ok: false, stage: 'validate', message }` (not `process.exit`) when mandatory fields are missing after merge.
-- `checkArchon` returns a result object (not `process.exit`) for every branch.
-- README quick-start references `/unic-archon-dlc:setup`.
-- CHANGELOG has `Added` and `Removed` entries.
-- ADR-0001 is referenced from `commands/setup.md` and `docs/adr/README.md`.
-
-## Out of scope
-
-- Migrating `.archon/commands/*` (Archon workflow templates) into Claude Code slash commands at `commands/`. Separate spec.
-- Wiring archon preflight into workflow commands. ADR-0001 explicitly leaves preflight in setup only.
-- `.gitignore` changes for `.archon/`. Config is team config and presumed committed; transient state lives under `$ARTIFACTS_DIR`.
-- A non-interactive CLI wrapper around `runInstall()` for CI/scripted use. If a future caller needs it, a 5-line `bin/setup-cli.mjs` is trivial; do not pre-build for it.
-- Changing the slash command name later (e.g. `:configure` or `:init`). Locked in ADR-0001 / Q4 of the grilling session.
diff --git a/apps/claude-code/unic-archon-dlc/docs/plans/README.md b/apps/claude-code/unic-archon-dlc/docs/plans/README.md
deleted file mode 100644
index d275020d..00000000
--- a/apps/claude-code/unic-archon-dlc/docs/plans/README.md
+++ /dev/null
@@ -1,9 +0,0 @@
-# unic-archon-dlc spec roadmap
-
-Plugin-scoped specs for unic-archon-dlc. The monorepo-level Spec Runner roadmap (specs 00–14) lives at the repo root in `docs/plans/`. Plugin-specific specs follow the same 8-section template documented in the root `CLAUDE.md` under *Spec-driven development*.
-
-## Status
-
-| #  | Spec                                          | Status | Depends on |
-| -- | --------------------------------------------- | ------ | ---------- |
-| 00 | Setup slash command (`/unic-archon-dlc:setup`) | done   | —          |
diff --git a/apps/claude-code/unic-archon-dlc/lib/agent-docs-writer.mjs b/apps/claude-code/unic-archon-dlc/lib/agent-docs-writer.mjs
index 5b01c292..55443356 100644
--- a/apps/claude-code/unic-archon-dlc/lib/agent-docs-writer.mjs
+++ b/apps/claude-code/unic-archon-dlc/lib/agent-docs-writer.mjs
@@ -1,6 +1,7 @@
 // @ts-check
 import { mkdirSync, readFileSync, writeFileSync } from 'node:fs'
 import { join, relative } from 'node:path'
+import { AGENT_DOC_BANNER, prependBanner, SKILLS_BLOCK_BANNER } from './dogfood-banner.mjs'
 
 const AGENT_SKILLS_BEGIN = '<!-- unic-archon-dlc:begin -->'
 const AGENT_SKILLS_END = '<!-- unic-archon-dlc:end -->'
@@ -9,7 +10,7 @@ const AGENT_SKILLS_LINKS = `- [issue-tracker.md](docs/agents/issue-tracker.md) 
 - [labels.md](docs/agents/labels.md) — three-tier label taxonomy: state, type, priority
 - [branching.md](docs/agents/branching.md) — branching strategy, branch names, PR targets
 - [domain.md](docs/agents/domain.md) — single-context vs multi-context, CONTEXT.md and ADR locations
-- [workflow.md](docs/agents/workflow.md) — six workflow phases, artifact outputs, docs/workflow/ paths`
+- [workflow.md](docs/agents/workflow.md) — seven workflow phases, artifact outputs, docs/workflow/ paths`
 
 /**
  * @typedef {import('./labels-config.mjs').LabelMapping} LabelMapping
@@ -37,11 +38,11 @@ export function writeAgentDocs(projectDir, config) {
 	const dir = join(projectDir, 'docs', 'agents')
 	mkdirSync(dir, { recursive: true })
 
-	writeFileSync(join(dir, 'issue-tracker.md'), buildIssueTrackerDoc(config))
-	writeFileSync(join(dir, 'labels.md'), buildLabelsDoc(config))
-	writeFileSync(join(dir, 'branching.md'), buildBranchingDoc(config))
-	writeFileSync(join(dir, 'domain.md'), buildDomainDoc(config, projectDir))
-	writeFileSync(join(dir, 'workflow.md'), buildWorkflowDoc())
+	writeFileSync(join(dir, 'issue-tracker.md'), prependBanner(AGENT_DOC_BANNER, buildIssueTrackerDoc(config)))
+	writeFileSync(join(dir, 'labels.md'), prependBanner(AGENT_DOC_BANNER, buildLabelsDoc(config)))
+	writeFileSync(join(dir, 'branching.md'), prependBanner(AGENT_DOC_BANNER, buildBranchingDoc(config)))
+	writeFileSync(join(dir, 'domain.md'), prependBanner(AGENT_DOC_BANNER, buildDomainDoc(config, projectDir)))
+	writeFileSync(join(dir, 'workflow.md'), prependBanner(AGENT_DOC_BANNER, buildWorkflowDoc()))
 }
 
 /**
@@ -53,7 +54,8 @@ export function writeAgentDocs(projectDir, config) {
 export function updateAgentSkillsBlock(projectDir) {
 	const claudePath = join(projectDir, 'CLAUDE.md')
 
-	const block = `## Agent skills\n\n${AGENT_SKILLS_BEGIN}\n${AGENT_SKILLS_LINKS}\n${AGENT_SKILLS_END}`
+	const innerBlock = `${SKILLS_BLOCK_BANNER}\n\n${AGENT_SKILLS_LINKS}`
+	const block = `## Agent skills\n\n${AGENT_SKILLS_BEGIN}\n${innerBlock}\n${AGENT_SKILLS_END}`
 
 	let content
 	try {
@@ -71,7 +73,7 @@ export function updateAgentSkillsBlock(projectDir) {
 
 	if (beginIdx !== -1 && endIdx !== -1) {
 		// Replace only the content between markers (inclusive)
-		content = `${content.slice(0, beginIdx) + AGENT_SKILLS_BEGIN}\n${AGENT_SKILLS_LINKS}\n${content.slice(endIdx)}`
+		content = `${content.slice(0, beginIdx) + AGENT_SKILLS_BEGIN}\n${innerBlock}\n${content.slice(endIdx)}`
 		writeFileSync(claudePath, content)
 		return
 	}
@@ -222,7 +224,7 @@ ${
 		? `This repository uses **multi-context** layout. Each package/app has its own \`CONTEXT.md\` file. A \`CONTEXT-MAP.md\` at the repo root maps each context to its location.
 
 - **Context map:** \`${relative(projectDir, join(projectDir, 'CONTEXT-MAP.md'))}\`
-- **ADRs:** \`docs/adr/\` (repo-level decisions)`
+- **ADRs:** monorepo-wide decisions live in root \`docs/adr/\`; each context may also keep its own \`docs/adr/\` for decisions scoped to that context.`
 		: `This repository uses **single-context** layout. One \`CONTEXT.md\` file lives at the repo root.
 
 - **Domain context:** \`CONTEXT.md\`
@@ -231,14 +233,18 @@ ${
 
 ## How agents use this
 
-Every agent working in this repo should read \`CONTEXT.md\` (and the ADRs in \`docs/adr/\`) before proposing terminology changes or architectural decisions.
+${
+	isMulti
+		? `Every agent working in this repo should read the relevant \`CONTEXT.md\` (located via \`CONTEXT-MAP.md\`) and the ADRs in root \`docs/adr/\` plus any context-scoped \`docs/adr/\` before proposing terminology changes or architectural decisions.`
+		: `Every agent working in this repo should read \`CONTEXT.md\` (and the ADRs in \`docs/adr/\`) before proposing terminology changes or architectural decisions.`
+}
 `
 }
 
 function buildWorkflowDoc() {
 	return `# Workflow Phases
 
-unic-archon-dlc ships six Archon workflow YAML DAGs. Each phase produces persistent artifacts committed to \`docs/workflow/<slug>/\`.
+unic-archon-dlc ships seven Archon workflow YAML DAGs. Six are lifecycle phases (explore → plan → build → qa → cleanup → triage) that produce persistent artifacts committed to \`docs/workflow/<slug>/\`. The \`review\` workflow is on-demand and posts (or updates) a structured review on the current PR via the configured tracker backend.
 
 | Phase | Command | Artifact outputs |
 |-------|---------|-----------------|
@@ -248,6 +254,7 @@ unic-archon-dlc ships six Archon workflow YAML DAGs. Each phase produces persist
 | qa | \`/unic-dlc-qa <slug>\` | merged PR |
 | cleanup | \`/unic-dlc-cleanup <slug>\` | \`docs/workflow/<slug>/arch-review.md\` |
 | triage | \`/unic-dlc-triage\` | \`HANDOFF.md\`, \`docs/workflow/ROADMAP.md\` |
+| review | \`/unic-dlc-review\` | PR comment (github/ado/jira) or \`docs/workflow/<slug>/review-comment.md\` (local-markdown); idempotent re-runs |
 
 ## State separation
 
diff --git a/apps/claude-code/unic-archon-dlc/lib/dogfood-banner.mjs b/apps/claude-code/unic-archon-dlc/lib/dogfood-banner.mjs
new file mode 100644
index 00000000..5cb93a84
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/lib/dogfood-banner.mjs
@@ -0,0 +1,23 @@
+// @ts-check
+
+export const AGENT_DOC_BANNER = `<!--
+  AUTO-GENERATED by unic-archon-dlc — DO NOT EDIT BY HAND.
+  Source: apps/claude-code/unic-archon-dlc/lib/agent-docs-writer.mjs
+  Regenerate: re-run \`/unic-archon-dlc:setup\` (or call \`runInstall()\` from \`lib/install-runner.mjs\`).
+  This file describes the unic-archon-dlc target workflow; the current driver
+  in any given Consumer repo may differ.
+-->`
+
+export const SKILLS_BLOCK_BANNER =
+	'<!-- AUTO-GENERATED by unic-archon-dlc — do not edit; re-run `/unic-archon-dlc:setup` to regenerate. -->'
+
+/**
+ * Prepend a banner to a body with a blank line separator.
+ * @param {string} banner
+ * @param {string} body
+ * @returns {string}
+ */
+export function prependBanner(banner, body) {
+	if (!body || !body.trim()) return banner
+	return `${banner}\n\n${body.trimStart()}`
+}
diff --git a/apps/claude-code/unic-archon-dlc/lib/install-runner.mjs b/apps/claude-code/unic-archon-dlc/lib/install-runner.mjs
index c32a8c2d..c5d21df0 100644
--- a/apps/claude-code/unic-archon-dlc/lib/install-runner.mjs
+++ b/apps/claude-code/unic-archon-dlc/lib/install-runner.mjs
@@ -1,5 +1,5 @@
 // @ts-check
-import { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs'
+import { copyFileSync, existsSync, mkdirSync, readdirSync, readFileSync, writeFileSync } from 'node:fs'
 import { join } from 'node:path'
 import { updateAgentSkillsBlock, writeAgentDocs } from './agent-docs-writer.mjs'
 import { getDefaultLabels } from './labels-config.mjs'
@@ -12,8 +12,8 @@ import { getDefaultLabels } from './labels-config.mjs'
  */
 
 /**
- * @typedef {{ ok: true, configPath: string }} RunInstallOk
- * @typedef {{ ok: false, stage: 'validate' | 'config' | 'docs' | 'claude-md', message: string }} RunInstallFail
+ * @typedef {{ ok: true, configPath: string, workflowsCopied: number, commandsCopied: number }} RunInstallOk
+ * @typedef {{ ok: false, stage: 'validate' | 'config' | 'workflows' | 'commands' | 'docs' | 'claude-md', message: string }} RunInstallFail
  * @typedef {RunInstallOk | RunInstallFail} RunInstallResult
  */
 
@@ -51,15 +51,17 @@ export function detectRepoLayout(projectDir) {
 }
 
 /**
- * Writes .archon/unic-dlc.config.json, docs/agents/, and the CLAUDE.md Agent skills block.
+ * Writes .archon/unic-dlc.config.json, copies bundled workflows, writes docs/agents/,
+ * and updates the CLAUDE.md Agent skills block.
  * Merge precedence: defaults < existing < partialAnswers.
  * Does not call checkArchon() or prompt — the slash command owns both.
  *
  * @param {string} projectDir
  * @param {Partial<DlcConfig>} [partialAnswers]
+ * @param {string | null} [pluginRoot] - plugin installation directory; defaults to CLAUDE_PLUGIN_ROOT env var
  * @returns {RunInstallResult}
  */
-export function runInstall(projectDir, partialAnswers = {}) {
+export function runInstall(projectDir, partialAnswers = {}, pluginRoot = process.env['CLAUDE_PLUGIN_ROOT'] ?? null) {
 	const configPath = join(projectDir, '.archon', 'unic-dlc.config.json')
 
 	let existing = /** @type {Record<string, unknown>} */ ({})
@@ -106,6 +108,50 @@ export function runInstall(projectDir, partialAnswers = {}) {
 		return { ok: false, stage: 'config', message: `Failed to write config: ${/** @type {Error} */ (err).message}` }
 	}
 
+	let workflowsCopied = 0
+	let commandsCopied = 0
+	if (pluginRoot) {
+		const archonAssets =
+			/** @type {Array<{ subdir: string, ext: string, stageName: 'workflows' | 'commands', counter: (n: number) => void }>} */ ([
+				{
+					subdir: 'workflows',
+					ext: '.yaml',
+					stageName: 'workflows',
+					counter: (n) => {
+						workflowsCopied = n
+					},
+				},
+				{
+					subdir: 'commands',
+					ext: '.md',
+					stageName: 'commands',
+					counter: (n) => {
+						commandsCopied = n
+					},
+				},
+			])
+		for (const { subdir, ext, stageName, counter } of archonAssets) {
+			const srcDir = join(pluginRoot, '.archon', subdir)
+			const destDir = join(projectDir, '.archon', subdir)
+			try {
+				if (existsSync(srcDir)) {
+					mkdirSync(destDir, { recursive: true })
+					const files = readdirSync(srcDir).filter((f) => f.endsWith(ext))
+					for (const file of files) {
+						copyFileSync(join(srcDir, file), join(destDir, file))
+					}
+					counter(files.length)
+				}
+			} catch (err) {
+				return {
+					ok: false,
+					stage: stageName,
+					message: `Config written to ${configPath}. Failed to copy ${subdir}: ${/** @type {Error} */ (err).message}`,
+				}
+			}
+		}
+	}
+
 	try {
 		writeAgentDocs(projectDir, {
 			tracker: /** @type {TrackerBackend} */ (merged.tracker),
@@ -132,5 +178,5 @@ export function runInstall(projectDir, partialAnswers = {}) {
 		}
 	}
 
-	return { ok: true, configPath }
+	return { ok: true, configPath, workflowsCopied, commandsCopied }
 }
diff --git a/apps/claude-code/unic-archon-dlc/package.json b/apps/claude-code/unic-archon-dlc/package.json
index 63222abd..be1bd020 100644
--- a/apps/claude-code/unic-archon-dlc/package.json
+++ b/apps/claude-code/unic-archon-dlc/package.json
@@ -1,6 +1,6 @@
 {
   "name": "unic-archon-dlc",
-  "version": "0.1.0",
+  "version": "0.1.2",
   "private": true,
   "license": "LGPL-3.0-or-later",
   "type": "module",
@@ -10,7 +10,7 @@
     "pnpm": ">=10"
   },
   "scripts": {
-    "test": "node --test test/config-loader.test.mjs test/setup-explorer.test.mjs test/labels-config.test.mjs test/install-agent-docs.test.mjs test/install-claude-md.test.mjs test/tracker-adapter.test.mjs test/handoff-generator.test.mjs test/findings-writer.test.mjs test/prd-writer.test.mjs test/spike-verdicts.test.mjs test/issues-schema.test.mjs test/dag-builder.test.mjs test/slopcheck.test.mjs test/stub-detector.test.mjs test/install-helpers.test.mjs test/archon-check.test.mjs test/install-runner.test.mjs",
+    "test": "node --test test/config-loader.test.mjs test/setup-explorer.test.mjs test/labels-config.test.mjs test/install-agent-docs.test.mjs test/install-claude-md.test.mjs test/tracker-adapter.test.mjs test/handoff-generator.test.mjs test/findings-writer.test.mjs test/prd-writer.test.mjs test/spike-verdicts.test.mjs test/issues-schema.test.mjs test/dag-builder.test.mjs test/slopcheck.test.mjs test/stub-detector.test.mjs test/install-helpers.test.mjs test/archon-check.test.mjs test/install-runner.test.mjs test/dogfood-banner.test.mjs",
     "typecheck": "tsc --noEmit --project tsconfig.json",
     "bump": "unic-bump",
     "sync-version": "unic-sync-version",
diff --git a/apps/claude-code/unic-archon-dlc/test/dogfood-banner.test.mjs b/apps/claude-code/unic-archon-dlc/test/dogfood-banner.test.mjs
new file mode 100644
index 00000000..c9e13314
--- /dev/null
+++ b/apps/claude-code/unic-archon-dlc/test/dogfood-banner.test.mjs
@@ -0,0 +1,58 @@
+// @ts-check
+
+import assert from 'node:assert/strict'
+import { test } from 'node:test'
+import { AGENT_DOC_BANNER, prependBanner, SKILLS_BLOCK_BANNER } from '../lib/dogfood-banner.mjs'
+
+test('AGENT_DOC_BANNER contains AUTO-GENERATED signal', () => {
+	assert.ok(AGENT_DOC_BANNER.includes('AUTO-GENERATED'), 'AGENT_DOC_BANNER must say AUTO-GENERATED')
+})
+
+test('AGENT_DOC_BANNER references the source module', () => {
+	assert.ok(AGENT_DOC_BANNER.includes('agent-docs-writer.mjs'), 'AGENT_DOC_BANNER must name the source module')
+})
+
+test('AGENT_DOC_BANNER contains the regenerate hint', () => {
+	assert.ok(
+		AGENT_DOC_BANNER.includes('/unic-archon-dlc:setup'),
+		'AGENT_DOC_BANNER must mention how to regenerate via /unic-archon-dlc:setup'
+	)
+})
+
+test('SKILLS_BLOCK_BANNER contains AUTO-GENERATED signal', () => {
+	assert.ok(SKILLS_BLOCK_BANNER.includes('AUTO-GENERATED'), 'SKILLS_BLOCK_BANNER must say AUTO-GENERATED')
+})
+
+test('SKILLS_BLOCK_BANNER contains the regenerate hint', () => {
+	assert.ok(SKILLS_BLOCK_BANNER.includes('setup'), 'SKILLS_BLOCK_BANNER must mention regeneration via setup')
+})
+
+test('prependBanner joins banner + blank line + body', () => {
+	const result = prependBanner('<!-- banner -->', '# Title\n\nBody.')
+	assert.ok(result.startsWith('<!-- banner -->'), 'should start with banner')
+	assert.ok(result.includes('\n\n'), 'should have blank line separator')
+	assert.ok(result.includes('# Title'), 'should include body')
+})
+
+test('prependBanner works when body already ends with newline', () => {
+	const result = prependBanner('<!-- banner -->', '# Title\n')
+	assert.ok(result.startsWith('<!-- banner -->'), 'should start with banner')
+	assert.ok(result.includes('# Title'), 'should include body')
+	// Exactly one blank-line between banner and body
+	assert.match(result, /<!-- banner -->\n\n# Title/)
+})
+
+test('prependBanner with empty body returns just the banner', () => {
+	const result = prependBanner('<!-- banner -->', '')
+	assert.equal(result.trim(), '<!-- banner -->', 'empty body should return just the banner')
+})
+
+test('prependBanner strips leading whitespace from body for consistent spacing', () => {
+	const result = prependBanner('<!-- banner -->', '\n\n# Title\n')
+	// Exactly one blank line between banner and first line of body
+	assert.match(
+		result,
+		/^<!-- banner -->\n\n# Title/,
+		'banner + single blank line + body, regardless of leading whitespace in body'
+	)
+})
diff --git a/apps/claude-code/unic-archon-dlc/test/install-agent-docs.test.mjs b/apps/claude-code/unic-archon-dlc/test/install-agent-docs.test.mjs
index 01800a36..ff78fd33 100644
--- a/apps/claude-code/unic-archon-dlc/test/install-agent-docs.test.mjs
+++ b/apps/claude-code/unic-archon-dlc/test/install-agent-docs.test.mjs
@@ -6,19 +6,22 @@ import { tmpdir } from 'node:os'
 import { join } from 'node:path'
 import { test } from 'node:test'
 import { writeAgentDocs } from '../lib/agent-docs-writer.mjs'
+import { AGENT_DOC_BANNER } from '../lib/dogfood-banner.mjs'
 import { getDefaultLabels } from '../lib/labels-config.mjs'
 
+const TEST_CONFIG = {
+	tracker: /** @type {const} */ ('local-markdown'),
+	pr_strategy: /** @type {const} */ ('merge'),
+	branching: /** @type {const} */ ('gitflow'),
+	repo_layout: 'single-context',
+	labels: getDefaultLabels('local-markdown'),
+}
+
 test('writeAgentDocs writes all 5 docs/agents/*.md files with expected content', () => {
 	const dir = join(tmpdir(), `unic-dlc-docs-${Date.now()}`)
 	mkdirSync(dir, { recursive: true })
 
-	writeAgentDocs(dir, {
-		tracker: 'local-markdown',
-		pr_strategy: 'merge',
-		branching: 'gitflow',
-		repo_layout: 'single-context',
-		labels: getDefaultLabels('local-markdown'),
-	})
+	writeAgentDocs(dir, TEST_CONFIG)
 
 	// All 5 files exist
 	for (const name of ['issue-tracker.md', 'labels.md', 'branching.md', 'domain.md', 'workflow.md']) {
@@ -38,10 +41,67 @@ test('writeAgentDocs writes all 5 docs/agents/*.md files with expected content',
 	// domain.md reflects single-context
 	const domain = readFileSync(join(dir, 'docs', 'agents', 'domain.md'), 'utf8')
 	assert.ok(domain.includes('single-context') || domain.includes('single'), 'domain.md should mention single-context')
+	assert.ok(
+		!domain.includes('each context may also keep its own'),
+		'single-context domain.md should not mention per-context ADR pattern'
+	)
+	assert.ok(
+		!domain.includes('located via `CONTEXT-MAP.md`'),
+		'single-context domain.md should not contain the multi-context "How agents use this" wording'
+	)
 
-	// workflow.md mentions all 6 workflow phases
+	// workflow.md mentions all 7 workflow phases
 	const workflow = readFileSync(join(dir, 'docs', 'agents', 'workflow.md'), 'utf8')
-	for (const phase of ['explore', 'plan', 'build', 'qa', 'cleanup', 'triage']) {
+	for (const phase of ['explore', 'plan', 'build', 'qa', 'cleanup', 'triage', 'review']) {
 		assert.ok(workflow.toLowerCase().includes(phase), `workflow.md should mention ${phase}`)
 	}
+	// 'review' is also a substring of 'arch-review' and appears in surrounding prose — anchor on the
+	// unique command string to make sure the review *row* is present, not just the word.
+	assert.ok(workflow.includes('/unic-dlc-review'), 'workflow.md should list the /unic-dlc-review command')
+})
+
+test('multi-context domain.md mentions per-context ADR pattern', () => {
+	const dir = join(tmpdir(), `unic-dlc-docs-multi-${Date.now()}`)
+	mkdirSync(dir, { recursive: true })
+
+	writeAgentDocs(dir, { ...TEST_CONFIG, repo_layout: 'multi-context' })
+
+	const domain = readFileSync(join(dir, 'docs', 'agents', 'domain.md'), 'utf8')
+	assert.ok(domain.includes('multi-context'), 'multi-context domain.md should mention multi-context')
+	assert.ok(
+		domain.includes('each context may also keep its own'),
+		'multi-context domain.md should mention per-context docs/adr/ pattern'
+	)
+	assert.ok(
+		domain.includes('located via `CONTEXT-MAP.md`'),
+		'multi-context domain.md "How agents use this" should branch on multi-context wording'
+	)
+	assert.ok(domain.startsWith(AGENT_DOC_BANNER), 'multi-context domain.md should still begin with AGENT_DOC_BANNER')
+})
+
+test('each generated docs/agents/*.md file begins with AGENT_DOC_BANNER', () => {
+	const dir = join(tmpdir(), `unic-dlc-docs-banner-${Date.now()}`)
+	mkdirSync(dir, { recursive: true })
+
+	writeAgentDocs(dir, TEST_CONFIG)
+
+	for (const name of ['issue-tracker.md', 'labels.md', 'branching.md', 'domain.md', 'workflow.md']) {
+		const content = readFileSync(join(dir, 'docs', 'agents', name), 'utf8')
+		assert.ok(content.startsWith(AGENT_DOC_BANNER), `${name} should begin with AGENT_DOC_BANNER`)
+	}
+})
+
+test('AGENT_DOC_BANNER appears exactly once per file on repeated writeAgentDocs calls', () => {
+	const dir = join(tmpdir(), `unic-dlc-docs-dedup-${Date.now()}`)
+	mkdirSync(dir, { recursive: true })
+
+	writeAgentDocs(dir, TEST_CONFIG)
+	writeAgentDocs(dir, TEST_CONFIG)
+	writeAgentDocs(dir, TEST_CONFIG)
+
+	for (const name of ['issue-tracker.md', 'labels.md', 'branching.md', 'domain.md', 'workflow.md']) {
+		const content = readFileSync(join(dir, 'docs', 'agents', name), 'utf8')
+		const occurrences = content.split(AGENT_DOC_BANNER).length - 1
+		assert.equal(occurrences, 1, `${name}: AGENT_DOC_BANNER should appear exactly once`)
+	}
 })
diff --git a/apps/claude-code/unic-archon-dlc/test/install-claude-md.test.mjs b/apps/claude-code/unic-archon-dlc/test/install-claude-md.test.mjs
index 47e96c02..05ad4ee0 100644
--- a/apps/claude-code/unic-archon-dlc/test/install-claude-md.test.mjs
+++ b/apps/claude-code/unic-archon-dlc/test/install-claude-md.test.mjs
@@ -6,6 +6,7 @@ import { tmpdir } from 'node:os'
 import { join } from 'node:path'
 import { test } from 'node:test'
 import { updateAgentSkillsBlock } from '../lib/agent-docs-writer.mjs'
+import { SKILLS_BLOCK_BANNER } from '../lib/dogfood-banner.mjs'
 
 let _seq = 0
 function tempDir() {
@@ -42,3 +43,75 @@ test('## Agent skills block is refreshed idempotently — never duplicated', ()
 	// Original content preserved
 	assert.ok(content.includes('Content.'), 'original content must not be destroyed')
 })
+
+test('SKILLS_BLOCK_BANNER appears inside the marker block after updateAgentSkillsBlock', () => {
+	const dir = tempDir()
+	writeFileSync(join(dir, 'CLAUDE.md'), '# Project\n\nSome content.\n')
+
+	updateAgentSkillsBlock(dir)
+
+	const content = readFileSync(join(dir, 'CLAUDE.md'), 'utf8')
+	const begin = content.indexOf('<!-- unic-archon-dlc:begin -->')
+	const end = content.indexOf('<!-- unic-archon-dlc:end -->')
+	assert.ok(begin !== -1 && end !== -1, 'markers must be present')
+	const block = content.slice(begin, end)
+	assert.ok(block.includes(SKILLS_BLOCK_BANNER), 'SKILLS_BLOCK_BANNER must appear inside the marker block')
+	// Surrounding content preserved
+	assert.ok(content.includes('Some content.'), 'original content must not be destroyed')
+})
+
+test('SKILLS_BLOCK_BANNER appears exactly once inside the block after multiple updateAgentSkillsBlock runs', () => {
+	const dir = tempDir()
+	writeFileSync(join(dir, 'CLAUDE.md'), '# Project\n\nContent.\n')
+
+	updateAgentSkillsBlock(dir)
+	updateAgentSkillsBlock(dir)
+	updateAgentSkillsBlock(dir)
+
+	const content = readFileSync(join(dir, 'CLAUDE.md'), 'utf8')
+	const begin = content.indexOf('<!-- unic-archon-dlc:begin -->')
+	const end = content.indexOf('<!-- unic-archon-dlc:end -->')
+	const block = content.slice(begin, end)
+	const bannerCount = block.split(SKILLS_BLOCK_BANNER).length - 1
+	assert.equal(bannerCount, 1, 'SKILLS_BLOCK_BANNER must appear exactly once inside the block')
+})
+
+test('updateAgentSkillsBlock replaces stale content inside markers (does not append)', () => {
+	const dir = tempDir()
+	const seeded = [
+		'# Project',
+		'',
+		'Pre-content.',
+		'',
+		'## Agent skills',
+		'',
+		'<!-- unic-archon-dlc:begin -->',
+		'OLD STUFF THAT MUST BE REPLACED',
+		'ANOTHER LINE',
+		'<!-- unic-archon-dlc:end -->',
+		'',
+		'Post-content.',
+		'',
+	].join('\n')
+	writeFileSync(join(dir, 'CLAUDE.md'), seeded)
+
+	updateAgentSkillsBlock(dir)
+
+	const content = readFileSync(join(dir, 'CLAUDE.md'), 'utf8')
+	assert.ok(!content.includes('OLD STUFF THAT MUST BE REPLACED'), 'stale content must be removed')
+	assert.ok(!content.includes('ANOTHER LINE'), 'stale content must be removed')
+
+	const begin = content.indexOf('<!-- unic-archon-dlc:begin -->')
+	const end = content.indexOf('<!-- unic-archon-dlc:end -->')
+	assert.ok(begin !== -1 && end !== -1, 'markers must still be present')
+	const block = content.slice(begin, end)
+	const bannerCount = block.split(SKILLS_BLOCK_BANNER).length - 1
+	assert.equal(bannerCount, 1, 'SKILLS_BLOCK_BANNER must appear exactly once between markers')
+
+	for (const link of ['issue-tracker.md', 'labels.md', 'branching.md', 'domain.md', 'workflow.md']) {
+		assert.ok(content.includes(link), `${link} link must be present after refresh`)
+	}
+
+	assert.ok(content.includes('Pre-content.'), 'surrounding pre-content must be preserved')
+	assert.ok(content.includes('Post-content.'), 'surrounding post-content must be preserved')
+})
diff --git a/apps/claude-code/unic-archon-dlc/test/install-runner.test.mjs b/apps/claude-code/unic-archon-dlc/test/install-runner.test.mjs
index 7584ba72..794a56f5 100644
--- a/apps/claude-code/unic-archon-dlc/test/install-runner.test.mjs
+++ b/apps/claude-code/unic-archon-dlc/test/install-runner.test.mjs
@@ -162,6 +162,81 @@ test('updateAgentSkillsBlock failure: stage is claude-md and message includes co
 	)
 })
 
+test('asset copy: workflows and commands are copied from pluginRoot', () => {
+	const dir = join(tmpdir(), `unic-dlc-runner-assets-${Date.now()}`)
+	const pluginRoot = join(tmpdir(), `unic-dlc-plugin-${Date.now()}`)
+	mkdirSync(join(pluginRoot, '.archon', 'workflows'), { recursive: true })
+	mkdirSync(join(pluginRoot, '.archon', 'commands'), { recursive: true })
+	writeFileSync(join(pluginRoot, '.archon', 'workflows', 'build.yaml'), 'name: unic-dlc-build\n')
+	writeFileSync(join(pluginRoot, '.archon', 'workflows', 'plan.yaml'), 'name: unic-dlc-plan\n')
+	writeFileSync(join(pluginRoot, '.archon', 'workflows', 'README.md'), '# ignored')
+	writeFileSync(join(pluginRoot, '.archon', 'commands', 'unic-dlc-build.md'), '# build command\n')
+	writeFileSync(join(pluginRoot, '.archon', 'commands', 'unic-dlc-plan.md'), '# plan command\n')
+	writeFileSync(join(pluginRoot, '.archon', 'commands', '.gitkeep'), '')
+	mkdirSync(dir, { recursive: true })
+
+	const result = runInstall(dir, { tracker: 'github', pr_strategy: 'squash', branching: 'gitflow' }, pluginRoot)
+
+	assert.ok(result.ok)
+	assert.equal(result.workflowsCopied, 2)
+	assert.equal(result.commandsCopied, 2)
+	assert.ok(existsSync(join(dir, '.archon', 'workflows', 'build.yaml')))
+	assert.ok(existsSync(join(dir, '.archon', 'workflows', 'plan.yaml')))
+	assert.ok(!existsSync(join(dir, '.archon', 'workflows', 'README.md')))
+	assert.ok(existsSync(join(dir, '.archon', 'commands', 'unic-dlc-build.md')))
+	assert.ok(existsSync(join(dir, '.archon', 'commands', 'unic-dlc-plan.md')))
+	assert.ok(!existsSync(join(dir, '.archon', 'commands', '.gitkeep')))
+})
+
+test('asset copy: skipped gracefully when pluginRoot is null', () => {
+	const dir = join(tmpdir(), `unic-dlc-runner-noassets-${Date.now()}`)
+	mkdirSync(dir, { recursive: true })
+
+	const result = runInstall(dir, { tracker: 'github', pr_strategy: 'squash', branching: 'gitflow' }, null)
+
+	assert.ok(result.ok)
+	assert.equal(result.workflowsCopied, 0)
+	assert.equal(result.commandsCopied, 0)
+	assert.ok(!existsSync(join(dir, '.archon', 'workflows')))
+	assert.ok(!existsSync(join(dir, '.archon', 'commands')))
+})
+
+test('asset copy: skipped gracefully when pluginRoot has no .archon subdirs', () => {
+	const dir = join(tmpdir(), `unic-dlc-runner-noarchon-${Date.now()}`)
+	const pluginRoot = join(tmpdir(), `unic-dlc-plugin-empty-${Date.now()}`)
+	mkdirSync(pluginRoot, { recursive: true })
+	mkdirSync(dir, { recursive: true })
+
+	const result = runInstall(dir, { tracker: 'github', pr_strategy: 'squash', branching: 'gitflow' }, pluginRoot)
+
+	assert.ok(result.ok)
+	assert.equal(result.workflowsCopied, 0)
+	assert.equal(result.commandsCopied, 0)
+})
+
+test('asset copy: overwrites existing files on re-run', () => {
+	const dir = join(tmpdir(), `unic-dlc-runner-overwrite-${Date.now()}`)
+	const pluginRoot = join(tmpdir(), `unic-dlc-plugin-ow-${Date.now()}`)
+	mkdirSync(join(pluginRoot, '.archon', 'workflows'), { recursive: true })
+	mkdirSync(join(pluginRoot, '.archon', 'commands'), { recursive: true })
+	writeFileSync(join(pluginRoot, '.archon', 'workflows', 'build.yaml'), 'name: unic-dlc-build\nruntime: bun\n')
+	writeFileSync(join(pluginRoot, '.archon', 'commands', 'unic-dlc-build.md'), '# updated command\n')
+	mkdirSync(join(dir, '.archon', 'workflows'), { recursive: true })
+	mkdirSync(join(dir, '.archon', 'commands'), { recursive: true })
+	writeFileSync(join(dir, '.archon', 'workflows', 'build.yaml'), 'name: unic-dlc-build\n')
+	writeFileSync(join(dir, '.archon', 'commands', 'unic-dlc-build.md'), '# old command\n')
+
+	const result = runInstall(dir, { tracker: 'github', pr_strategy: 'squash', branching: 'gitflow' }, pluginRoot)
+
+	assert.ok(result.ok)
+	assert.equal(result.workflowsCopied, 1)
+	assert.equal(result.commandsCopied, 1)
+	const workflow = readFileSync(join(dir, '.archon', 'workflows', 'build.yaml'), 'utf8')
+	assert.ok(workflow.includes('runtime: bun'), 'stale workflow should be overwritten')
+	const command = readFileSync(join(dir, '.archon', 'commands', 'unic-dlc-build.md'), 'utf8')
+	assert.ok(command.includes('# updated command'), 'stale command should be overwritten')
+})
+
 test('custom labels in existing config are preserved on reconfigure', () => {
 	const dir = join(tmpdir(), `unic-dlc-runner-labels-${Date.now()}`)
 	mkdirSync(join(dir, '.archon'), { recursive: true })
diff --git a/apps/claude-code/unic-confluence/AGENTS.md b/apps/claude-code/unic-confluence/AGENTS.md
new file mode 100644
index 00000000..57b490ff
--- /dev/null
+++ b/apps/claude-code/unic-confluence/AGENTS.md
@@ -0,0 +1,87 @@
+# AGENTS.md — unic-confluence
+
+Guidance for any AI agent working inside this Plugin directory. `CLAUDE.md` in this directory is a symlink to this file.
+
+## What this Plugin is
+
+`unic-confluence` is a Claude Code Plugin in the [`unic-agents-plugins`](../../../AGENTS.md) monorepo. It Publishes Markdown files into Confluence pages by injecting rendered HTML into named Injection Zones, via the Confluence v2 REST API. The Plugin ships both as a Claude Code slash command (`/unic-confluence`) and as a plain Node.js script that Consumers can wire into their own pipelines. See [`CONTEXT.md`](CONTEXT.md) for the domain vocabulary (Publish, Injection Zone, Zone Label, Page Map, Page Alias, Auto-aliasing).
+
+## Where to start
+
+Root docs (monorepo-wide conventions live here — pnpm scripts, Gitflow, SemVer, Conventional Commits, code conventions, LICENSE policy, cross-platform requirement):
+
+- [Root AGENTS.md](../../../AGENTS.md) — source of truth for cross-cutting rules
+- [Root CONTEXT.md](../../../CONTEXT.md) — monorepo-wide vocabulary (Plugin, Workspace Package, Release, Feature, Consumer)
+- [Root CONTEXT-MAP.md](../../../CONTEXT-MAP.md) — index of all bounded contexts in the repo
+- [Root docs/adr/](../../../docs/adr/) — monorepo-wide architecture decisions
+- [Root docs/process/](../../../docs/process/) — process and workflow guides
+
+This Plugin's own decisions:
+
+- [Plugin docs/adr/](docs/adr/) — Plugin-specific architecture decisions
+
+## Commands
+
+Plugin-specific pnpm scripts (run from this directory or with `pnpm --filter unic-confluence <script>` from the repo root):
+
+```sh
+pnpm test                 # run node:test suite over the pure-function library
+pnpm typecheck            # tsc --noEmit over the Plugin's .mjs sources
+pnpm confluence           # run the Publish CLI (node scripts/push-to-confluence.mjs)
+pnpm bump <patch|minor|major>   # bump plugin.json version + promote CHANGELOG
+pnpm sync-version         # mirror plugin.json version into marketplace.json + package.json
+pnpm tag                  # create the unic-confluence@<version> git tag locally
+pnpm verify:changelog     # check CHANGELOG entry for the current version
+```
+
+Monorepo-wide commands (`pnpm install`, `pnpm check`, `pnpm format`, `pnpm ci:check`) are documented in the [root AGENTS.md](../../../AGENTS.md).
+
+## Layout
+
+```tree
+.claude-plugin/           # Plugin manifest (plugin.json) and marketplace listing
+commands/                 # Claude Code slash command definition (unic-confluence.md)
+scripts/                  # push-to-confluence.mjs (the CLI) and pure-function library
+docs/                     # Plugin-specific documentation
+  adr/                    # Plugin Architecture Decision Records
+```
+
+## Plugin doctrines
+
+Load-bearing invariants. These either originate in a Plugin ADR or are policy decisions that are not obvious from the code.
+
+- **Refuse to Publish without Injection Zone markers** — unless the legacy anchor-macro path or explicit append fallback applies. The Plugin never silently overwrites a page body. See [ADR-0001](docs/adr/0001-refuse-publish-without-markers.md).
+- **Three-strategy injection priority.** Plain-text Injection Zone markers → legacy anchor macros → append fallback, in that order. The first matching strategy wins. See [ADR-0002](docs/adr/0002-three-strategy-injection-priority.md).
+- **Structured macro for code blocks.** Markdown fenced code blocks render to Confluence `<ac:structured-macro ac:name="code">`, not raw `<pre>`. See [ADR-0003](docs/adr/0003-structured-macro-for-code-blocks.md).
+- **Dry-run is read-only.** `--dry-run` never issues a `PUT` and never mutates credentials, the Page Map, or any remote state. See [ADR-0004](docs/adr/0004-dry-run-read-only.md).
+- **Ping-check auth, not per-page verify.** Authentication is checked once via a lightweight ping; the Plugin does not pre-flight every page id. See [ADR-0005](docs/adr/0005-ping-check-auth-over-per-page-verify.md).
+- **Hard HTTP timeout.** Every Confluence request runs under a hard timeout so a stalled API call cannot hang the CLI. See [ADR-0006](docs/adr/0006-hard-http-timeout.md).
+- **`CliError` class for user-facing failures.** Recoverable, user-actionable errors throw `CliError` with an exit code; unexpected errors keep their stack. See [ADR-0007](docs/adr/0007-clierror-class.md).
+- **Pure-functions lib with tests.** Markdown→HTML, marker detection, and Page Map mutations are pure functions covered by `node:test`. See [ADR-0008](docs/adr/0008-pure-functions-lib-with-tests.md).
+- **Bare-integer Page id schema.** Page Map values are bare integer Confluence page ids, not objects. Any schema change is breaking. See [ADR-0009](docs/adr/0009-bare-integer-page-id-schema.md).
+- **No pnpm catalog for runtime deps.** This Plugin pins its own runtime deps directly rather than going through the workspace catalog. See [ADR-0010](docs/adr/0010-no-catalog-for-runtime-deps.md).
+- **Auto-aliasing on raw-id Publish.** A Publish addressed by raw numeric id writes a new Page Alias into the Page Map so the next Publish can use a slug. See [ADR-0011](docs/adr/0011-alias-auto-population.md).
+- **Do-not-add scope guard.** The list below is canonical and load-bearing — these features have been considered and rejected with reasons. See [ADR-0012](docs/adr/0012-do-not-add-scope-guard.md).
+
+## External dependencies
+
+- **Confluence v2 REST API** at the Consumer's Confluence instance (default `https://uniccom.atlassian.net`).
+- **Credentials file** at `~/.unic-confluence.json` (chmod 600), with fields `url`, `username`, `token`. Overridable per-run via `CONFLUENCE_URL`, `CONFLUENCE_USER`, `CONFLUENCE_TOKEN` env vars.
+- **`marked`** (runtime npm dep, pinned in this Plugin's `package.json` — see [ADR-0010](docs/adr/0010-no-catalog-for-runtime-deps.md)). Used to render Markdown to HTML before injection.
+
+## Do not add
+
+These have been considered and explicitly rejected. Open a Feature in the issue tracker with a concrete use case before reopening any of them; expect to overturn [ADR-0012](docs/adr/0012-do-not-add-scope-guard.md).
+
+- **Image upload / attachments.** Walking the Markdown AST for local image references and uploading via `/wiki/rest/api/content/{id}/child/attachment` is significant work that requires a new subcommand, content-negotiation path, and multi-part form handling. Defer.
+- **Create-page support.** The Plugin only updates existing pages. Adding `POST /wiki/api/v2/pages` with `spaceId` + `parentId` forces a schema change to the Page Map (value becomes an object, not a bare integer) and complicates every read path.
+- **Multi-space or cross-instance publishing.** Page ids are unique per Confluence instance; multiple instances would need a `baseUrl`-per-entry schema and credential routing.
+- **MCP server.** The slash command and the CLI are the correct and sufficient surfaces. An MCP server would add lifecycle, transport, and protocol-schema overhead for zero user-visible benefit.
+- **Agents or sub-agents.** Publish is a deterministic one-shot — read file → convert Markdown → GET page → inject → PUT page. There is no branching, no tool selection, no iteration; agent autonomy has no value here.
+- **Recursive directory Publishing.** Publishing all Markdown files under a tree requires mapping every file to a page id, handling partial failures, and defining rollback semantics. The complexity grows faster than the value.
+- **Changesets / release-please / semantic-release.** `@unic/release-tools` is sufficient for one Plugin with two version fields. Do not add a release-management framework.
+- **Watch mode / file-watcher.** Confluence is not a live preview target; each Publish increments page version and creates a Confluence revision. Accidental rapid Publishes would pollute revision history.
+
+## Plugin ADRs
+
+Plugin-specific architecture decisions live in [docs/adr/](docs/adr/).
diff --git a/apps/claude-code/unic-confluence/CLAUDE.md b/apps/claude-code/unic-confluence/CLAUDE.md
deleted file mode 100644
index 53f843e5..00000000
--- a/apps/claude-code/unic-confluence/CLAUDE.md
+++ /dev/null
@@ -1,89 +0,0 @@
-# CLAUDE.md
-
-This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
-
-## What this repo is
-
-A Claude Code plugin that publishes Markdown files to Confluence pages via the Confluence v2 REST API. It ships both as a Claude Code plugin (registered via `marketplace.json`) and as a plain npm package (consumed via `scripts/push-to-confluence.mjs`).
-
-No build step. No TypeScript. No test suite. Pure ESM Node.js.
-
-## Key files
-
-| Path                              | Purpose                                                    |
-| --------------------------------- | ---------------------------------------------------------- |
-| `scripts/push-to-confluence.mjs`  | Single-file CLI — all logic lives here                     |
-| `commands/unic-confluence.md`     | Claude Code slash command definition (`/unic-confluence`)  |
-| `.claude-plugin/plugin.json`      | Plugin manifest (name, version, command list)              |
-| `.claude-plugin/marketplace.json` | Marketplace listing (used by `claude plugins marketplace`) |
-
-## Running the script
-
-```sh
-# One-time credential setup (writes ~/.unic-confluence.json)
-node scripts/push-to-confluence.mjs --setup
-
-# Verify all page IDs in confluence-pages.json resolve
-node scripts/push-to-confluence.mjs --verify
-
-# Publish a file (key from confluence-pages.json, or raw numeric page ID)
-node scripts/push-to-confluence.mjs <page-key-or-id> <file.md>
-```
-
-No `pnpm run` wrapper is defined in this repo's `package.json` — consuming repos add a `"confluence"` script themselves. When testing locally, call the script with `node` directly.
-
-## Content injection strategies
-
-The script uses three strategies (in priority order) to place Markdown content on an existing Confluence page:
-
-1. **Plain-text markers** — `[AUTO_INSERT_START: label]` / `[AUTO_INSERT_END: label]` anywhere in the page body. Labels must match exactly.
-2. **Anchor macros (legacy)** — Confluence `<ac:structured-macro>` anchors named `md-start` / `md-end`.
-3. **Append** — no markers found; content is appended to the end of the existing page body.
-
-## Credentials
-
-Stored in `~/.unic-confluence.json` (chmod 600). Three fields: `url`, `username`, `token`. Can be overridden per-run with env vars `CONFLUENCE_URL`, `CONFLUENCE_USER`, `CONFLUENCE_TOKEN`.
-
-Default Confluence URL pre-filled in setup: `https://uniccom.atlassian.net`.
-
-## Plugin versioning
-
-`.claude-plugin/plugin.json` is the single source of truth. Every `feat(spec-NN)` / `fix(spec-NN)` commit includes a version bump and a dated `CHANGELOG.md` entry, via `pnpm bump <patch|minor|major>`. Never hand-edit `.claude-plugin/marketplace.json` — `pnpm bump` mirrors into it automatically.
-
-**SemVer policy:**
-
-- **major**: breaking change to CLI flags, exit codes, or on-disk contracts (`confluence-pages.json` schema, `~/.unic-confluence.json` format).
-- **minor**: new flag, subcommand, or user-visible feature.
-- **patch**: bug fix, refactor, docs, internal tooling.
-
-`pnpm verify:changelog` (and CI on PRs) rejects changes that modify source or user-facing docs without a version bump and CHANGELOG entry.
-
-## Naming convention
-
-| Surface           | Value                        |
-| ----------------- | ---------------------------- |
-| GitHub repo       | `unic-claude-code-<service>` |
-| Plugin identifier | `unic-<service>`             |
-| npm package name  | `unic-<service>`             |
-
-## Do not add
-
-The following are explicitly out of scope for this plugin. Do not implement them without first opening a GitHub issue and getting explicit sign-off from the maintainer:
-
-- **Image upload / attachments** — walking the Markdown AST to find local image references and uploading them via `/wiki/rest/api/content/{id}/child/attachment` is significant work that requires a new CLI subcommand, a new content-negotiation path, and multi-part form handling. Defer until a user explicitly requests it with a concrete use case.
-
-- **Create-page support** — the script only updates existing pages. Adding `POST /wiki/api/v2/pages` with `spaceId` + `parentId` requires a schema change to `confluence-pages.json` (value becomes an object, not just a page ID integer), complicating every code path that reads the file. Defer unless there is real demand.
-
-- **Multi-space or cross-instance publishing** — `confluence-pages.json` maps keys to page IDs, and page IDs are unique per Confluence instance. Supporting multiple Confluence instances would require a different config schema (e.g. a `baseUrl` field per entry) and credential routing logic. Out of scope.
-
-- **MCP server** — there is no benefit to wrapping this plugin's functionality in a Model Context Protocol server. The slash command is the correct and sufficient surface. An MCP server would add a process lifecycle, a transport layer, and a versioned protocol schema for zero user-visible benefit.
-
-- **Agents or sub-agents** — the publish task is a deterministic one-shot sequence: read file → convert Markdown → GET page → inject → PUT page. There is no branching, no tool-selection, and no iteration. Agent autonomy adds complexity without value here.
-
-- **Recursive directory publishing** — publishing all Markdown files under a directory tree in one command (e.g. `node push-to-confluence.mjs docs/`) requires mapping every file to a page ID, handling partial failures, and defining rollback semantics. The complexity grows faster than the value. Publish one file at a time.
-
-- **Changesets or release-please** — the `sync-version.mjs` script (spec 06) is sufficient for one package with two version fields. Do not add a release-management framework (Changesets, release-please, semantic-release) — the overhead exceeds the benefit for a single-file plugin.
-
-- **Watch mode / file-watcher** — a flag like `--watch` that re-publishes on file change is not appropriate for a Confluence publishing tool. Confluence is not a live preview target; each publish increments the page version and creates a revision in Confluence's history. Accidental rapid publishes would pollute the revision history.
-
-When in doubt: if the feature is not in the existing `scripts/push-to-confluence.mjs` command set and is not listed in an open `docs/plans/` spec, it is out of scope. Open a GitHub issue before starting implementation.
diff --git a/apps/claude-code/unic-confluence/CLAUDE.md b/apps/claude-code/unic-confluence/CLAUDE.md
new file mode 120000
index 00000000..47dc3e3d
--- /dev/null
+++ b/apps/claude-code/unic-confluence/CLAUDE.md
@@ -0,0 +1 @@
+AGENTS.md
\ No newline at end of file
diff --git a/apps/claude-code/unic-confluence/CONTRIBUTING.md b/apps/claude-code/unic-confluence/CONTRIBUTING.md
index b2fd9473..f7f2b0ef 100644
--- a/apps/claude-code/unic-confluence/CONTRIBUTING.md
+++ b/apps/claude-code/unic-confluence/CONTRIBUTING.md
@@ -1,141 +1,40 @@
 # Contributing
 
-This project uses a spec-driven development workflow. New features and fixes are described in self-contained spec files under `docs/plans/`. Implementation is either automated with **Ralph Orchestrator** (recommended) or done by hand following the same steps.
+Contributions start as GitHub Issues in the [unic/unic-agents-plugins](https://github.com/unic/unic-agents-plugins/issues) tracker. Open or claim an issue before starting implementation.
 
 ## Prerequisites
 
-| Tool            | Version                                         | How to get it                                                          |
-| --------------- | ----------------------------------------------- | ---------------------------------------------------------------------- |
-| Node.js         | ≥ 22 (see `.nvmrc` for the recommended version) | [nodejs.org](https://nodejs.org)                                       |
-| pnpm            | ≥ 10                                            | `npm install -g pnpm`                                                  |
-| Claude Code CLI | latest                                          | [claude.ai/code](https://claude.ai/code) — required as Ralph's backend |
+| Tool            | Version                                         | How to get it                            |
+| --------------- | ----------------------------------------------- | ---------------------------------------- |
+| Node.js         | ≥ 22 (see `.nvmrc` for the recommended version) | [nodejs.org](https://nodejs.org)         |
+| pnpm            | ≥ 10                                            | `npm install -g pnpm`                    |
+| Claude Code CLI | latest                                          | [claude.ai/code](https://claude.ai/code) |
 
-Everything else (Ralph Orchestrator, Biome, TypeScript) is a project devDependency and installs with:
+Everything else (Biome, TypeScript) is a project devDependency and installs with:
 
 ```sh
 pnpm install
 ```
 
-**ralph-loop** is a Claude Code plugin and must be installed once globally:
+## Planning
 
-```sh
-claude plugins install anthropics/claude-plugins-official/plugins/ralph-loop
-```
-
-This registers the `/ralph-loop` skill and the session loop hook in Claude Code.
-
-## Planning — Writing a spec
-
-All work starts with a spec file. This applies whether you plan to run Ralph or implement manually.
-
-### 1. Pick a number
-
-Check the execution order table in [`docs/plans/README.md`](docs/plans/README.md) for the current highest spec number. Create `docs/plans/NN-short-slug.md` with the next sequential number.
-
-### 2. Fill in the required headers
-
-```markdown
-# NN. Title
-
-**Priority:** P0 | P1 | P2
-**Effort:** XS | S | M | L
-**Version impact:** patch | minor | major
-**Depends on:** (spec numbers this must land after, or "none")
-**Touches:** (comma-separated list of files/dirs affected)
-```
-
-**Version impact** is used by Ralph (and human contributors) to know which `pnpm bump` argument to pass when committing. Follow the SemVer policy in [`docs/plans/README.md`](docs/plans/README.md).
-
-### 3. Fill in the required sections
-
-| Section                   | What to write                                                                                       |
-| ------------------------- | --------------------------------------------------------------------------------------------------- |
-| `## Context`              | Why this change is needed; what problem it solves                                                   |
-| `## Current behaviour`    | Exact code snippets or behaviour _before_ the change. Ralph uses this to verify the starting state. |
-| `## Target behaviour`     | What the code/behaviour should look like _after_                                                    |
-| `## Affected files`       | Table: file path → Create / Modify / Delete                                                         |
-| `## Implementation steps` | Numbered steps with exact "before → after" diffs or code to write. No ambiguity.                    |
-| `## Verification`         | Shell commands to run and their expected output                                                     |
-| `## Acceptance criteria`  | Checkbox list — Ralph checks each item before committing                                            |
-| `## Out of scope`         | Explicit list of things NOT to change in this spec                                                  |
-
-Good specs are **self-contained** — Ralph has no memory of previous runs or conversations. Include actual code snapshots; don't rely on "as discussed".
-
-### 4. Register the spec
-
-In `docs/plans/README.md`:
-
-- Add a row to the **Execution order** table.
-- Add any ordering constraints to the **Cross-cutting dependencies** section.
-
-## Implementing with Ralph Orchestrator
-
-Once specs are in place, Ralph implements them one at a time in a loop.
-
-```sh
-# From the repo root:
-pnpm ralph
-```
-
-Ralph reads `ralph.yml`, which points to `PROMPT.md` as its prompt. Each loop iteration:
-
-1. Claude Code scans `docs/plans/` in order and finds the first spec without `**Status: done**`.
-2. Reads the spec fully and verifies the "Current behaviour" snapshot matches.
-3. Implements the "Implementation steps" exactly.
-4. Adds a CHANGELOG entry, runs `pnpm bump <type>`, and runs `pnpm verify:changelog`.
-5. Marks the spec `**Status: done — YYYY-MM-DD**` and commits `feat(spec-NN): description (vX.Y.Z)`.
-6. Outputs a summary, then stops. Ralph feeds the prompt again for the next spec.
-
-### Stopping and resuming
-
-- **Stop early**: `Ctrl+C`. The current spec may be partially implemented — check `git status` before continuing.
-- **Resume**: `pnpm ralph` again. Ralph scans for the first unfinished spec and picks up there.
-- **Paused for review**: if a spec has a `## Questions` section, Claude emits `LOOP_COMPLETE` and Ralph stops. Read the question, update the spec with your answer, remove the `## Questions` section, and run `pnpm ralph` again.
-- **All done**: when all specs are `**Status: done**`, Claude emits `LOOP_COMPLETE` and Ralph exits.
-
-### After a Ralph run
-
-```sh
-git log --oneline   # review what was committed
-git push            # push when satisfied
-```
-
-Ralph never pushes — that is always a manual step.
-
-### Configuration
-
-The `pnpm ralph` script runs `ralph run -c ralph.yml -H builtin:code-assist`. Key settings (`ralph.yml`):
-
-| Setting / flag                   | Value           | Meaning                                                                        |
-| -------------------------------- | --------------- | ------------------------------------------------------------------------------ |
-| `-c ralph.yml`                   | config file     | Explicit path to the Ralph config (avoids ambiguity if multiple configs exist) |
-| `-H builtin:code-assist`         | hat             | Runs Claude Code in code-assist mode, optimised for implementation tasks       |
-| `cli.backend`                    | `claude`        | Uses the Claude Code CLI as the underlying agent                               |
-| `event_loop.prompt_file`         | `PROMPT.md`     | The orchestration prompt fed to Claude on every iteration                      |
-| `event_loop.completion_promise`  | `LOOP_COMPLETE` | Sentinel string that ends the loop                                             |
-| `event_loop.max_iterations`      | `100`           | Hard cap on loop iterations                                                    |
-| `event_loop.max_runtime_seconds` | `14400`         | 4-hour safety timeout                                                          |
-
-## Implementing manually (without Ralph)
+All work starts as a GitHub Issue in [unic/unic-agents-plugins](https://github.com/unic/unic-agents-plugins/issues). Open or claim an issue there before touching code. See [`docs/agents/workflow.md`](../../../../docs/agents/workflow.md) in the repo root for the full seven-phase workflow (triage → specs → implementation → PR → release).
 
-Follow the same steps Ralph would take:
+## Implementing
 
-1. Find the first unstarted spec (no `**Status: done**`) in the execution order table.
-2. Read the spec fully — check "Depends on" and "Current behaviour" before touching any code.
-3. Follow the "Implementation steps" exactly. If the "before" snapshot doesn't match what you see, note the discrepancy in a `## Deviations` section.
-4. Add one bullet under the appropriate `## [Unreleased]` subsection in `CHANGELOG.md`.
-5. Run `pnpm bump <patch|minor|major>` (matches the spec's `**Version impact:**` line).
-6. Run the spec's "Verification" commands. Fix all failures.
-7. Check every item in "Acceptance criteria".
-8. Mark the spec done — add `**Status: done — YYYY-MM-DD**` as the second line (after the title).
-9. Commit everything: `git add -A && git commit -m "feat(spec-NN): description (vX.Y.Z)"`.
+1. Branch from `develop` following the Gitflow naming convention (`feature/<slug>`).
+2. Implement the changes described in the issue.
+3. Add one bullet under the appropriate `## [Unreleased]` subsection in `CHANGELOG.md`.
+4. Run `pnpm bump <patch|minor|major>` to reflect the semantic impact.
+5. Run the issue's verification/acceptance criteria. Fix all failures.
+6. Commit with a conventional commit message: `feat(unic-confluence): description`.
 
 ## Finishing your work
 
 1. Verify everything is clean
 
-   - `pnpm format`: Ensure formatting is correct
-   - `pnpm check`: biome lint + format check
+   - `pnpm -w format`: format all files (run from repo root)
+   - `pnpm -w check`: Biome lint + format check (run from repo root)
    - `pnpm test`: run tests
    - `pnpm typecheck`: type check
 
@@ -147,11 +46,11 @@ Follow the same steps Ralph would take:
    - `git add -A`
    - `git commit -m "<Convention Commit>"`
 
-The `pnpm bump <patch|minor|major>` step is mandatory — `pnpm verify:changelog` (enforced in CI and the pre-push hook) will reject the commit if source files changed without a version bump and dated CHANGELOG entry.
+The `pnpm bump <patch|minor|major>` step is mandatory — `pnpm verify:changelog` (enforced in CI) will reject a PR if source files changed without a version bump and dated CHANGELOG entry.
 
 ## Code standards
 
-Ground rules are documented in [`docs/plans/README.md`](docs/plans/README.md) and [`CLAUDE.md`](CLAUDE.md). Key points:
+Ground rules are documented in [`CLAUDE.md`](CLAUDE.md). Key points:
 
 - **Tabs** for indentation, **LF** line endings (enforced by `.editorconfig`)
 - **pnpm** — not npm; all scripts use `pnpm run <name>`
@@ -166,7 +65,7 @@ PRs must pass:
 ```sh
 pnpm ci:check        # Biome lint + format
 pnpm test            # Node built-in test runner
-pnpm verify:changelog  # version bumped + CHANGELOG entry present (available after spec 20)
+pnpm verify:changelog  # version bumped + CHANGELOG entry present
 ```
 
 A PR that modifies source or user-facing docs without bumping the version will fail CI.
diff --git a/apps/claude-code/unic-confluence/PROMPT.md b/apps/claude-code/unic-confluence/PROMPT.md
deleted file mode 100644
index 565737fd..00000000
--- a/apps/claude-code/unic-confluence/PROMPT.md
+++ /dev/null
@@ -1,115 +0,0 @@
-# Ralph Orchestrator — unic-confluence Roadmap
-
-You are implementing the roadmap for the `unic-confluence` Claude Code plugin, one spec at a time.
-
-## Step 1 — Determine what's next
-
-Check the execution order in `docs/plans/README.md`. Then scan spec files in order (00 → 18) and find the first file that does NOT contain the string `**Status: done`.
-
-If **all** specs contain `**Status: done`, output exactly:
-
-```
-<promise>LOOP_COMPLETE</promise>
-```
-
-…and stop. Do nothing else.
-
-## Step 2 — Read the spec completely
-
-Read the entire spec file before writing any code. Pay special attention to:
-
-- **Current behaviour** — verify the code actually works this way before starting; if it doesn't, add a `## Deviations` section at the bottom of the spec file documenting the discrepancy
-- **Out of scope** — list of things you must NOT change in this iteration
-- **Depends on** — if a dependency spec is not yet marked done, skip this spec and move to the next one that can be done
-
-## Step 3 — Implement
-
-Check the spec's `**Version impact:**` line to choose the implementation approach:
-
-**`none` (workspace/infrastructure spec)** — implement directly. Follow the "Implementation steps" exactly. If a step's "before" snapshot doesn't match the current file, consult the "Deviations" section (if you wrote one) or document the discrepancy and adapt minimally.
-
-**`patch` / `minor` / `major` (plugin spec)** — use `/tdd` to drive implementation. Treat the spec's "Implementation steps" as guidance (key files to touch, rough order), not a recipe. The spec's **Acceptance criteria** are the target; the red-green-refactor cycle drives the path there.
-
-Ground rules (from `docs/plans/README.md`):
-
-- Use `pnpm` for all package operations
-- Tabs for indentation, LF line endings (per `.editorconfig`)
-- Conventional commits: `feat(scope): description`, `fix(scope): description`, `chore(scope): description`
-- Cross-platform: use Node.js APIs (`node:path`, `node:fs`, `node:os`) instead of shell commands; no bash/sh assumptions
-- **Never hand-edit** `.claude-plugin/marketplace.json` version — use `pnpm bump`
-- If something can't be followed as written: document it in `## Deviations`, don't silently deviate
-
-**Supporting skill:** `/diagnose` — if a failure persists and the root cause appears to be outside the spec's scope (an existing module, tooling, or dependency the spec doesn't touch), use `/diagnose` to isolate it before continuing.
-
-## Step 4 — Verify
-
-Run the exact commands in the spec's **Verification** section. Fix any failures before proceeding.
-
-Check every item in **Acceptance criteria**. If any item fails, fix it.
-
-Then run the always-on repo hygiene checks regardless of spec type:
-
-```sh
-pnpm -w check   # Biome + Prettier (workspace root) — fix any failures before proceeding
-```
-
-## Step 4.5 — Bump version + CHANGELOG
-
-1. Read the spec's `**Version impact:** patch|minor|major` line at the top of the spec file.
-   If the line is absent, infer: breaking CLI/contract change → `major`; new flag/feature → `minor`; bug fix, refactor, docs → `patch`.
-
-2. Append **one bullet** to the matching subsection under `## [Unreleased]` in `CHANGELOG.md`, replacing the `- (none)` placeholder on first use:
-
-   - `### Breaking` — CLI flag renamed/removed, exit-code change, on-disk file schema change
-   - `### Added` — new flag, subcommand, or user-visible feature
-   - `### Fixed` — bug fix, refactor, docs, internal tooling
-     Wording: one line, user-facing, present-tense description (e.g. `- \`pnpm bump\` command for atomic version bumping`).
-
-3. Run:
-
-   ```sh
-   pnpm bump <patch|minor|major>
-   ```
-
-   This atomically: increments `plugin.json` version, mirrors into `marketplace.json`, and promotes `[Unreleased]` → a new dated section.
-
-4. Run `pnpm verify:changelog` to confirm the check passes.
-
-## Step 5 — Mark done and commit
-
-1. Add this line immediately after the spec's `# NN. Title` heading (as the second line):
-
-```
-**Status: done — YYYY-MM-DD**
-```
-
-Replace `YYYY-MM-DD` with today's date.
-
-2. Stage and commit all changes:
-
-```sh
-git add -A
-git commit -m "feat(spec-NN): <short description of what was implemented> (vX.Y.Z)"
-```
-
-Replace `NN` with the spec number (e.g. `00`, `03`) and write a clear description.
-Replace `X.Y.Z` with the version output by `pnpm bump`.
-
-3. **Do not push.** Commits only.
-
-## Step 6 — Stop for this iteration
-
-Output a brief summary of what was implemented and committed. Then stop — Ralph will feed this prompt again for the next spec.
-
----
-
-## Important constraints
-
-- Implement **one spec per iteration**. Do not implement multiple specs in a single run.
-- If a spec requires human judgment (e.g. choosing between two design approaches not covered by the spec), stop, document the question in a `## Questions` section at the bottom of the spec file, and output:
-  ```
-  <promise>LOOP_COMPLETE</promise>
-  ```
-  This pauses the loop for human review.
-- Do not create any files outside of what the spec describes.
-- Do not modify any spec files except to add `**Status: done**`, `## Deviations`, or `## Questions` sections.
diff --git a/apps/claude-code/unic-confluence/docs/plans/00-convert-repo-tooling-to-pnpm.md b/apps/claude-code/unic-confluence/docs/plans/00-convert-repo-tooling-to-pnpm.md
deleted file mode 100644
index a7beddf4..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/00-convert-repo-tooling-to-pnpm.md
+++ /dev/null
@@ -1,237 +0,0 @@
-# 00. Convert Repo Tooling to pnpm
-**Status: done — 2026-04-24**
-
-**Priority:** P0
-**Effort:** S
-**Depends on:** none
-**Touches:** `package.json`, `package-lock.json`, `pnpm-workspace.yaml`, `.npmrc`, `commands/unic-confluence.md`, `README.md`
-
-## Context
-
-The repo currently uses npm with a `package-lock.json`. The organisation standard is pnpm, which the README already mentions for installation (`pnpm install`). Having `package-lock.json` alongside pnpm usage creates confusion and means two lockfile formats can silently diverge. Moving to pnpm workspace mode with a catalog reference for `marked` gives supply-chain hardening (pinned version in one place, `save-exact`, engine checks) and aligns the developer workflow with what the README already describes.
-
-## Current behaviour
-
-- `package-lock.json` exists at repo root, managed by npm.
-- `package.json` (root) has no `packageManager` field and no `engines` field.
-- `marked` is declared as `"marked": "17.0.5"` in `dependencies` (not a catalog reference).
-- `commands/unic-confluence.md`, step 0, runs `npm install --prefix "${CLAUDE_PLUGIN_ROOT}"`.
-- `README.md` documents `pnpm install` for repo setup but does not mention `pnpm confluence` for running the command (uses `npm run confluence --`).
-
-Relevant file paths and content:
-
-`package.json` (full current content):
-```json
-{
-  "name": "unic-confluence",
-  "private": true,
-  "license": "LGPL-3.0-or-later",
-  "type": "module",
-  "dependencies": {
-    "marked": "17.0.5"
-  }
-}
-```
-
-`commands/unic-confluence.md`, step 0 (lines 10-11):
-```
-### 0. Ensure dependencies are installed
-test -d "${CLAUDE_PLUGIN_ROOT}/node_modules/marked" || npm install --prefix "${CLAUDE_PLUGIN_ROOT}"
-```
-
-## Target behaviour
-
-- `package-lock.json` is deleted.
-- `pnpm-workspace.yaml` exists at repo root with catalog entry for `marked`.
-- `.npmrc` enforces `engine-strict=true` and `save-exact=true`.
-- `package.json` declares `packageManager` (exact pnpm version), `engines` (`node>=24`, `pnpm>=10`), and references `marked` via `catalog:`.
-- `pnpm-lock.yaml` is generated by running `pnpm install`.
-- `commands/unic-confluence.md` step 0 uses `pnpm install`.
-- `README.md` uses `pnpm confluence` (not `npm run confluence --`) for all run examples and has no remaining `npm install` references in install sections.
-- `node scripts/push-to-confluence.mjs` (no args) exits 1 with a usage message — not a `require()` crash — confirming `marked` still loads correctly after the lockfile swap.
-
-## Implementation steps
-
-### Step 1 — Delete `package-lock.json`
-
-```sh
-rm /Users/oriol.torrent/Sites/UNIC/unic-claude-code-confluence/package-lock.json
-```
-
-### Step 2 — Create `pnpm-workspace.yaml`
-
-Create the file at repo root with this exact content (spaces for YAML, not tabs):
-
-```yaml
-packages:
-  - "."
-
-catalog:
-  marked: "17.0.5"
-
-catalogMode: prefer
-preferWorkspacePackages: true
-strictDepBuilds: true
-minimumReleaseAge: 1440
-trustPolicy: no-downgrade
-useNodeVersion: "24.x.y"   # Ralph: replace with latest Node 24.x LTS patch from https://nodejs.org/en/download
-```
-
-> Ralph: Before writing this file, fetch the latest Node 24.x LTS patch version from https://nodejs.org/en/download and substitute `24.x.y` with the real version string (e.g. `24.0.1`).
-
-### Step 3 — Create `.npmrc`
-
-Create the file at repo root:
-
-```
-engine-strict=true
-save-exact=true
-```
-
-### Step 4 — Update `package.json`
-
-Replace the full content of `package.json` with:
-
-```json
-{
-	"name": "unic-confluence",
-	"private": true,
-	"license": "LGPL-3.0-or-later",
-	"type": "module",
-	"packageManager": "pnpm@10.x.y",
-	"engines": {
-		"node": ">=24",
-		"pnpm": ">=10"
-	},
-	"dependencies": {
-		"marked": "catalog:"
-	}
-}
-```
-
-> Ralph: Before writing this file, run `pnpm --version` and substitute `10.x.y` in `packageManager` with the exact version string returned (e.g. `"pnpm@10.11.0"`).
-
-### Step 5 — Run `pnpm install`
-
-```sh
-pnpm install
-```
-
-This generates `pnpm-lock.yaml` and installs `marked` via the catalog. Confirm `node_modules/marked` is present.
-
-### Step 6 — Verify the script still loads `marked`
-
-```sh
-node scripts/push-to-confluence.mjs
-```
-
-Expected: exits 1 with a usage message such as `Usage: npm run confluence -- {pageId} {file.md}`. There must be no `Error [ERR_MODULE_NOT_FOUND]` or `Cannot find module` crash. If a crash occurs, check that `pnpm install` completed and `node_modules/marked/src/marked.cjs` (or equivalent) exists.
-
-### Step 7 — Update `commands/unic-confluence.md` step 0
-
-File: `commands/unic-confluence.md`
-
-Before (step 0 body):
-```
-test -d "${CLAUDE_PLUGIN_ROOT}/node_modules/marked" || npm install --prefix "${CLAUDE_PLUGIN_ROOT}"
-```
-
-After:
-```
-test -d "${CLAUDE_PLUGIN_ROOT}/node_modules/marked" || pnpm install --prefix "${CLAUDE_PLUGIN_ROOT}"
-```
-
-### Step 8 — Update `README.md` run examples
-
-Scan `README.md` for every occurrence of `npm run confluence --` and replace with `pnpm confluence`. Also confirm no `npm install` references remain in any install/setup section (the README already uses `pnpm install` for setup, but verify).
-
-Replace pattern (all occurrences):
-```
-npm run confluence --
-```
-With:
-```
-pnpm confluence
-```
-
-If any line reads `npm install` in an install/setup context, replace with `pnpm install`.
-
-> Note: `package.json` has no `"scripts"` field today, so `pnpm confluence` will not work until a scripts section is added. If the README references `pnpm confluence` as a shorthand that requires a script alias, add a `scripts` entry to `package.json`:
-> ```json
-> "scripts": {
->   "confluence": "node scripts/push-to-confluence.mjs"
-> }
-> ```
-> This also aligns with the existing `npm run confluence --` documentation pattern.
-
-### Step 9 — Commit `pnpm-lock.yaml`
-
-Stage and commit all changed files including the newly generated `pnpm-lock.yaml`:
-
-```sh
-git add pnpm-workspace.yaml pnpm-lock.yaml .npmrc package.json commands/unic-confluence.md README.md
-git rm package-lock.json
-git commit -m "chore: migrate from npm to pnpm with workspace catalog"
-```
-
-## Test cases
-
-| Input | Expected stdout/stderr | Exit code |
-|---|---|---|
-| `pnpm install --frozen-lockfile` on a fresh clone | No error output; `node_modules/marked` present | 0 |
-| `node scripts/push-to-confluence.mjs` (no args) | `Usage: ...` message to stderr | 1 |
-| `node scripts/push-to-confluence.mjs` (no args) | No `ERR_MODULE_NOT_FOUND` or `Cannot find module` in stderr | — |
-| `pnpm confluence` (no args) | Same usage message as above | 1 |
-
-## Acceptance criteria
-
-- `package-lock.json` does not exist in the repo root.
-- `pnpm-lock.yaml` exists and is committed.
-- `pnpm-workspace.yaml` exists with a `catalog.marked` entry matching the version in `package.json`'s `catalog:` reference.
-- `.npmrc` exists with `engine-strict=true` and `save-exact=true`.
-- `package.json` has `packageManager`, `engines`, and `"marked": "catalog:"`.
-- `commands/unic-confluence.md` uses `pnpm install`, not `npm install`.
-- `README.md` contains no `npm run confluence --` occurrences.
-- `pnpm install --frozen-lockfile` exits 0.
-- `node scripts/push-to-confluence.mjs` exits 1 with usage message (no crash).
-- Lint passes (if a lint script exists).
-- Slash command `/unic-confluence` still works end-to-end (credentials + publish flow unaffected by lockfile change).
-
-## Verification
-
-```sh
-# 1. Confirm no package-lock.json
-ls package-lock.json 2>&1 | grep "No such file"
-
-# 2. Confirm pnpm-lock.yaml exists
-ls pnpm-lock.yaml
-
-# 3. Confirm catalog reference in package.json
-grep '"marked": "catalog:"' package.json
-
-# 4. Confirm pnpm install is clean (frozen)
-pnpm install --frozen-lockfile
-
-# 5. Confirm script loads marked correctly
-node scripts/push-to-confluence.mjs; echo "Exit: $?"
-# Expected: exit 1 with usage text, no module-not-found crash
-
-# 6. Confirm README has no npm run confluence reference
-grep -n "npm run confluence" README.md && echo "FAIL: npm run confluence still in README" || echo "OK"
-```
-
-## Out of scope
-
-- No CI pipeline changes (spec 10).
-- No Biome or other linter setup (spec 10).
-- No version bump to `plugin.json` or `marketplace.json`.
-- No changes to `scripts/push-to-confluence.mjs` logic.
-- No changes to `.claude-plugin/` files.
-- No addition of test infrastructure.
-
-## Follow-ups
-
-- Once CI exists (spec 10), add a `pnpm install --frozen-lockfile` step to the workflow to catch lockfile drift.
-- Consider adding Renovate or Dependabot config to auto-update the `marked` catalog entry.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/01-drop-verify-from-slash-command.md b/apps/claude-code/unic-confluence/docs/plans/01-drop-verify-from-slash-command.md
deleted file mode 100644
index 152c0aa7..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/01-drop-verify-from-slash-command.md
+++ /dev/null
@@ -1,189 +0,0 @@
-# 01. Drop --verify from Slash Command, Replace with Cheap Auth Probe
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** S
-**Depends on:** none (spec 07 is a follow-up, not a prerequisite)
-**Touches:** `scripts/push-to-confluence.mjs`, `commands/unic-confluence.md`
-
-## Context
-
-Every time a user runs `/unic-confluence`, the slash command calls `--verify` before publishing. The `--verify` subcommand iterates over every entry in `confluence-pages.json` and performs an HTTP GET for each page to confirm it still exists. On a repo with 20 mapped pages this fires 20 HTTP requests before even starting the actual publish. Latency scales linearly with the number of mapped pages and the pre-flight check provides marginal safety value — if a specific page is missing, the subsequent publish GET/PUT will surface that error anyway. The fix is to replace `--verify` in the slash command with a single cheap auth probe: one GET to `/wiki/api/v2/pages?limit=1`. This proves credentials are valid without touching any specific pages.
-
-Spec 07 will later formalise this as a `--check-auth` subcommand. This spec implements a temporary `--ping` arg that is cheap to add now and easy to rename in spec 07.
-
-## Current behaviour
-
-`commands/unic-confluence.md`, step 2 (exact text):
-
-```markdown
-### 2. Check credentials
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --verify
-
-If credentials missing, instruct user to run --setup.
-```
-
-`scripts/push-to-confluence.mjs`, `main()` starting at line 388:
-
-```js
-async function main() {
-	const args = process.argv.slice(2);
-	if (args[0] === "--setup") { await runSetup(); process.exit(0); }
-	if (args[0] === "--verify") { await runVerify(); return; }
-	if (args.length < 2) { console.error("Usage: npm run confluence -- {pageId} {file.md}"); process.exit(1); }
-```
-
-The `--verify` branch (line 391) calls `runVerify()`, which GETs every page in `confluence-pages.json` (function defined somewhere between lines 252-388 — not shown in the provided excerpt but known to perform N HTTP GETs). This happens on every publish invocation through the slash command.
-
-## Target behaviour
-
-### `scripts/push-to-confluence.mjs`
-
-A new `--ping` branch is added in `main()` immediately before the existing `--verify` branch. When `--ping` is passed:
-
-1. Load credentials via `loadCredentials()` (exits 1 with setup hint if credentials are missing).
-2. Make a single GET to `{baseUrl}/wiki/api/v2/pages?limit=1`.
-3. If the response is 2xx → exit 0, no stdout.
-4. If the network call throws → print `err.message` to stderr and exit 1.
-5. For non-2xx responses → call `handleHttpError(res.status, "")` (exits 1 with a specific message for 401/403/404/other).
-
-The existing `--verify` subcommand is left unchanged — it remains useful as an interactive diagnostic tool.
-
-### `commands/unic-confluence.md`
-
-Step 2 is replaced: the expensive `--verify` call becomes `--ping`. The heading and explanatory text are updated to describe the lightweight nature of the check.
-
-### Edge cases
-
-- `--ping` with no credentials file and no env vars → `loadCredentials()` prints the setup hint and exits 1 (unchanged behaviour, propagated from the existing helper).
-- `--ping` with a valid token but network unreachable → exits 1 with the "Cannot reach Confluence — check VPN/network connectivity" message from `httpsRequest`'s error handler (line 85).
-- `--ping` with a 401 → exits 1 with the API token rejected message from `handleHttpError` (line 99).
-- `--ping` with a 403 → exits 1 with the access denied message (line 100).
-
-## Implementation steps
-
-### Step 1 — Add `--ping` branch to `main()` in `scripts/push-to-confluence.mjs`
-
-File: `scripts/push-to-confluence.mjs`
-
-Locate `main()` starting at line 388. The current top of the function reads:
-
-```js
-async function main() {
-	const args = process.argv.slice(2);
-	if (args[0] === "--setup") { await runSetup(); process.exit(0); }
-	if (args[0] === "--verify") { await runVerify(); return; }
-	if (args.length < 2) { console.error("Usage: npm run confluence -- {pageId} {file.md}"); process.exit(1); }
-```
-
-Insert the `--ping` branch between `--setup` and `--verify`:
-
-```js
-async function main() {
-	const args = process.argv.slice(2);
-	if (args[0] === "--setup") { await runSetup(); process.exit(0); }
-	if (args[0] === "--ping") {
-		const { url: baseUrl, username, token } = loadCredentials();
-		const authHeader = makeBasicAuth(username, token);
-		let res;
-		try {
-			res = await httpsRequest("GET", `${baseUrl.replace(/\/$/, "")}/wiki/api/v2/pages?limit=1`, authHeader);
-		} catch (err) {
-			console.error(err.message);
-			process.exit(1);
-		}
-		if (res.status >= 200 && res.status < 300) { process.exit(0); }
-		handleHttpError(res.status, "");
-	}
-	if (args[0] === "--verify") { await runVerify(); return; }
-	if (args.length < 2) { console.error("Usage: npm run confluence -- {pageId} {file.md}"); process.exit(1); }
-```
-
-No other changes to this function at this step.
-
-### Step 2 — Update step 2 in `commands/unic-confluence.md`
-
-File: `commands/unic-confluence.md`
-
-Before (step 2, exact text):
-
-```markdown
-### 2. Check credentials
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --verify
-
-If credentials missing, instruct user to run --setup.
-```
-
-After:
-
-```markdown
-### 2. Check credentials (quick auth probe)
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --ping
-
-Performs a single HTTP GET to verify credentials are valid. If credentials are missing or rejected, instruct user to run --setup.
-```
-
-No other changes to `commands/unic-confluence.md`.
-
-## Test cases
-
-| Command | Precondition | Expected stderr | Expected stdout | Exit code |
-|---|---|---|---|---|
-| `node scripts/push-to-confluence.mjs --ping` | Valid credentials, network reachable | (empty) | (empty) | 0 |
-| `node scripts/push-to-confluence.mjs --ping` | No credentials file, no env vars | `Run \`npm run confluence -- --setup\` to configure credentials` | (empty) | 1 |
-| `node scripts/push-to-confluence.mjs --ping` | Credentials present, bad token (401) | `API token rejected — generate a new one at...` | (empty) | 1 |
-| `node scripts/push-to-confluence.mjs --ping` | Credentials present, 403 response | `Access denied — check that your API token...` | (empty) | 1 |
-| `node scripts/push-to-confluence.mjs --ping` | Credentials present, network down | `Cannot reach Confluence — check VPN/network connectivity` | (empty) | 1 |
-| `node scripts/push-to-confluence.mjs --verify` | Any | Unchanged — `--verify` still works as before | — | unchanged |
-
-## Acceptance criteria
-
-- `--ping` is handled in `main()` before `--verify` (to avoid accidental fallthrough).
-- `--ping` with valid credentials makes exactly one HTTP request (GET `/wiki/api/v2/pages?limit=1`).
-- `--ping` exits 0 on a valid 2xx response with no output.
-- `--ping` exits 1 for any error or non-2xx response with an appropriate message to stderr.
-- `--verify` behaviour is unchanged (this spec does not touch `runVerify`).
-- `commands/unic-confluence.md` step 2 calls `--ping`, not `--verify`.
-- Lint passes.
-- Slash command `/unic-confluence` still works end-to-end: the pre-flight auth probe fires, and a valid publish completes successfully.
-
-## Verification
-
-```sh
-# 1. Confirm --ping branch exists in the script
-grep -n "\-\-ping" scripts/push-to-confluence.mjs
-
-# 2. Confirm commands file uses --ping
-grep "\-\-ping" commands/unic-confluence.md
-
-# 3. Confirm --verify is still present (not removed)
-grep -n "\-\-verify" scripts/push-to-confluence.mjs
-
-# 4. Smoke test with no credentials
-unset CONFLUENCE_URL CONFLUENCE_USER CONFLUENCE_TOKEN
-node scripts/push-to-confluence.mjs --ping; echo "Exit: $?"
-# Expected: exit 1 with setup hint
-
-# 5. With valid credentials set in env (replace with real values for manual test):
-# CONFLUENCE_URL=https://yoursite.atlassian.net \
-# CONFLUENCE_USER=you@example.com \
-# CONFLUENCE_TOKEN=your-token \
-# node scripts/push-to-confluence.mjs --ping; echo "Exit: $?"
-# Expected: exit 0
-```
-
-## Out of scope
-
-- Do not modify or remove `--verify` / `runVerify()`. The `--verify` subcommand remains fully functional.
-- Do not implement `--check-auth` as a formal subcommand — that is spec 07. This spec adds `--ping` as a minimal stepping stone.
-- Do not add any new dependencies.
-- No version bump.
-- No changes to `confluence-pages.json` or credential storage logic.
-- No changes to `handleHttpError` or `httpsRequest`.
-
-## Follow-ups
-
-- Spec 07 will formalise `--check-auth` as the canonical auth-probe subcommand. At that point `--ping` can be aliased to `--check-auth` or replaced.
-- If the Confluence API returns 200 even for invalid tokens on the `pages?limit=1` endpoint (e.g. returns an empty list with 200), consider probing `/wiki/api/v2/spaces?limit=1` or checking the response body for a `results` array instead.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/02-document-marker-syntax-readme.md b/apps/claude-code/unic-confluence/docs/plans/02-document-marker-syntax-readme.md
deleted file mode 100644
index 8f54d8bf..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/02-document-marker-syntax-readme.md
+++ /dev/null
@@ -1,241 +0,0 @@
-# 02. Document Marker Syntax in README
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** S
-**Depends on:** none
-**Touches:** `README.md`
-
-## Context
-
-The `[AUTO_INSERT_START:label]` / `[AUTO_INSERT_END:label]` marker syntax is the primary mechanism for safely injecting content into Confluence pages without overwriting surrounding content. It is implemented in `injectContent()` at lines 121-148 of `scripts/push-to-confluence.mjs`, but it is completely absent from `README.md`. Users discovering the plugin for the first time have no way to know they need to set up markers on their Confluence pages before publishing — they will trigger the silent-append foot-gun (Strategy 3, lines 197-199) and discover doubled content after a second publish run. Similarly, the `confluence-pages.json` mapping file is referenced in the script (line 213) but is not documented in the README alongside the usage example. This is a docs-only change; no script logic is touched.
-
-## Current behaviour
-
-`README.md` has a "Usage" section that shows how to invoke the script but does not mention:
-
-- The `[AUTO_INSERT_START:label]` / `[AUTO_INSERT_END:label]` markers.
-- How to add markers to a Confluence page.
-- That labels are case-sensitive.
-- What happens when markers are absent (today: silently appends; after spec 03: errors with a hint).
-- The legacy `md-start` / `md-end` anchor-macro fallback.
-- The `confluence-pages.json` file format or where it lives.
-
-The "Per-repo setup" section (if it exists) or equivalent does not cover page preparation.
-
-Relevant source code for accuracy:
-
-`scripts/push-to-confluence.mjs`, lines 121-148 (marker detection and injection):
-```js
-const TEXT_START_RE = /(?:<p>\s*)?\[AUTO_INSERT_START:\s*([^\]]+?)\s*\](?:\s*<\/p>)?/;
-const TEXT_END_RE   = /(?:<p>\s*)?\[AUTO_INSERT_END:\s*([^\]]+?)\s*\](?:\s*<\/p>)?/;
-
-function injectContent(existingBody, newHtml, title) {
-	const hasStart = TEXT_START_RE.test(existingBody);
-	const hasEnd   = TEXT_END_RE.test(existingBody);
-
-	// Strategy 1: plain-text markers
-	if (hasStart || hasEnd) {
-		if (hasStart !== hasEnd) {
-			console.error(`Found [AUTO_INSERT_START] without a matching [AUTO_INSERT_END]...`);
-			process.exit(1);
-		}
-		// ... slices body and injects newHtml between markers
-	}
-	// Strategy 2: anchor macros (legacy fallback)
-	// ...
-	// Strategy 3: append (no markers found) — FOOT-GUN: doubles content on each run
-	return existingBody + "\n" + newHtml;
-}
-```
-
-`scripts/push-to-confluence.mjs`, lines 207-252 (`resolvePageId` reading `confluence-pages.json`):
-```js
-function resolvePageId(arg) {
-	if (isNumericId(arg)) { ... }
-	const pagesPath = path.join(process.cwd(), "confluence-pages.json");
-	if (!existsSync(pagesPath)) { console.error("confluence-pages.json not found ..."); process.exit(1); }
-	// ... parses JSON and returns pages[arg]
-}
-```
-
-## Target behaviour
-
-`README.md` gains a "Page setup" section (placed between the existing "Usage" section and "Per-repo setup", or appended to "Usage" if "Per-repo setup" does not exist). The section covers:
-
-1. **What markers look like** — show the raw text a user types into Confluence.
-2. **How to add them** — step-by-step: open the Confluence page → Edit → type the marker lines → Save.
-3. **Label rules** — labels are case-sensitive; the label in `[AUTO_INSERT_START:label]` must exactly match the label in `[AUTO_INSERT_END:label]`.
-4. **What happens without markers** — currently silently appends (doubles content on re-run); after spec 03 this will become an explicit error with a `--replace-all` opt-out.
-5. **Legacy anchor-macro fallback** — mention that old pages may use Confluence anchor macros named `md-start` / `md-end` instead of text markers; this is deprecated and new pages should use text markers.
-6. **`confluence-pages.json`** — document the file format and that it lives at the repo root; show a minimal example alongside the usage example.
-
-The section must include the exact copy-paste example block shown below so users can drop it straight into Confluence:
-
-```
-[AUTO_INSERT_START: my-docs]
-
-(Claude Code will inject content here)
-
-[AUTO_INSERT_END: my-docs]
-```
-
-## Implementation steps
-
-### Step 1 — Read the current `README.md`
-
-Read the full file to understand the existing section order, heading levels, and writing style before editing.
-
-File: `README.md` (repo root).
-
-### Step 2 — Add `confluence-pages.json` example to the Usage section
-
-Locate the Usage section. After the existing usage invocation example (e.g. `pnpm confluence my-page docs/my-file.md`), add a note explaining the `my-page` argument can be a numeric Confluence page ID **or** a key defined in `confluence-pages.json`. Append a minimal example:
-
-```markdown
-#### `confluence-pages.json` (optional key mapping)
-
-Place this file at your repo root to use short keys instead of numeric page IDs:
-
-```json
-{
-  "my-docs": 123456789,
-  "another-page": 987654321
-}
-```
-
-Then publish using the key:
-
-```sh
-pnpm confluence my-docs docs/my-file.md
-```
-```
-
-### Step 3 — Add the "Page setup" section
-
-After the Usage section (or after the `confluence-pages.json` sub-section added in step 2), insert the following new section. Match the existing heading level convention in the README (use `##` if top-level sections use `##`, `###` if they use `###`).
-
-Suggested content (adapt heading level to match the existing document):
-
-```markdown
-## Page setup — injection markers
-
-The script injects your Markdown content into a Confluence page rather than replacing the whole page. To control *where* the content lands, add injection markers directly to the Confluence page body.
-
-### Adding markers to a Confluence page
-
-1. Open the target Confluence page and click **Edit**.
-2. Place your cursor where you want the injected content to appear.
-3. Type the start marker on its own line:
-   ```
-   [AUTO_INSERT_START: my-docs]
-   ```
-4. Leave a blank line (optional placeholder text helps with visual orientation):
-   ```
-   (Claude Code will inject content here)
-   ```
-5. Type the end marker on its own line:
-   ```
-   [AUTO_INSERT_END: my-docs]
-   ```
-6. **Save** the page.
-
-Full copy-paste block:
-
-```
-[AUTO_INSERT_START: my-docs]
-
-(Claude Code will inject content here)
-
-[AUTO_INSERT_END: my-docs]
-```
-
-### Label rules
-
-- The label (`my-docs` above) is **case-sensitive**. `My-Docs` and `my-docs` are different labels.
-- The label in `[AUTO_INSERT_START:label]` must exactly match the label in `[AUTO_INSERT_END:label]`.
-- Whitespace around the label is trimmed — `[AUTO_INSERT_START: my-docs]` and `[AUTO_INSERT_START:my-docs]` are equivalent.
-- A page can have only one marker pair. Multiple pairs are not supported.
-
-### What happens without markers
-
-If the page has no markers, the script currently appends the new HTML after all existing content. **Running the publish command twice will double the content.** A future update (spec 03) will change this to an explicit error with a `--replace-all` opt-out flag.
-
-### Legacy anchor-macro fallback (deprecated)
-
-Pages set up before text markers were introduced may use Confluence anchor macros instead:
-
-- Start anchor: macro named `md-start`
-- End anchor: macro named `md-end`
-
-This fallback is still supported but deprecated. Migrate legacy pages to text markers when convenient.
-```
-
-### Step 4 — Verify no unintended whitespace or broken Markdown
-
-After editing, re-read `README.md` and confirm:
-- No raw `[AUTO_INSERT_START` / `[AUTO_INSERT_END` markers appear outside fenced code blocks (they would be mistaken for actual page markers if a tool renders the README as Confluence content).
-- All fenced code blocks are properly closed.
-- All heading levels are consistent with the rest of the document.
-
-## Test cases
-
-This is a docs-only change; there are no script test cases.
-
-| Check | Expected result |
-|---|---|
-| `grep "AUTO_INSERT_START" README.md` | At least one occurrence (in the docs example) |
-| `grep "confluence-pages.json" README.md` | At least one occurrence with a JSON example |
-| `grep "md-start" README.md` | At least one occurrence in the legacy section |
-| `grep "case-sensitive" README.md` | Mentioned |
-| README renders correctly in GitHub Markdown preview | No broken fences, all code blocks closed |
-
-## Acceptance criteria
-
-- `README.md` contains a "Page setup" (or equivalent title) section documenting the marker syntax.
-- The exact copy-paste block `[AUTO_INSERT_START: my-docs]` / `[AUTO_INSERT_END: my-docs]` appears verbatim in a fenced code block.
-- `confluence-pages.json` format is documented with a JSON example.
-- Label case-sensitivity is explicitly stated.
-- Legacy `md-start` / `md-end` anchor macros are mentioned as deprecated.
-- The consequence of missing markers (silent append / future error) is noted.
-- No changes made to any file other than `README.md`.
-- Lint passes (Markdown linting if configured; otherwise manual review).
-- Slash command `/unic-confluence` still works end-to-end (no script changes).
-
-## Verification
-
-```sh
-# 1. Confirm marker docs are present
-grep -c "AUTO_INSERT_START" README.md
-# Expected: >= 1
-
-# 2. Confirm confluence-pages.json is documented
-grep -c "confluence-pages.json" README.md
-# Expected: >= 1
-
-# 3. Confirm legacy anchor section exists
-grep -c "md-start" README.md
-# Expected: >= 1
-
-# 4. Confirm no script files were modified
-git diff --name-only
-# Expected: only README.md
-```
-
-## Out of scope
-
-- Do not modify `scripts/push-to-confluence.mjs`.
-- Do not modify `commands/unic-confluence.md`.
-- Do not document `--dry-run` — that flag does not exist yet (spec 04).
-- Do not document `--replace-all` — that flag does not exist yet (spec 03).
-- Do not document `--check-auth` — that subcommand does not exist yet (spec 07).
-- Do not change marker behaviour, regex patterns, or any logic.
-- No version bump.
-
-## Follow-ups
-
-- After spec 03 lands, update the "What happens without markers" paragraph to describe the new error behaviour and mention `--replace-all`.
-- After spec 04 lands, add a short `--dry-run` note to the Usage section.
-- Consider adding a troubleshooting section covering the 401/403/404 error messages that `handleHttpError` produces (lines 98-106).
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/03-refuse-publish-without-markers.md b/apps/claude-code/unic-confluence/docs/plans/03-refuse-publish-without-markers.md
deleted file mode 100644
index 4e6045e6..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/03-refuse-publish-without-markers.md
+++ /dev/null
@@ -1,353 +0,0 @@
-# 03. Refuse Publish Without Markers (Breaking Change → v2.0.0)
-**Status: done — 2026-04-24**
-
-**Priority:** P0
-**Effort:** M
-**Depends on:** 02 (README should document markers before this error starts firing)
-**Touches:** `scripts/push-to-confluence.mjs`, `commands/unic-confluence.md`, `.claude-plugin/plugin.json`, `.claude-plugin/marketplace.json`
-
-## Context
-
-`injectContent()` at lines 197-199 of `scripts/push-to-confluence.mjs` has a silent foot-gun: when neither text markers nor anchor macros are found on a Confluence page, it appends the new HTML after all existing page content. Running the publish command a second time doubles the content; a third run triples it. This is the worst-case default because it is silent, cumulative, and affects production Confluence pages. The correct default is to refuse and tell the user what to do. An explicit `--replace-all` opt-in flag gives a controlled escape hatch for users who genuinely want to overwrite the full body, with a local backup created automatically as a safety net. This is a breaking change because existing workflows that relied on the append behaviour will now error. The version is bumped to 2.0.0.
-
-## Current behaviour
-
-`scripts/push-to-confluence.mjs`, lines 197-199 (Strategy 3 inside `injectContent`):
-
-```js
-// Strategy 3: append (no markers found) — FOOT-GUN: doubles content on each run
-return existingBody + "\n" + newHtml;
-```
-
-`injectContent` function signature at line 124:
-
-```js
-function injectContent(existingBody, newHtml, title) {
-```
-
-Call site in `main()`, approximately line 469:
-
-```js
-const newBody = injectContent(existingBody, html, title);
-```
-
-`import` destructure at line 11:
-
-```js
-import {
-	chmodSync,
-	existsSync,
-	readFileSync,
-	statSync,
-	writeFileSync,
-} from "fs";
-```
-
-`.claude-plugin/plugin.json` current version:
-
-```json
-{
-  "version": "1.0.2"
-}
-```
-
-`.claude-plugin/marketplace.json` current version:
-
-```json
-{
-  "plugins": [{
-    "version": "1.0.2"
-  }]
-}
-```
-
-## Target behaviour
-
-### `scripts/push-to-confluence.mjs`
-
-1. `injectContent` signature changes to accept an options object:
-   ```
-   injectContent(existingBody, newHtml, title, { replaceAll = false, pageId, version } = {})
-   ```
-
-2. Strategy 3 (the append foot-gun at lines 197-199) is replaced with:
-   - If `replaceAll` is `true`: write the current `existingBody` to a backup file at `~/.unic-confluence/backups/{pageId}-v{version}.html`, log the backup path to stdout, and return `newHtml` as the full replacement body.
-   - If `replaceAll` is `false` (default): print an error message to stderr explaining that no markers were found and exit 1.
-
-3. `main()` parses `--replace-all` and `--dry-run` (see spec 04) from `process.argv` using a flags-first pattern that also filters positional args:
-   ```js
-   const replaceAll = args.includes("--replace-all");
-   const positionalArgs = args.filter(a => !a.startsWith("--"));
-   ```
-
-4. The `injectContent` call site passes `{ replaceAll, pageId, version }` so the backup filename is deterministic.
-
-5. `mkdirSync` is added to the existing `fs` import destructure.
-
-6. The backup uses `mkdirSync` (with `{ recursive: true }`) and `writeFileSync` — keeping `injectContent` synchronous.
-
-### `commands/unic-confluence.md`
-
-Step 2 (the pre-flight check) is updated to mention `--replace-all` as an option if the user gets the no-markers error.
-
-### `.claude-plugin/plugin.json` and `.claude-plugin/marketplace.json`
-
-Version bumped from `1.0.2` to `2.0.0`.
-
-### Edge cases
-
-- Page with valid marker pair → `replaceAll` is never reached; markers take precedence. No backup is created.
-- Mismatched labels (start without matching end) → existing error message at lines 130-132 fires before Strategy 3 is reached. Unchanged.
-- `--replace-all` with a marker page → markers are found in Strategy 1 or 2 and content is injected normally; `replaceAll` is never consulted. The flag is silently ignored when markers exist.
-- `pageId` or `version` undefined when `replaceAll = true` → backup filename degrades gracefully: use `"unknown-page"` for pageId and `"unknown-version"` for version. Log a warning.
-
-## Implementation steps
-
-### Step 1 — Add `mkdirSync` to the `fs` import
-
-File: `scripts/push-to-confluence.mjs`, lines 11-17.
-
-Before:
-```js
-import {
-	chmodSync,
-	existsSync,
-	readFileSync,
-	statSync,
-	writeFileSync,
-} from "fs";
-```
-
-After:
-```js
-import {
-	chmodSync,
-	existsSync,
-	mkdirSync,
-	readFileSync,
-	statSync,
-	writeFileSync,
-} from "fs";
-```
-
-### Step 2 — Update `injectContent` signature
-
-File: `scripts/push-to-confluence.mjs`, line 124.
-
-Before:
-```js
-function injectContent(existingBody, newHtml, title) {
-```
-
-After:
-```js
-function injectContent(existingBody, newHtml, title, { replaceAll = false, pageId, version } = {}) {
-```
-
-### Step 3 — Replace Strategy 3 (lines 197-199)
-
-File: `scripts/push-to-confluence.mjs`, lines 197-199.
-
-Before:
-```js
-	// Strategy 3: append (no markers found) — FOOT-GUN: doubles content on each run
-	return existingBody + "\n" + newHtml;
-```
-
-After:
-```js
-	// Strategy 3: no markers found
-	if (replaceAll) {
-		const safePageId = pageId ?? "unknown-page";
-		const safeVersion = version ?? "unknown-version";
-		const backupDir = path.join(os.homedir(), ".unic-confluence", "backups");
-		mkdirSync(backupDir, { recursive: true });
-		const backupPath = path.join(backupDir, `${safePageId}-v${safeVersion}.html`);
-		writeFileSync(backupPath, existingBody, "utf8");
-		console.log(`Backup saved to ${backupPath}`);
-		return newHtml;
-	}
-	console.error(
-		`No [AUTO_INSERT_START:label] / [AUTO_INSERT_END:label] markers found on page "${title}". ` +
-		`Add markers to the Confluence page, or use --replace-all to overwrite the full body.`
-	);
-	process.exit(1);
-```
-
-### Step 4 — Update `main()` to parse `--replace-all`
-
-File: `scripts/push-to-confluence.mjs`, `main()` starting at line 388.
-
-Locate the top of `main()`. The current arg parsing reads:
-
-```js
-	const args = process.argv.slice(2);
-	if (args[0] === "--setup") { await runSetup(); process.exit(0); }
-	if (args[0] === "--ping") { ... }   // added in spec 01
-	if (args[0] === "--verify") { await runVerify(); return; }
-	if (args.length < 2) { console.error("Usage: npm run confluence -- {pageId} {file.md}"); process.exit(1); }
-	const [pageArg, filePath] = args;
-```
-
-Replace the positional-arg extraction portion (the last two lines shown) with:
-
-```js
-	const replaceAll = args.includes("--replace-all");
-	const positionalArgs = args.filter(a => !a.startsWith("--"));
-	if (positionalArgs.length < 2) {
-		console.error("Usage: node scripts/push-to-confluence.mjs [--replace-all] {pageId} {file.md}");
-		process.exit(1);
-	}
-	const [pageArg, filePath] = positionalArgs;
-```
-
-> Note: If spec 04 (`--dry-run`) has already landed when Ralph picks up this ticket, `dryRun` will already be parsed here. In that case, add `replaceAll` to the existing flags block rather than replacing it. The positional-args filter already handles multiple flags.
-
-### Step 5 — Update the `injectContent` call site
-
-File: `scripts/push-to-confluence.mjs`, approximately line 469 (inside `main()`, after the GET response is parsed to get `existingBody`, `version`, and after `html` is prepared from the Markdown conversion).
-
-Before:
-```js
-	const newBody = injectContent(existingBody, html, title);
-```
-
-After:
-```js
-	const newBody = injectContent(existingBody, html, title, { replaceAll, pageId, version });
-```
-
-> Ralph: `pageId` and `version` are resolved earlier in `main()` (pageId from `resolvePageId(pageArg)` around line 394; version from the GET response body — look for `page.version.number` or equivalent in the response parsing block). Confirm the exact variable names by reading lines 394-469 of the script.
-
-### Step 6 — Update `commands/unic-confluence.md`
-
-File: `commands/unic-confluence.md`.
-
-In step 2 (the pre-flight check description), add a note about `--replace-all` after the existing setup hint:
-
-Before (step 2 body, as updated by spec 01):
-```markdown
-### 2. Check credentials (quick auth probe)
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --ping
-
-Performs a single HTTP GET to verify credentials are valid. If credentials are missing or rejected, instruct user to run --setup.
-```
-
-After:
-```markdown
-### 2. Check credentials (quick auth probe)
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --ping
-
-Performs a single HTTP GET to verify credentials are valid. If credentials are missing or rejected, instruct user to run --setup.
-
-If the publish step fails with "No markers found", the target Confluence page needs [AUTO_INSERT_START:label] / [AUTO_INSERT_END:label] markers. To overwrite the full page body instead, pass --replace-all as part of $ARGUMENTS.
-```
-
-> Note: If spec 01 has NOT yet landed, the step 2 text above will differ. In that case locate the `--verify` text and add the `--replace-all` note below it rather than below `--ping`.
-
-### Step 7 — Version bump
-
-File: `.claude-plugin/plugin.json`
-
-Before:
-```json
-{
-  "author": { "name": "Unic" },
-  "commands": ["./commands/unic-confluence.md"],
-  "description": "Publish markdown files to Confluence pages via the Confluence v2 API.",
-  "name": "unic-confluence",
-  "version": "1.0.2"
-}
-```
-
-After:
-```json
-{
-  "author": { "name": "Unic" },
-  "commands": ["./commands/unic-confluence.md"],
-  "description": "Publish markdown files to Confluence pages via the Confluence v2 API.",
-  "name": "unic-confluence",
-  "version": "2.0.0"
-}
-```
-
-File: `.claude-plugin/marketplace.json`
-
-Locate `"version": "1.0.2"` inside the `plugins` array entry and change to `"version": "2.0.0"`. Also update any top-level `version` field if present.
-
-> Note: Spec 06 will introduce a sync script that derives versions automatically. Until then, update both files manually.
-
-## Test cases
-
-| Scenario | Command | Expected stderr | Expected stdout | Exit code |
-|---|---|---|---|---|
-| Page has valid marker pair | `node scripts/push-to-confluence.mjs <id> file.md` | (none) | `Published …` or similar success message | 0 |
-| Page has no markers, no `--replace-all` | `node scripts/push-to-confluence.mjs <id> file.md` | `No [AUTO_INSERT_START:label] / [AUTO_INSERT_END:label] markers found on page "…". Add markers to the Confluence page, or use --replace-all to overwrite the full body.` | (none) | 1 |
-| Page has no markers, `--replace-all` | `node scripts/push-to-confluence.mjs --replace-all <id> file.md` | (none) | `Backup saved to ~/.unic-confluence/backups/<id>-v<version>.html` | 0 (then PUT proceeds) |
-| `--replace-all` with marker page | `node scripts/push-to-confluence.mjs --replace-all <id> file.md` | (none) | Normal publish success | 0 |
-| Mismatched labels | `node scripts/push-to-confluence.mjs <id> file.md` | `Marker label mismatch on page "…"` (unchanged) | (none) | 1 |
-| Backup directory check | After `--replace-all` run | — | `~/.unic-confluence/backups/<id>-v<N>.html` exists with original page HTML | — |
-
-## Acceptance criteria
-
-- Strategy 3 no longer silently appends.
-- A page with no markers and no `--replace-all` exits 1 with an error message containing both "Add markers" and "--replace-all".
-- A page with no markers and `--replace-all` creates a backup at `~/.unic-confluence/backups/{pageId}-v{version}.html` and returns the new HTML as the full page body.
-- A page with valid markers is unaffected by the `--replace-all` flag.
-- `mkdirSync` is imported from `fs` (no new top-level imports added).
-- `injectContent` remains a synchronous function.
-- The call site at approximately line 469 passes `{ replaceAll, pageId, version }`.
-- `.claude-plugin/plugin.json` version is `2.0.0`.
-- `.claude-plugin/marketplace.json` version is `2.0.0`.
-- Lint passes.
-- Slash command `/unic-confluence` still works end-to-end for a page that has markers.
-
-## Verification
-
-```sh
-# 1. Confirm mkdirSync is imported
-grep "mkdirSync" scripts/push-to-confluence.mjs | head -3
-
-# 2. Confirm Strategy 3 no longer has the append
-grep -n "existingBody.*newHtml\|newHtml.*existingBody" scripts/push-to-confluence.mjs
-# Expected: no match (the append line is gone)
-
-# 3. Confirm error message text is present
-grep -n "Add markers to the Confluence page" scripts/push-to-confluence.mjs
-
-# 4. Confirm replaceAll is parsed in main()
-grep -n "replaceAll" scripts/push-to-confluence.mjs
-
-# 5. Confirm version bump
-grep '"version"' .claude-plugin/plugin.json
-grep '"version"' .claude-plugin/marketplace.json
-# Both should show "2.0.0"
-
-# 6. Confirm injectContent call site passes options
-grep -n "injectContent(" scripts/push-to-confluence.mjs
-# Expected: call includes replaceAll, pageId, version
-
-# 7. Manual test — no-markers case (requires a real page ID or mock):
-# CONFLUENCE_URL=... CONFLUENCE_USER=... CONFLUENCE_TOKEN=... \
-# node scripts/push-to-confluence.mjs <page-without-markers> file.md
-# Expected: exit 1, "No markers found" error
-```
-
-## Out of scope
-
-- Do not add a `--dry-run` flag (spec 04).
-- Do not add `--check-auth` (spec 07).
-- Do not modify the anchor-macro (Strategy 2) detection logic.
-- Do not modify `handleHttpError`, `httpsRequest`, or credential loading.
-- No new npm/pnpm dependencies.
-- No test framework (tests are manual CLI invocations as listed above).
-- Do not implement multiple marker pairs per page.
-
-## Follow-ups
-
-- Ralph: confirm the exact variable names for `pageId` and `version` at the `injectContent` call site (around line 469) and verify the version is the current page version number from the GET response, not the plugin version.
-- Consider a `--replace-all --force` to skip the backup for power users who know what they are doing (deferred).
-- After spec 06 (version sync script) lands, remove manual version bump instructions from this spec's implementation steps.
-- Update README (spec 02) "What happens without markers" paragraph to reflect the new error + `--replace-all` behaviour once this spec lands.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/04-dry-run-flag.md b/apps/claude-code/unic-confluence/docs/plans/04-dry-run-flag.md
deleted file mode 100644
index 6d0d78f0..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/04-dry-run-flag.md
+++ /dev/null
@@ -1,233 +0,0 @@
-# 04. Add --dry-run Flag
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** S
-**Depends on:** 03 (positional-arg parsing refactor in spec 03 is a prerequisite; `--dry-run` reuses the same flags-first pattern)
-**Touches:** `scripts/push-to-confluence.mjs`, `commands/unic-confluence.md`
-
-## Context
-
-Publishing to Confluence is a write operation against a live page. There is currently no way to preview what the script would publish without actually making the PUT request. This is particularly risky when using `--replace-all` (spec 03), where the full page body is replaced. A `--dry-run` flag lets users — and the slash command in future specs — inspect the computed final HTML before committing it to Confluence. The flag does not suppress the GET request (credentials and the current page body are still needed to compute the final output), but it skips the PUT entirely and dumps the computed body to stdout.
-
-## Current behaviour
-
-`scripts/push-to-confluence.mjs`, `main()` starting at line 388. After `resolvePageId`, `readFileSync`, Markdown conversion, the GET request, and `injectContent`, the code immediately makes a PUT request with no opportunity to preview the result. There is no `--dry-run` handling anywhere in the file.
-
-Current top of `main()` (after spec 03 has landed):
-
-```js
-async function main() {
-	const args = process.argv.slice(2);
-	if (args[0] === "--setup") { await runSetup(); process.exit(0); }
-	if (args[0] === "--ping") { ... }
-	if (args[0] === "--verify") { await runVerify(); return; }
-	const replaceAll = args.includes("--replace-all");
-	const positionalArgs = args.filter(a => !a.startsWith("--"));
-	if (positionalArgs.length < 2) {
-		console.error("Usage: node scripts/push-to-confluence.mjs [--replace-all] {pageId} {file.md}");
-		process.exit(1);
-	}
-	const [pageArg, filePath] = positionalArgs;
-```
-
-> If spec 03 has NOT yet landed, the positional-arg extraction is still the original two-liner:
-> ```js
-> if (args.length < 2) { console.error("Usage: npm run confluence -- {pageId} {file.md}"); process.exit(1); }
-> const [pageArg, filePath] = args;
-> ```
-> In that case Ralph must apply the flags-first refactor from spec 03 step 4 first (or as part of this step).
-
-The `injectContent` call site is approximately line 469:
-
-```js
-	const newBody = injectContent(existingBody, html, title, { replaceAll, pageId, version });
-```
-
-The PUT request follows immediately after this line.
-
-## Target behaviour
-
-### `scripts/push-to-confluence.mjs`
-
-1. `--dry-run` is parsed alongside `--replace-all` in the flags block at the top of `main()`.
-2. After `newBody` is computed by `injectContent`, a dry-run branch is evaluated:
-   - If `dryRun` is `true`: print a header, the full `newBody` HTML, and a footer to stdout; exit 0. The PUT request is never made. No backup is written (even if `--replace-all` is also passed).
-   - If `dryRun` is `false`: proceed as normal (make the PUT request).
-3. The usage error message is updated to include both flags.
-
-### `commands/unic-confluence.md`
-
-Steps 3 and 4 are updated to mention `--dry-run` as an opt-in flag that users can include in `$ARGUMENTS`. The flag is NOT enabled by default in the slash command.
-
-### Edge cases
-
-- `--dry-run --replace-all` together: dry-run takes precedence. `injectContent` still receives `replaceAll = true` and will compute the full-body replacement (or write a backup if no markers). Wait — to avoid side effects during a dry run, `--dry-run` should suppress the backup write too. Pass `replaceAll: dryRun ? false : replaceAll` to `injectContent` so that when dry-running, no backup is created and no markers-not-found error fires from the `--replace-all` path.
-
-  > Ralph: reconsider this. An alternative is to pass `replaceAll` as-is and let the backup write happen — the user asked for `--replace-all` and a backup is a safe side effect. However the simpler rule is: `--dry-run` is purely read-only, so suppress both the PUT and the backup write. Use `effectiveReplaceAll = dryRun ? false : replaceAll` when calling `injectContent`.
-
-- `--dry-run` with missing markers (post-spec-03, no `--replace-all`): `injectContent` still exits 1 with the "No markers found" error because `effectiveReplaceAll` is `false`. The dry-run branch is reached only if `injectContent` returns successfully.
-- `--dry-run` with missing credentials: `loadCredentials()` exits 1 before any HTML is computed.
-- `--dry-run` with a network error on the GET: the catch block in `httpsRequest` rejects and the error propagates normally (exit 1).
-
-## Implementation steps
-
-### Step 1 — Add `--dry-run` to the flags block in `main()`
-
-File: `scripts/push-to-confluence.mjs`, `main()` starting at line 388.
-
-If spec 03 has already landed, the flags block looks like:
-
-```js
-	const replaceAll = args.includes("--replace-all");
-	const positionalArgs = args.filter(a => !a.startsWith("--"));
-	if (positionalArgs.length < 2) {
-		console.error("Usage: node scripts/push-to-confluence.mjs [--replace-all] {pageId} {file.md}");
-		process.exit(1);
-	}
-```
-
-Update to:
-
-```js
-	const dryRun = args.includes("--dry-run");
-	const replaceAll = args.includes("--replace-all");
-	const positionalArgs = args.filter(a => !a.startsWith("--"));
-	if (positionalArgs.length < 2) {
-		console.error("Usage: node scripts/push-to-confluence.mjs [--dry-run] [--replace-all] {pageId} {file.md}");
-		process.exit(1);
-	}
-```
-
-If spec 03 has NOT yet landed, apply the full flags-first refactor first (per spec 03 step 4), then add `dryRun` to the same block.
-
-### Step 2 — Pass `effectiveReplaceAll` to `injectContent`
-
-File: `scripts/push-to-confluence.mjs`, approximately line 469 (the `injectContent` call site).
-
-Before (after spec 03):
-```js
-	const newBody = injectContent(existingBody, html, title, { replaceAll, pageId, version });
-```
-
-After:
-```js
-	const effectiveReplaceAll = dryRun ? false : replaceAll;
-	const newBody = injectContent(existingBody, html, title, { replaceAll: effectiveReplaceAll, pageId, version });
-```
-
-### Step 3 — Insert the dry-run branch after `newBody` is computed
-
-File: `scripts/push-to-confluence.mjs`, immediately after the `injectContent` call and before the PUT request (approximately line 471+).
-
-Insert:
-
-```js
-	if (dryRun) {
-		console.log("=== DRY RUN — Page would be updated to: ===\n");
-		console.log(newBody);
-		console.log("\n=== END DRY RUN ===");
-		process.exit(0);
-	}
-```
-
-The PUT request code that follows is not modified — it is simply never reached when `dryRun` is true.
-
-### Step 4 — Update `commands/unic-confluence.md` step 3
-
-File: `commands/unic-confluence.md`.
-
-Locate step 3 (the publish step). Before (representative text):
-
-```markdown
-### 3. Run the publish script
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" <page-key-or-id> <markdown-file>
-```
-
-After:
-
-```markdown
-### 3. Run the publish script
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" <page-key-or-id> <markdown-file>
-
-Optional flags (pass as part of $ARGUMENTS):
-- `--dry-run` — compute the final HTML and print it to stdout without making the PUT request
-- `--replace-all` — overwrite the full page body (creates a local backup; requires no markers on the page)
-```
-
-### Step 5 — Update `commands/unic-confluence.md` step 4
-
-File: `commands/unic-confluence.md`.
-
-Locate step 4 ("Report result"). Append a note:
-
-```markdown
-### 4. Report result
-If --dry-run was passed, the computed page HTML is printed above. No changes were made to Confluence.
-Otherwise, report success or relay any error message to the user.
-```
-
-## Test cases
-
-| Scenario | Command | Expected stdout (partial) | Expected stderr | Exit code |
-|---|---|---|---|---|
-| `--dry-run` with valid markers | `node scripts/push-to-confluence.mjs --dry-run <id> file.md` | `=== DRY RUN ===` followed by HTML, then `=== END DRY RUN ===` | (none) | 0 |
-| `--dry-run` — no PUT made | Same as above | (check: no Confluence page version bump after the run) | (none) | 0 |
-| `--dry-run --replace-all` | `node scripts/push-to-confluence.mjs --dry-run --replace-all <id> file.md` | Full replacement HTML between DRY RUN markers | (none) | 0 |
-| `--dry-run --replace-all` — no backup | Same as above | No `Backup saved to` line in stdout | (none) | 0 |
-| `--dry-run` with missing markers (post-spec-03) | `node scripts/push-to-confluence.mjs --dry-run <id> file.md` | (none) | `No [AUTO_INSERT_START:label] / [AUTO_INSERT_END:label] markers found…` | 1 |
-| `--dry-run` with no credentials | `node scripts/push-to-confluence.mjs --dry-run <id> file.md` | (none) | `Run \`npm run confluence -- --setup\`…` | 1 |
-| No flags (normal publish) | `node scripts/push-to-confluence.mjs <id> file.md` | Success message | (none) | 0 |
-
-## Acceptance criteria
-
-- `--dry-run` is parsed in `main()` from `process.argv` using `args.includes("--dry-run")`.
-- When `--dry-run` is set, the PUT request is never made (verify by checking the Confluence page version is unchanged after the run).
-- When `--dry-run` is set, no backup file is written even if `--replace-all` is also passed.
-- The dry-run output includes the literal strings `=== DRY RUN — Page would be updated to: ===` and `=== END DRY RUN ===` as delimiters.
-- `--dry-run` with missing markers (no `--replace-all`) still exits 1 with the markers error (spec 03 behaviour is preserved).
-- The usage error message lists `[--dry-run]` and `[--replace-all]`.
-- `commands/unic-confluence.md` mentions `--dry-run` as an opt-in flag in step 3.
-- Lint passes.
-- Slash command `/unic-confluence` still works end-to-end for a normal publish (no `--dry-run`).
-
-## Verification
-
-```sh
-# 1. Confirm --dry-run is parsed
-grep -n "dry-run\|dryRun" scripts/push-to-confluence.mjs
-
-# 2. Confirm dry-run branch exists after injectContent call
-grep -n "DRY RUN" scripts/push-to-confluence.mjs
-
-# 3. Confirm effectiveReplaceAll guards the backup
-grep -n "effectiveReplaceAll" scripts/push-to-confluence.mjs
-
-# 4. Confirm updated usage message
-grep -n "Usage:" scripts/push-to-confluence.mjs
-# Expected: includes [--dry-run] and [--replace-all]
-
-# 5. Confirm commands file mentions --dry-run
-grep "dry-run" commands/unic-confluence.md
-
-# 6. Manual smoke test (requires a mapped page and valid credentials):
-# node scripts/push-to-confluence.mjs --dry-run <page-id> path/to/file.md
-# Expected: HTML output to stdout, exit 0, page version unchanged in Confluence
-```
-
-## Out of scope
-
-- No HTML diff output (e.g. showing what changed between the current body and the new body). A full HTML dump is sufficient for this spec. Diffing would require an additional dependency.
-- No colour-coded terminal output for the dry-run dump.
-- Do not suppress the GET request during dry-run — the GET is needed to compute the final body.
-- No changes to `--verify`, `--setup`, or `--ping`.
-- No version bump (this is a non-breaking additive change).
-- No changes to `confluence-pages.json` parsing or credential loading.
-
-## Follow-ups
-
-- A future spec could pipe `--dry-run` output through a HTML-to-terminal renderer for readability.
-- Consider exposing `--dry-run` as a named option in the slash command arguments documentation so users can pass it naturally as `$ARGUMENTS`.
-- If `injectContent` is ever made async (e.g. for spec 03's backup write), confirm the dry-run branch placement is still correct relative to the `await`.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/05-confluence-code-macro-fidelity.md b/apps/claude-code/unic-confluence/docs/plans/05-confluence-code-macro-fidelity.md
deleted file mode 100644
index a1ba6544..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/05-confluence-code-macro-fidelity.md
+++ /dev/null
@@ -1,301 +0,0 @@
-# 05. Confluence Code Macro Fidelity
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** S
-**Depends on:** none
-**Touches:** `scripts/push-to-confluence.mjs`
-
-## Context
-
-`marked` with default settings converts fenced code blocks to `<pre><code class="language-xxx">…</code></pre>`. When Confluence ingests this HTML via the storage format, it renders the block as monospace plain text — no syntax highlighting, no copy button, no collapse widget. Confluence has its own first-class code block component: the `ac:structured-macro ac:name="code"` element. Any storage-format HTML that uses this macro instead of raw `<pre><code>` gets full syntax highlighting (using Confluence's built-in Prism theme), the copy-to-clipboard button, and the optional title/linenumbers features. This spec adds a `postProcessHtml` function that rewrites `marked`'s `<pre><code>` output into the Confluence macro format immediately after conversion, before the HTML is injected into the page body.
-
-## Current behaviour
-
-**File:** `scripts/push-to-confluence.mjs`, lines 420–423
-
-```js
-const rawContent = readFileSync(resolvedPath, "utf8");
-const stripped = stripFrontmatter(rawContent);
-const html = marked(stripped);                     // line 423 — default options
-```
-
-`marked` emits the following for a fenced JavaScript block:
-
-```html
-<pre><code class="language-javascript">console.log(&quot;hi&quot;);\n</code></pre>
-```
-
-- The `class="language-xxx"` attribute carries the language hint but Confluence's storage-format parser ignores HTML `class` attributes on `<pre>` / `<code>` elements.
-- Characters like `<`, `>`, `&`, `"` inside the fenced block are HTML-entity-encoded by `marked` (`&lt;`, `&gt;`, `&amp;`, `&quot;`). Confluence's CDATA section expects raw text, not HTML entities — if entities are left in place, Confluence will double-escape them and the page will display literal `&amp;lt;` strings.
-- Unlabelled fenced blocks (no language specified) produce `<pre><code>content</code></pre>` with no `class` attribute.
-
-**File:** `package.json`, `dependencies`:
-
-```json
-"marked": "17.0.5"
-```
-
-`marked` v17 ships with GFM mode **on by default** (`marked.defaults.gfm === true`). No explicit option is required.
-
-## Target behaviour
-
-After this spec lands:
-
-1. Every fenced code block — labelled or unlabelled — is rendered as a Confluence `ac:structured-macro` in the storage format.
-2. The language parameter is included only when a language was specified in the fence.
-3. HTML entities inside the code content are decoded back to raw text before being placed inside the CDATA section, so Confluence renders literal `<`, `>`, and `&` characters correctly.
-4. Non-code HTML produced by `marked` (paragraphs, headings, tables, lists, inline code spans) is unchanged.
-
-**Labelled fence** (e.g. ` ```javascript … ``` `):
-
-```xml
-<ac:structured-macro ac:name="code">
-  <ac:parameter ac:name="language">javascript</ac:parameter>
-  <ac:plain-text-body><![CDATA[console.log("hi");
-]]></ac:plain-text-body>
-</ac:structured-macro>
-```
-
-**Unlabelled fence** (` ``` … ``` ` with no language):
-
-```xml
-<ac:structured-macro ac:name="code">
-  <ac:plain-text-body><![CDATA[plain text content
-]]></ac:plain-text-body>
-</ac:structured-macro>
-```
-
-### Edge cases
-
-- A code block whose content contains `]]>` would break the CDATA section. This is vanishingly rare in practice but Ralph should note: Confluence itself escapes `]]>` as `]]]]><![CDATA[>` when round-tripping storage format. For now, the implementation does **not** handle this — add a comment noting it as a known limitation. A follow-up spec can address it if it surfaces in practice.
-- Inline code spans (`<code>` not wrapped in `<pre>`) must **not** be transformed. The regex is anchored to `<pre><code` so inline `<code>` elements are unaffected.
-- Nested backticks inside a fenced block are handled correctly by `marked` before `postProcessHtml` sees the output — no special handling needed here.
-
-## Implementation steps
-
-### Step 1 — Verify GFM default in marked v17
-
-Before adding any `marked.use()` call, verify that GFM is already enabled:
-
-```js
-// Temporary diagnostic — remove before committing
-console.log("marked GFM default:", marked.defaults.gfm);
-```
-
-Run `node -e "const {createRequire}=require('module');const r=createRequire(import.meta.url);const {marked}=r('marked');console.log(marked.defaults.gfm)"` — if it prints `true`, no `marked.use()` call is needed. If it prints `false` or `undefined`, add `marked.use({ gfm: true })` at module initialisation time (after line 25). In practice, `marked` ≥ 1.0 defaults `gfm: true`, so this step is expected to be a no-op.
-
-### Step 2 — Add `postProcessHtml` function
-
-Insert the following function in `scripts/push-to-confluence.mjs` immediately after the `stripFrontmatter` function (after line 117, before the `// ── Content injection strategies` comment block at line 119):
-
-```js
-// ── HTML post-processing ───────────────────────────────────────────────────────
-
-/**
- * Rewrites <pre><code> blocks emitted by `marked` into Confluence storage-format
- * `ac:structured-macro ac:name="code"` elements so that syntax highlighting
- * and the copy-to-clipboard button work in Confluence.
- *
- * Known limitation: code blocks whose content contains the CDATA close sequence
- * `]]>` are not handled — this is exceptionally rare and can be addressed in a
- * follow-up if it surfaces.
- *
- * @param {string} html — HTML string as emitted by `marked`
- * @returns {string} — HTML with <pre><code> blocks replaced by Confluence macros
- */
-function postProcessHtml(html) {
-	function decodeEntities(str) {
-		return str
-			.replace(/&lt;/g, "<")
-			.replace(/&gt;/g, ">")
-			.replace(/&amp;/g, "&")
-			.replace(/&quot;/g, '"')
-			.replace(/&#39;/g, "'");
-	}
-
-	return html.replace(
-		/<pre><code(?:\s+class="language-([^"]+)")?>([\s\S]*?)<\/code><\/pre>/g,
-		(_, lang, code) => {
-			const rawCode = decodeEntities(code);
-			const langParam = lang
-				? `\n  <ac:parameter ac:name="language">${lang}</ac:parameter>`
-				: "";
-			return `<ac:structured-macro ac:name="code">${langParam}\n  <ac:plain-text-body><![CDATA[${rawCode}]]></ac:plain-text-body>\n</ac:structured-macro>`;
-		},
-	);
-}
-```
-
-### Step 3 — Call `postProcessHtml` after `marked`
-
-Change lines 422–423 from:
-
-```js
-const stripped = stripFrontmatter(rawContent);
-const html = marked(stripped);
-```
-
-to:
-
-```js
-const stripped = stripFrontmatter(rawContent);
-const rawHtml = marked(stripped);
-const html = postProcessHtml(rawHtml);
-```
-
-This is a two-line change: rename the `marked()` result to `rawHtml` and add a `postProcessHtml` call. The `html` variable used downstream (line 469 `injectContent`, line 424 empty-check) remains the same name and type.
-
-### Step 4 — Verify the empty-HTML guard still works
-
-Line 424 currently reads:
-
-```js
-if (!html || !html.trim()) {
-```
-
-After step 3, `html` is the post-processed string. A document with only fenced code blocks will produce non-empty HTML (the macro markup), so the guard remains correct. No change needed.
-
-## Test cases
-
-All tests below are manual until spec 09 adds the test harness. The expected output can be verified by pushing to a test Confluence page and inspecting the rendered result.
-
-### TC-01: Labelled code block
-
-**Input markdown:**
-
-````markdown
-```javascript
-const x = 1;
-console.log(x);
-```
-````
-
-**Expected Confluence storage output (in page body):**
-
-```xml
-<ac:structured-macro ac:name="code">
-  <ac:parameter ac:name="language">javascript</ac:parameter>
-  <ac:plain-text-body><![CDATA[const x = 1;
-console.log(x);
-]]></ac:plain-text-body>
-</ac:structured-macro>
-```
-
-**Expected Confluence render:** syntax-highlighted JavaScript block with copy button.
-
-### TC-02: Unlabelled code block
-
-**Input markdown:**
-
-````markdown
-```
-plain text
-no language
-```
-````
-
-**Expected Confluence storage output:**
-
-```xml
-<ac:structured-macro ac:name="code">
-  <ac:plain-text-body><![CDATA[plain text
-no language
-]]></ac:plain-text-body>
-</ac:structured-macro>
-```
-
-No `<ac:parameter ac:name="language">` line should be present.
-
-### TC-03: HTML special characters in code block
-
-**Input markdown:**
-
-````markdown
-```html
-<div class="foo">bar & baz</div>
-```
-````
-
-`marked` emits: `<pre><code class="language-html">&lt;div class=&quot;foo&quot;&gt;bar &amp; baz&lt;/div&gt;\n</code></pre>`
-
-**Expected CDATA content:** `<div class="foo">bar & baz</div>` — raw characters, no entities.
-
-**Verification:** Inspect raw page storage via `GET /wiki/api/v2/pages/{id}?body-format=storage` and confirm the CDATA section contains `<`, `>`, and `&` not `&lt;`, `&gt;`, `&amp;`.
-
-### TC-04: Inline code span — must not be transformed
-
-**Input markdown:**
-
-```markdown
-Use `console.log()` to debug.
-```
-
-**Expected output:** `<p>Use <code>console.log()</code> to debug.</p>` — the inline `<code>` is **not** wrapped in a macro.
-
-### TC-05: Mixed document — code blocks and prose
-
-**Input markdown:**
-
-````markdown
-## Setup
-
-Run the following:
-
-```bash
-npm install
-```
-
-Then edit `config.json`.
-````
-
-**Expected output:** One macro for the bash block; prose paragraphs and the inline `config.json` code span are unchanged.
-
-### TC-06: Exit code on empty file
-
-Push an empty Markdown file → process exits with code 1 and message "Markdown converted to empty HTML — check the source file is not empty". `postProcessHtml("")` returns `""`, the guard on line 424 fires correctly.
-
-## Acceptance criteria
-
-- A fenced code block with a language label renders with syntax highlighting in Confluence after publishing.
-- A fenced code block without a language label renders as a plain code block (no highlighting, no error).
-- HTML special characters (`<`, `>`, `&`, `"`) inside code blocks appear correctly in Confluence — not double-escaped.
-- Inline `<code>` spans (not inside `<pre>`) are unaffected.
-- The rest of the page (headings, paragraphs, tables, lists) is unaffected.
-- `postProcessHtml("")` returns `""` without throwing.
-- `postProcessHtml` is a pure function — no side effects, no I/O.
-
-## Verification
-
-```sh
-# 1. Confirm postProcessHtml is exported (after spec 09 moves it) or inline-testable:
-node -e "
-import('./scripts/push-to-confluence.mjs').catch(() => {});
-" 2>&1 | head -5
-
-# 2. Manual smoke test — push a file with a fenced JS block to a test page:
-node scripts/push-to-confluence.mjs test-page /tmp/test-code.md
-
-# 3. Inspect the resulting page storage:
-# GET https://{base}/wiki/api/v2/pages/{id}?body-format=storage
-# Verify ac:structured-macro is present and CDATA contains raw characters.
-
-# 4. Regression — push a file with only prose (no code blocks):
-node scripts/push-to-confluence.mjs test-page /tmp/test-prose.md
-# Verify the page renders correctly with no macro elements.
-```
-
-## Out of scope
-
-- Do not change table rendering.
-- Do not add a `marked` Renderer extension or custom token for code blocks — the post-process regex approach is simpler and avoids coupling to `marked` internals.
-- Do not upgrade the `marked` version.
-- Do not add `ac:parameter ac:name="title"` or `ac:parameter ac:name="linenumbers"` — those are follow-up enhancements.
-- Do not handle the `]]>` CDATA escape edge case in this spec (noted as a known limitation in the function JSDoc).
-- Do not change how inline code spans (`backtick` in prose) are rendered.
-
-## Follow-ups
-
-- **CDATA `]]>` escape**: If any code blocks in the wild contain `]]>`, the macro will produce malformed XML. Fix: split the CDATA on `]]>` and rejoin with `]]]]><![CDATA[>`. Defer until it surfaces in practice.
-- **Code block title and line numbers**: Confluence supports `<ac:parameter ac:name="title">…</ac:parameter>` and `<ac:parameter ac:name="linenumbers">true</ac:parameter>`. A future spec could parse extended fence info strings (e.g., ` ```js title="server.js" linenumbers `) to populate these.
-- **Spec 09 test coverage**: Once `postProcessHtml` is extracted to `scripts/lib/html.mjs`, add unit tests for TC-01 through TC-06 as automated `node --test` cases.
diff --git a/apps/claude-code/unic-confluence/docs/plans/06-version-sync-script-and-changelog.md b/apps/claude-code/unic-confluence/docs/plans/06-version-sync-script-and-changelog.md
deleted file mode 100644
index 1726eaf6..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/06-version-sync-script-and-changelog.md
+++ /dev/null
@@ -1,344 +0,0 @@
-# 06. Version Sync Script and Changelog
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** S
-**Depends on:** spec 00 (pnpm migration — `pnpm release` command assumes pnpm is the runner)
-**Touches:** `scripts/sync-version.mjs` (new), `package.json`, `.claude-plugin/marketplace.json`, `CHANGELOG.md` (new), `README.md`
-
-## Context
-
-Two files carry the plugin version number: `.claude-plugin/plugin.json` (the authoritative source, read by Claude Code at install time) and `.claude-plugin/marketplace.json` (consumed by the marketplace registry, nested at `plugins[0].version`). Commit `ce6e2ad` ("feat: update version numbers and correct plugin metadata") already shows these drifting — someone bumped one file without the other. Because both files are JSON and both live in the same repo, the only reliable fix is automation: make `plugin.json` the single source of truth and provide a script that reads it and writes the derived version into `marketplace.json`. This spec also introduces a `CHANGELOG.md` in Keep-a-Changelog format and documents the release process in `README.md` so that the version bump workflow is self-contained and does not depend on tribal knowledge.
-
-## Current behaviour
-
-**`.claude-plugin/plugin.json`** (relevant excerpt):
-
-```json
-{
-  "name": "unic-confluence",
-  "version": "1.0.2",
-  ...
-}
-```
-
-**`.claude-plugin/marketplace.json`** (relevant excerpt):
-
-```json
-{
-  "plugins": [
-    {
-      "name": "unic-confluence",
-      "version": "1.0.2",
-      ...
-    }
-  ]
-}
-```
-
-**`package.json`** `scripts` field: does not exist. The file has no `"scripts"` key at all — there are only `"name"`, `"private"`, `"license"`, `"type"`, and `"dependencies"` fields.
-
-There is no `CHANGELOG.md`. Version history is reconstructed from git log only.
-
-The `README.md` has no "Releasing" section.
-
-There is no automated check to prevent the two version fields from diverging.
-
-## Target behaviour
-
-After this spec lands:
-
-1. `scripts/sync-version.mjs` reads `.claude-plugin/plugin.json`, extracts `.version`, and writes it back into `.claude-plugin/marketplace.json` at `plugins[0].version`. The file is written with 2-space indent and a trailing newline. The script exits 0 on success and 1 with a descriptive message if either file is missing or malformed.
-2. `package.json` gains a `"scripts"` block with at minimum `"sync-version"` and `"release"` entries.
-3. `CHANGELOG.md` exists at the repo root in Keep-a-Changelog / SemVer format and documents all versions from 1.0.0 through the current 1.0.2.
-4. `README.md` gains a short "Releasing" section (3–5 lines) explaining the manual steps.
-
-A developer releasing version 1.1.0 does the following — and nothing else:
-
-```sh
-# 1. Bump the version in the single source of truth:
-#    edit .claude-plugin/plugin.json  →  "version": "1.1.0"
-
-# 2. Update the changelog:
-#    move [Unreleased] items to ## [1.1.0] — YYYY-MM-DD
-
-# 3. Sync and commit:
-pnpm release
-```
-
-## Implementation steps
-
-### Step 1 — Create `scripts/sync-version.mjs`
-
-Create the file at `scripts/sync-version.mjs` with the following content:
-
-```js
-#!/usr/bin/env node
-// SPDX-License-Identifier: LGPL-3.0-or-later
-import { readFileSync, writeFileSync } from "fs";
-import path from "path";
-import { fileURLToPath } from "url";
-
-const root = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..");
-
-// Read the single source of truth
-const pluginPath = path.join(root, ".claude-plugin/plugin.json");
-let pluginJson;
-try {
-	pluginJson = JSON.parse(readFileSync(pluginPath, "utf8"));
-} catch (err) {
-	console.error(`sync-version: cannot read ${pluginPath}: ${err.message}`);
-	process.exit(1);
-}
-
-const version = pluginJson.version;
-if (!version || typeof version !== "string") {
-	console.error(`sync-version: .version is missing or not a string in ${pluginPath}`);
-	process.exit(1);
-}
-
-// Write derived version into marketplace.json
-const marketplacePath = path.join(root, ".claude-plugin/marketplace.json");
-let marketplace;
-try {
-	marketplace = JSON.parse(readFileSync(marketplacePath, "utf8"));
-} catch (err) {
-	console.error(`sync-version: cannot read ${marketplacePath}: ${err.message}`);
-	process.exit(1);
-}
-
-if (!Array.isArray(marketplace.plugins) || marketplace.plugins.length === 0) {
-	console.error(`sync-version: marketplace.json has no plugins[] array`);
-	process.exit(1);
-}
-
-const prev = marketplace.plugins[0].version;
-marketplace.plugins[0].version = version;
-writeFileSync(marketplacePath, JSON.stringify(marketplace, null, 2) + "\n", "utf8");
-
-if (prev === version) {
-	console.log(`sync-version: marketplace.json already at version ${version} (no change)`);
-} else {
-	console.log(`sync-version: marketplace.json updated ${prev} → ${version}`);
-}
-```
-
-Key design decisions:
-- Uses `fileURLToPath` + `path.dirname` to locate the repo root relative to the script — works regardless of the cwd when the script is invoked.
-- Writes a trailing newline after the JSON to satisfy POSIX text-file conventions and avoid noisy git diffs.
-- Logs whether an actual change was made — useful in CI to detect accidental double-runs.
-- All error paths exit 1 with a human-readable message.
-
-### Step 2 — Add `scripts` to `package.json`
-
-The current `package.json` has no `"scripts"` field. Add it. The full updated `package.json`:
-
-**Before (full file):**
-
-```json
-{
-  "name": "unic-confluence",
-  "private": true,
-  "license": "LGPL-3.0-or-later",
-  "type": "module",
-  "dependencies": {
-    "marked": "17.0.5"
-  }
-}
-```
-
-**After:**
-
-```json
-{
-  "name": "unic-confluence",
-  "private": true,
-  "license": "LGPL-3.0-or-later",
-  "type": "module",
-  "scripts": {
-    "confluence": "node scripts/push-to-confluence.mjs",
-    "sync-version": "node scripts/sync-version.mjs",
-    "release": "node scripts/sync-version.mjs && git add .claude-plugin/marketplace.json CHANGELOG.md && git commit -m \"chore: release v$(node -e \"import('./.claude-plugin/plugin.json',{assert:{type:'json'}}).then(m=>process.stdout.write(m.default.version))\")\""
-  },
-  "dependencies": {
-    "marked": "17.0.5"
-  }
-}
-```
-
-**Note for Ralph on the `release` script**: The inline `node -e` uses dynamic `import()` with a JSON assertion to read the version — this works in Node 18+ ESM context without requiring `createRequire`. However, the `assert:` JSON import assertion syntax was deprecated in Node 22 in favour of `with:`. Test the exact `node -e` invocation against the Node version in use:
-
-```sh
-# Node 18-21:
-node -e "import('./.claude-plugin/plugin.json', {assert:{type:'json'}}).then(m=>process.stdout.write(m.default.version))"
-
-# Node 22+:
-node -e "import('./.claude-plugin/plugin.json', {with:{type:'json'}}).then(m=>process.stdout.write(m.default.version))"
-```
-
-If neither works cleanly in a pnpm script context (shell escaping of nested quotes is fragile), use this simpler alternative that avoids import assertions entirely:
-
-```sh
-node -e "const fs=require('fs');const p=JSON.parse(fs.readFileSync('.claude-plugin/plugin.json','utf8'));process.stdout.write(p.version)"
-```
-
-This uses CommonJS `require` which is available in `node -e` even when `package.json` has `"type":"module"` (the `-e` flag does not use the package type). Confirm which form works before committing.
-
-The `confluence` script entry is added here so that `pnpm confluence -- pageId file.md` works — this is the primary user-facing command and was previously undocumented in `package.json`.
-
-### Step 3 — Create `CHANGELOG.md`
-
-Create `CHANGELOG.md` at the repo root:
-
-```markdown
-# Changelog
-
-All notable changes to this project will be documented in this file.
-
-The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
-and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
-
-## [Unreleased]
-
-### Breaking
-- (none)
-
-### Added
-- (none)
-
-### Fixed
-- (none)
-
-## [1.0.2] — 2026-04-23
-
-### Fixed
-- Add `package-lock.json` for dependency management
-- Correct plugin metadata in `marketplace.json`
-- Add pnpm commands to README
-
-## [1.0.1] — 2026-04-22
-
-### Fixed
-- Correct install instructions in README
-
-## [1.0.0] — 2026-04-22
-
-### Added
-- Initial release: `/unic-confluence` slash command for publishing Markdown to Confluence via v2 API
-- Interactive `--setup` flow for credential configuration (`~/.unic-confluence.json`)
-- `--verify` subcommand to health-check all pages in `confluence-pages.json`
-- Plain-text marker injection strategy (`[AUTO_INSERT_START: label]` / `[AUTO_INSERT_END: label]`)
-- Anchor-macro fallback injection strategy (`md-start` / `md-end` Confluence anchors)
-- Append-to-page fallback when no markers are present
-- YAML frontmatter stripping
-- 5 MB file size guard
-- LGPL-3.0-or-later licence
-```
-
-**Note for Ralph**: The `[1.0.1]` entry is inferred from commit `53cddac` ("fix: add marketplace.json and correct install instructions"). If the git log shows no 1.0.1 tag and 1.0.0 went straight to 1.0.2, collapse the two Fixed entries into the `[1.0.2]` section and remove the `[1.0.1]` entry. Check with `git tag -l` before writing the changelog.
-
-### Step 4 — Add "Releasing" section to `README.md`
-
-Open `README.md` and add the following section. Insert it after the existing "Installation" / "Usage" sections, before "Contributing" or "Licence" if those exist, or at the end of the file if they do not.
-
-```markdown
-## Releasing
-
-1. Edit `"version"` in `.claude-plugin/plugin.json` — this is the single source of truth.
-2. Update `CHANGELOG.md`: move items from `## [Unreleased]` to a new `## [X.Y.Z] — YYYY-MM-DD` section.
-3. Run `pnpm release` — this syncs `marketplace.json`, stages both files, and commits with a `chore: release vX.Y.Z` message.
-4. Push and create a GitHub release tag: `git tag vX.Y.Z && git push origin vX.Y.Z`.
-```
-
-## Test cases
-
-### TC-01: Sync with matching versions (no-op)
-
-```sh
-# Ensure both files are at the same version, then run:
-node scripts/sync-version.mjs
-# Expected stdout: "sync-version: marketplace.json already at version 1.0.2 (no change)"
-# Expected exit code: 0
-# marketplace.json: unchanged
-```
-
-### TC-02: Sync after bumping plugin.json
-
-```sh
-# Manually set plugin.json version to "1.1.0", leave marketplace.json at "1.0.2"
-node scripts/sync-version.mjs
-# Expected stdout: "sync-version: marketplace.json updated 1.0.2 → 1.1.0"
-# Expected exit code: 0
-# marketplace.json plugins[0].version: "1.1.0"
-# Restore plugin.json to "1.0.2" after the test.
-```
-
-### TC-03: Missing plugin.json
-
-```sh
-mv .claude-plugin/plugin.json /tmp/plugin.json.bak
-node scripts/sync-version.mjs
-# Expected: exits 1, message contains "cannot read" and the file path
-mv /tmp/plugin.json.bak .claude-plugin/plugin.json
-```
-
-### TC-04: Malformed marketplace.json
-
-```sh
-echo '{"plugins":[]}' > /tmp/mktplace_test.json
-# Temporarily point the script at a marketplace with empty plugins array
-# (modify the script path in a test copy, or test the guard condition by inspection)
-# Expected: exits 1, message "marketplace.json has no plugins[] array"
-```
-
-### TC-05: Release script produces correct commit message
-
-```sh
-pnpm release
-# Expected: git log --oneline -1 shows "chore: release v1.0.2" (or current version)
-# Expected: staged files include .claude-plugin/marketplace.json and CHANGELOG.md
-```
-
-## Acceptance criteria
-
-- `node scripts/sync-version.mjs` exits 0 and writes the correct version to `marketplace.json` when `plugin.json` has a valid version.
-- `pnpm sync-version` is equivalent to the above.
-- The script exits 1 with a non-empty error message for all error conditions (missing file, malformed JSON, missing `plugins[]`).
-- `marketplace.json` always has a trailing newline after the script runs (no git diff noise).
-- `CHANGELOG.md` exists and is parseable as Keep-a-Changelog Markdown.
-- The `[Unreleased]` section has the three subsections: Breaking, Added, Fixed.
-- `README.md` contains a "Releasing" section with `pnpm release` in the instructions.
-- The `release` script produces a commit with message `chore: release vX.Y.Z`.
-
-## Verification
-
-```sh
-# Sync script runs cleanly:
-node scripts/sync-version.mjs
-
-# pnpm script alias works:
-pnpm sync-version
-
-# Confirm marketplace.json trailing newline (xxd last 3 bytes should be 0a):
-xxd .claude-plugin/marketplace.json | tail -1
-
-# Confirm CHANGELOG.md has [Unreleased] section:
-grep '## \[Unreleased\]' CHANGELOG.md
-
-# Confirm README.md has Releasing section:
-grep '## Releasing' README.md
-```
-
-## Out of scope
-
-- Do not actually bump the version as part of this spec — the current version (1.0.2) stays as-is.
-- Do not add Changesets, release-please, or semantic-release automation.
-- Do not add a git tag step to the `release` script — the tag is a manual `git tag vX.Y.Z && git push origin vX.Y.Z` documented in the README.
-- Do not add CI enforcement (e.g., a GitHub Actions job that fails if the versions are out of sync) — that is a follow-up.
-- Do not modify `.claude-plugin/plugin.json` structure or add new fields to it.
-
-## Follow-ups
-
-- **CI version drift check**: Add a GitHub Actions step that runs `node scripts/sync-version.mjs` in `--dry-run` mode (or a simple `diff` check) and fails the build if the versions are out of sync.
-- **Git tag automation**: Extend `pnpm release` to also run `git tag v$(version) && git push origin v$(version)` — deferred because tag + push should be an explicit, reviewed action.
-- **Version in README badge**: A `[![version](…)](…)` badge sourced from `marketplace.json` or a GitHub release would surface the current version without having to open the JSON files.
diff --git a/apps/claude-code/unic-confluence/docs/plans/07-http-timeout-and-check-auth.md b/apps/claude-code/unic-confluence/docs/plans/07-http-timeout-and-check-auth.md
deleted file mode 100644
index b45b4cdc..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/07-http-timeout-and-check-auth.md
+++ /dev/null
@@ -1,365 +0,0 @@
-# 07. HTTP Timeout and `--check-auth` Subcommand
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** S
-**Depends on:** spec 01 (which introduced `--ping`; this spec formalises it as `--check-auth`)
-**Touches:** `scripts/push-to-confluence.mjs`, `commands/unic-confluence.md`
-
-## Context
-
-`httpsRequest` (lines 64–95) creates an HTTPS request but never sets a timeout. If the connection hangs — due to a VPN drop mid-flight, a firewall that silently drops packets without sending TCP RST, or a Confluence server that accepts the connection but never sends response bytes — the process waits indefinitely. There is no way to interrupt it short of `Ctrl+C`. For a CLI tool used inside Claude Code slash commands (which have their own timeout budget), a hung sub-process is particularly damaging because it holds the Claude context open without making progress.
-
-The fix is a 30-second socket timeout using Node's `req.setTimeout()`, which emits a `timeout` event and allows the code to call `req.destroy(err)` to surface a clean error message. The existing `req.on("error", reject)` handler already catches the resulting error, so no new control flow is needed.
-
-Spec 01 introduced a lightweight `--ping` branch for cheap credential probing. This spec formalises that as `--check-auth`, which is the name used in the updated slash command documentation. If spec 01 landed with `--ping`, this spec renames it. If spec 01 has not landed, this spec adds `--check-auth` from scratch.
-
-## Current behaviour
-
-### `httpsRequest` — no timeout (lines 64–95)
-
-```js
-function httpsRequest(method, urlStr, authHeader, bodyObj) {
-	return new Promise((resolve, reject) => {
-		const parsed = new URL(urlStr);
-		const bodyStr = bodyObj ? JSON.stringify(bodyObj) : null;
-		const options = {
-			method,
-			hostname: parsed.hostname,
-			path: parsed.pathname + parsed.search,
-			headers: {
-				Authorization: authHeader,
-				"Content-Type": "application/json",
-				Accept: "application/json",
-				...(bodyStr ? { "Content-Length": Buffer.byteLength(bodyStr) } : {}),
-			},
-		};
-
-		const req = https.request(options, (res) => {
-			let data = "";
-			res.on("data", (chunk) => (data += chunk));
-			res.on("end", () => resolve({ status: res.statusCode, body: data }));
-		});
-
-		req.on("error", () => {
-			reject(
-				new Error("Cannot reach Confluence — check VPN/network connectivity"),
-			);
-		});
-
-		if (bodyStr) req.write(bodyStr);
-		req.end();
-	});
-}
-```
-
-There is no `req.setTimeout()` call. If the server never responds, the Promise never settles and the process hangs forever.
-
-### `main()` — no `--check-auth` (lines 388–398)
-
-```js
-async function main() {
-	const args = process.argv.slice(2);
-
-	if (args[0] === "--setup") { await runSetup(); process.exit(0); }
-	if (args[0] === "--verify") { await runVerify(); return; }
-	// ... no --ping or --check-auth branch
-```
-
-### `commands/unic-confluence.md` — step 2 uses `--verify`
-
-```markdown
-### 2. Check credentials
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --verify
-
-If credentials missing, instruct user to run --setup.
-```
-
-(This was the target for spec 01; check whether spec 01 has landed and already replaced `--verify` with `--ping`.)
-
-## Target behaviour
-
-### `httpsRequest` — 30-second timeout
-
-Every call to `httpsRequest` times out after 30 seconds. On timeout:
-
-- The error message is: `"Request timed out after 30s — check VPN/network connectivity"`.
-- The process exits with code 1 (through the existing `catch (err) { console.error(err.message); process.exit(1); }` pattern used at every call site).
-- The underlying socket is destroyed, releasing file descriptors.
-
-### `--check-auth` subcommand
-
-Calling `node scripts/push-to-confluence.mjs --check-auth`:
-
-- Loads credentials via `loadCredentials()`. If credentials are not configured, exits 1 with "Run `npm run confluence -- --setup` to configure credentials" (the message from `loadCredentials`).
-- Makes a single GET to `{baseUrl}/wiki/api/v2/pages?limit=1`.
-- If the response status is 2xx: prints "✓ Credentials valid" to stdout and exits 0.
-- If the response status is 4xx/5xx: calls `handleHttpError(status, "")` and exits 1.
-- If the request throws (network unreachable, timeout): prints `err.message` to stderr and exits 1.
-
-### `commands/unic-confluence.md` — step 2 uses `--check-auth`
-
-```markdown
-### 2. Check credentials
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --check-auth
-
-If this fails with a credential error, instruct the user to run --setup first.
-```
-
-## Implementation steps
-
-### Step 1 — Add timeout to `httpsRequest`
-
-**File:** `scripts/push-to-confluence.mjs`
-
-**Location:** Inside `httpsRequest`, after the `const req = https.request(...)` call (line 80), before the `req.on("error", ...)` call (line 86).
-
-**Before** (lines 80–94):
-
-```js
-		const req = https.request(options, (res) => {
-			let data = "";
-			res.on("data", (chunk) => (data += chunk));
-			res.on("end", () => resolve({ status: res.statusCode, body: data }));
-		});
-
-		req.on("error", () => {
-			reject(
-				new Error("Cannot reach Confluence — check VPN/network connectivity"),
-			);
-		});
-
-		if (bodyStr) req.write(bodyStr);
-		req.end();
-```
-
-**After:**
-
-```js
-		const req = https.request(options, (res) => {
-			let data = "";
-			res.on("data", (chunk) => (data += chunk));
-			res.on("end", () => resolve({ status: res.statusCode, body: data }));
-		});
-
-		req.setTimeout(30_000, () => {
-			req.destroy(
-				new Error("Request timed out after 30s — check VPN/network connectivity"),
-			);
-		});
-
-		req.on("error", (err) => {
-			reject(err);
-		});
-
-		if (bodyStr) req.write(bodyStr);
-		req.end();
-```
-
-Two changes in this diff:
-
-1. Add `req.setTimeout(30_000, callback)` — the callback calls `req.destroy(err)`. `req.destroy(err)` emits the `"error"` event with the provided error, which causes the existing `req.on("error", ...)` handler to fire and `reject` the promise with that error.
-
-2. Change `req.on("error", () => { reject(new Error("Cannot reach...")) })` to `req.on("error", (err) => { reject(err) })` — the error is now forwarded directly rather than replaced with a generic message. This preserves the timeout message ("timed out after 30s") as well as any other OS-level errors (ECONNREFUSED, ENOTFOUND) verbatim. The old generic message masked useful diagnostic information.
-
-**Why `req.destroy(new Error(...))` triggers `req.on("error")`**: Node.js `ClientRequest.destroy(err)` emits the `"error"` event with the provided error object. This is documented behaviour in Node.js `net.Socket` (which `ClientRequest` inherits from). Confirmed in Node.js docs: "If `error` is provided, an 'error' event will be emitted and `error` will be passed as an argument to any listeners on that event." Do not add a separate `req.on("timeout", ...)` listener — the `req.setTimeout()` callback is cleaner for this pattern.
-
-### Step 2 — Add `--check-auth` branch to `main()`
-
-**File:** `scripts/push-to-confluence.mjs`
-
-**Location:** In `main()`, after the `--verify` branch (or after the `--ping` branch if spec 01 landed).
-
-**Before** (lines 391–398):
-
-```js
-	if (args[0] === "--setup") { await runSetup(); process.exit(0); }
-	if (args[0] === "--verify") { await runVerify(); return; }
-
-	if (args.length < 2) {
-```
-
-**After** (if spec 01 has NOT landed — add both `--ping` removal and `--check-auth`):
-
-```js
-	if (args[0] === "--setup") { await runSetup(); process.exit(0); }
-	if (args[0] === "--verify") { await runVerify(); return; }
-	if (args[0] === "--check-auth") {
-		const { url: baseUrl, username, token } = loadCredentials();
-		const authHeader = makeBasicAuth(username, token);
-		let res;
-		try {
-			res = await httpsRequest(
-				"GET",
-				`${baseUrl.replace(/\/$/, "")}/wiki/api/v2/pages?limit=1`,
-				authHeader,
-			);
-		} catch (err) {
-			console.error(err.message);
-			process.exit(1);
-		}
-		if (res.status >= 200 && res.status < 300) {
-			console.log("✓ Credentials valid");
-			process.exit(0);
-		}
-		handleHttpError(res.status, "");
-	}
-
-	if (args.length < 2) {
-```
-
-**After** (if spec 01 has ALREADY landed with `--ping` — rename it):
-
-Replace the `--ping` branch with the `--check-auth` block shown above. The implementation is identical; only the argument name changes.
-
-**Note for Ralph**: Check whether spec 01 landed. Look for an `args[0] === "--ping"` block. If present, rename `"--ping"` to `"--check-auth"` in the condition. If absent, add the full block shown above.
-
-### Step 3 — Update `commands/unic-confluence.md`
-
-**File:** `commands/unic-confluence.md`
-
-Find the line that reads (one of the following, depending on what spec 01 did):
-
-```
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --verify
-```
-
-or:
-
-```
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --ping
-```
-
-Replace it with:
-
-```
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --check-auth
-```
-
-Also update any surrounding text that mentions `--verify` or `--ping` in the context of the credential check step. The `--verify` subcommand is still valid for its original purpose (full page enumeration health check) — do not remove it from the script, only from the slash command's default pre-flight path.
-
-## Test cases
-
-### TC-01: Timeout fires after 30 seconds (manual / observational)
-
-Simulate a hung connection by blocking the Confluence hostname at the firewall level or by connecting to a known non-responsive host, then running:
-
-```sh
-time node scripts/push-to-confluence.mjs my-page /tmp/test.md
-```
-
-Expected: process exits after ~30 seconds with exit code 1 and stderr message "Request timed out after 30s — check VPN/network connectivity".
-
-A lighter test: use Node's `--inspect` to pause the response callback, or use a local netcat listener that accepts the TCP connection but never sends a response:
-
-```sh
-# Terminal 1: listen but never respond
-nc -l 8443
-
-# Terminal 2: (temporarily edit credentials to point at localhost:8443)
-CONFLUENCE_URL=https://localhost:8443 CONFLUENCE_USER=u CONFLUENCE_TOKEN=t \
-  node scripts/push-to-confluence.mjs --check-auth
-```
-
-Expected: after 30s, exits 1 with the timeout message.
-
-### TC-02: `--check-auth` with valid credentials
-
-```sh
-node scripts/push-to-confluence.mjs --check-auth
-# Expected exit code: 0
-# Expected stdout: "✓ Credentials valid"
-```
-
-Requires valid `~/.unic-confluence.json` or `CONFLUENCE_URL` / `CONFLUENCE_USER` / `CONFLUENCE_TOKEN` env vars.
-
-### TC-03: `--check-auth` with wrong API token
-
-Set `CONFLUENCE_TOKEN` to an invalid value:
-
-```sh
-CONFLUENCE_TOKEN=wrong node scripts/push-to-confluence.mjs --check-auth
-# Expected exit code: 1
-# Expected stderr: "API token rejected — generate a new one at ..."
-```
-
-### TC-04: `--check-auth` with no credentials configured
-
-Remove `~/.unic-confluence.json` and unset env vars:
-
-```sh
-node scripts/push-to-confluence.mjs --check-auth
-# Expected exit code: 1
-# Expected stderr: "Run `npm run confluence -- --setup` to configure credentials"
-```
-
-### TC-05: `--check-auth` with network unreachable
-
-Disconnect from VPN / network so that DNS resolution fails:
-
-```sh
-node scripts/push-to-confluence.mjs --check-auth
-# Expected exit code: 1
-# Expected stderr: error message from the OS (ENOTFOUND or similar)
-# Expected: exits within 30 seconds (not hung)
-```
-
-Note: DNS failures typically return immediately (ENOTFOUND is synchronous with OS DNS). The 30-second timeout covers TCP-level hangs, not DNS failures.
-
-### TC-06: Regular publish still works (regression)
-
-```sh
-node scripts/push-to-confluence.mjs my-page /tmp/test.md
-# Expected: publishes normally, exits 0, no timeout side effects
-```
-
-The timeout is transparent for connections that complete within 30 seconds.
-
-## Acceptance criteria
-
-- `httpsRequest` no longer hangs indefinitely — it rejects after 30 seconds with a descriptive error.
-- The timeout error message contains "30s" and "VPN/network".
-- The underlying socket is destroyed on timeout (no open file descriptors after the process exits).
-- `--check-auth` exits 0 with "✓ Credentials valid" when credentials are correct.
-- `--check-auth` exits 1 with the appropriate `handleHttpError` message for 401, 403, 404, 409.
-- `--check-auth` exits 1 with the `loadCredentials` error message when no credentials are configured.
-- `commands/unic-confluence.md` uses `--check-auth` (not `--verify` or `--ping`) for the pre-flight credential probe.
-- All existing functionality (publish, `--setup`, `--verify`) is unaffected by the timeout addition.
-
-## Verification
-
-```sh
-# 1. Confirm req.setTimeout is present in the source:
-grep -n "setTimeout" scripts/push-to-confluence.mjs
-# Expected: one line containing "req.setTimeout(30_000,"
-
-# 2. Confirm --check-auth branch exists:
-grep -n "check-auth" scripts/push-to-confluence.mjs
-# Expected: one line with args[0] === "--check-auth"
-
-# 3. Confirm commands file references --check-auth:
-grep -n "check-auth" commands/unic-confluence.md
-# Expected: at least one hit
-
-# 4. Confirm --ping or --verify is not used in the commands file for the credential step:
-grep -n "\-\-ping\|\-\-verify" commands/unic-confluence.md
-# Expected: no hits (or any hits are in comments/explanatory text, not as the invoked command)
-
-# 5. Live check-auth test (requires valid credentials):
-node scripts/push-to-confluence.mjs --check-auth
-# Expected: "✓ Credentials valid" and exit 0
-```
-
-## Out of scope
-
-- Do not change the 30-second timeout value — it is the conventional network operation timeout for interactive CLI tools. If the value needs to be configurable, that is a follow-up.
-- Do not add retry logic — retrying automatically on timeout could exacerbate the problem (e.g., a server that is overloaded). Retrying is a user decision.
-- Do not add a `--timeout` flag to override the default.
-- Do not add response-streaming timeout (time between chunks) — only the initial connection / response-start timeout is addressed here.
-- Do not change `runVerify()` — it keeps the sequential-GET behaviour and is still accessible via `--verify` for users who want the full health check.
-
-## Follow-ups
-
-- **Response body timeout**: Currently the 30-second timeout covers the time until the first response byte. A very large page body could then hang indefinitely while streaming. A `res.setTimeout()` on the response stream would address this — deferred.
-- **Configurable timeout**: If users on particularly slow VPNs report false positives, add a `--timeout <ms>` flag or respect a `CONFLUENCE_TIMEOUT_MS` env var.
-- **`--check-auth` in `runSetup()`**: The setup flow currently does its own inline credential validation GET (lines 368–383). After this spec lands, consider refactoring `runSetup()` to call the same code path as `--check-auth`. Deferred to avoid scope creep.
diff --git a/apps/claude-code/unic-confluence/docs/plans/08-tighten-strip-frontmatter.md b/apps/claude-code/unic-confluence/docs/plans/08-tighten-strip-frontmatter.md
deleted file mode 100644
index 686270ae..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/08-tighten-strip-frontmatter.md
+++ /dev/null
@@ -1,322 +0,0 @@
-# 08. Tighten `stripFrontmatter` Regex
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** S
-**Depends on:** none
-**Touches:** `scripts/push-to-confluence.mjs`
-
-## Context
-
-`stripFrontmatter` uses the `s` ("dotAll") flag, which makes `.` match newline characters. Combined with lazy `.*?`, the regex tries to find the shortest match for the content between the two `---` delimiters. In well-formed frontmatter this is correct: the regex matches from the opening `---` to the first subsequent `---` line and strips only the block. However, if a document's YAML frontmatter is never closed — the author forgot to write the closing `---` — the `.*?` with `s` will scan forward through the entire document until it finds the last `---` anywhere in the body, potentially matching and silently deleting real content. The `s` flag removes the natural line-boundary protection that makes frontmatter parsing safe. Capping the frontmatter block to a maximum of 50 lines provides defence-in-depth without breaking any legitimate use case (YAML front matter is never 50 key–value pairs in practice).
-
-## Current behaviour
-
-**File:** `scripts/push-to-confluence.mjs`, lines 113–117
-
-```js
-function stripFrontmatter(content) {
-	// The opening --- must be at byte 0 to avoid treating a CommonMark HR in the body as frontmatter
-	// \r?\n handles both LF and CRLF line endings
-	return content.replace(/^---\r?\n.*?\r?\n---\s*\r?\n/s, "");
-}
-```
-
-The regex: `/^---\r?\n.*?\r?\n---\s*\r?\n/s`
-
-Breakdown:
-- `^` — anchors to start of string (`String.prototype.replace` with `/^/` without `m` flag matches byte 0 only — correct).
-- `---\r?\n` — matches opening `---` line.
-- `.*?` — matches any characters (including newlines, because of `s` flag) lazily.
-- `\r?\n---\s*\r?\n` — matches the closing `---` line.
-
-**The failure scenario:**
-
-Given this input:
-
-```
----
-title: My Document
-(author forgot to add closing ---)
-
-# Introduction
-
-Some text with a horizontal rule below:
-
----
-
-## Second Section
-```
-
-The regex will match from byte 0 to the `---` line above "## Second Section", consuming the entire "Introduction" section. The result is:
-
-```
-
-## Second Section
-```
-
-The "Introduction" section and everything up to the last `---` is silently deleted. No error is raised. The user pushes a truncated document to Confluence without realising it.
-
-**The `.*?` laziness does not fully protect against this**: laziness means the engine prefers shorter matches, but it must still find a match. If the only `---` after the opening is in the document body, the engine will match to that `---` even though it is far from the opening. There is no maximum match length.
-
-**Note on the "common" case**: A document whose body contains `---` used as a CommonMark horizontal rule (e.g., `\n---\n` between two paragraphs, not at the very beginning of a line after a blank line) is handled differently by different Markdown parsers, but the regex does not care about Markdown semantics — it just scans for the next `\r?\n---\s*\r?\n` pattern anywhere in the string.
-
-## Target behaviour
-
-After this spec, `stripFrontmatter` caps the frontmatter block to at most 50 lines using `(?:[^\n]*\r?\n){0,50}?` instead of `.*?` with the `s` flag.
-
-**New function:**
-
-```js
-function stripFrontmatter(content) {
-	// Match only if opening --- is at byte 0.
-	// Allow up to 50 key:value lines inside the block — legitimate YAML frontmatter
-	// is never 50 lines; this cap prevents runaway matches on unclosed frontmatter.
-	// [^\n]*\r?\n matches one line (LF or CRLF) without crossing to the next.
-	// The s flag is intentionally absent: [^\n] provides the same line-at-a-time
-	// semantics without granting .* permission to range across the whole document.
-	return content.replace(
-		/^---\r?\n(?:[^\n]*\r?\n){0,50}?---\s*\r?\n/,
-		"",
-	);
-}
-```
-
-### Behaviour for all input categories
-
-| Input | Old behaviour | New behaviour |
-|-------|--------------|---------------|
-| No frontmatter | Returns content unchanged | Returns content unchanged (no match) |
-| Valid 2-field frontmatter | Strips correctly | Strips correctly |
-| Valid 40-field frontmatter | Strips correctly | Strips correctly (within 50-line cap) |
-| Frontmatter with `---` HR in body | May consume body up to last `---` | Strips only the frontmatter block |
-| Unclosed frontmatter (no closing `---` within 50 lines) | Matches to last `---` in document | No match — returns content unchanged |
-| Unclosed frontmatter with a `---` at line 55+ | Matches to that `---` | No match — cap of 50 lines prevents reaching it |
-| CRLF frontmatter | Strips correctly | Strips correctly |
-| Frontmatter where closing `---` is on line 51 | Strips correctly (no cap) | No match — returns content unchanged |
-
-The 51-line case is an acceptable trade-off: a YAML front matter with 51 fields is malformed by any reasonable standard, and returning the content unchanged (rather than stripping a wrong block) is the safer failure mode.
-
-### Regex mechanics
-
-`/^---\r?\n(?:[^\n]*\r?\n){0,50}?---\s*\r?\n/`
-
-- `^` — start of string (no `m` or `s` flags, so `^` is byte 0).
-- `---\r?\n` — opening delimiter, LF or CRLF.
-- `(?:[^\n]*\r?\n){0,50}?` — match 0 to 50 lines, lazily. Each iteration consumes `[^\n]*` (any chars except LF) followed by `\r?\n` (the line terminator). The `?` after `{0,50}` makes the quantifier prefer fewer iterations — the engine tries to find a match with the fewest lines first, stopping as soon as the next token (`---`) can match.
-- `---\s*\r?\n` — closing delimiter line.
-- No `s` flag — `[^\n]` does not match `\n`, so each `[^\n]*` is bounded to a single line by construction.
-
-**Why `[^\n]*` instead of `.` with `s`**: `[^\n]*` explicitly excludes the newline character. The `\r?\n` at the end of each non-capturing group consumes the line terminator. This means the alternation `[^\n]*\r?\n` matches exactly one line and cannot span multiple lines, regardless of flags.
-
-**Why `{0,50}?` is lazy**: Without the `?`, `{0,50}` would greedily consume up to 50 lines before checking for `---`. With `?`, the engine tries 0 lines first (i.e., checks if the very next token is `---`), then 1 line, etc. For a 2-line frontmatter block this means 2 iterations before the engine finds the closing `---`, which is correct and efficient.
-
-## Implementation steps
-
-### Step 1 — Replace the function body
-
-**File:** `scripts/push-to-confluence.mjs`
-
-**Before** (lines 113–117):
-
-```js
-function stripFrontmatter(content) {
-	// The opening --- must be at byte 0 to avoid treating a CommonMark HR in the body as frontmatter
-	// \r?\n handles both LF and CRLF line endings
-	return content.replace(/^---\r?\n.*?\r?\n---\s*\r?\n/s, "");
-}
-```
-
-**After:**
-
-```js
-function stripFrontmatter(content) {
-	// Match only if opening --- is at byte 0.
-	// Allow up to 50 key:value lines inside the block — legitimate YAML frontmatter
-	// is never 50 lines; this cap prevents runaway matches on unclosed frontmatter.
-	// [^\n]*\r?\n matches one line (LF or CRLF) without crossing to the next.
-	// The s flag is intentionally absent: [^\n] provides the same line-at-a-time
-	// semantics without granting .* permission to range across the whole document.
-	return content.replace(
-		/^---\r?\n(?:[^\n]*\r?\n){0,50}?---\s*\r?\n/,
-		"",
-	);
-}
-```
-
-This is a single-function change with no other modifications required in this spec.
-
-### Step 2 — Update `scripts/lib/frontmatter.mjs` if spec 09 has already landed
-
-If spec 09 has already been implemented and `stripFrontmatter` was extracted to `scripts/lib/frontmatter.mjs`, apply the same change to that file instead of (or in addition to) the inline definition in `push-to-confluence.mjs`. If spec 09 has not landed yet, change only `push-to-confluence.mjs` — spec 09 will copy the updated version when it extracts the function.
-
-## Test cases
-
-These test cases are expressed as Node.js assertions for use with spec 09's `node --test` harness. For now they can be run manually.
-
-### TC-01: No frontmatter — identity
-
-```js
-const input = "# Hello\n\nSome content.\n";
-const result = stripFrontmatter(input);
-assert.strictEqual(result, input);
-```
-
-**Expected:** content returned unchanged.
-
-### TC-02: Valid 2-field frontmatter — stripped
-
-```js
-const input = "---\ntitle: My Doc\ndate: 2026-01-01\n---\n# Hello\n\nContent.\n";
-const result = stripFrontmatter(input);
-assert.strictEqual(result, "# Hello\n\nContent.\n");
-```
-
-**Expected:** frontmatter block removed, body intact.
-
-### TC-03: Valid 1-field frontmatter (minimal)
-
-```js
-const input = "---\ntitle: Foo\n---\nBody\n";
-const result = stripFrontmatter(input);
-assert.strictEqual(result, "Body\n");
-```
-
-### TC-04: CRLF frontmatter — stripped
-
-```js
-const input = "---\r\ntitle: Doc\r\ndate: 2026\r\n---\r\nBody\n";
-const result = stripFrontmatter(input);
-assert.strictEqual(result, "Body\n");
-```
-
-**Expected:** CRLF line endings in frontmatter handled correctly; body is intact.
-
-### TC-05: `---` horizontal rule in body after valid frontmatter — HR preserved
-
-```js
-const input = "---\ntitle: Doc\n---\n# Section 1\n\nText.\n\n---\n\n# Section 2\n";
-const result = stripFrontmatter(input);
-assert.strictEqual(result, "# Section 1\n\nText.\n\n---\n\n# Section 2\n");
-```
-
-**Expected:** Only the frontmatter `---` block is stripped. The `---` HR in the body is preserved. This is the key regression test — the old regex could incorrectly match this case if the body `---` came before the frontmatter closing `---` in the document.
-
-**Note:** In this specific test the frontmatter is well-formed (has a closing `---` on the second line), so both old and new regexes produce the correct result. The test is included to guard against regressions if the regex is modified.
-
-### TC-06: Unclosed frontmatter — no content stripped
-
-```js
-// No closing --- anywhere
-const input = "---\ntitle: Doc\ndate: 2026\n\n# Body\n\nContent here.\n";
-const result = stripFrontmatter(input);
-assert.strictEqual(result, input);
-```
-
-**Expected:** content returned unchanged. The old regex also returns unchanged here because there is no second `---` at all, so neither regex matches. This test confirms the baseline.
-
-### TC-07: Unclosed frontmatter with `---` in body — new behaviour
-
-```js
-// Closing --- missing; there IS a --- further down in the body
-const input = "---\ntitle: Doc\n\n# Introduction\n\nSome content.\n\n---\n\n# Section 2\n";
-const result = stripFrontmatter(input);
-// Old regex: strips from byte 0 to the "---" before "# Section 2" — loses "Introduction" section
-// New regex: no match — returns content unchanged
-assert.strictEqual(result, input);
-```
-
-**Expected (new):** content returned unchanged. The regex requires a closing `---` within 50 lines of the opening `---`. The `---` in the body is at line 8 (counting from the opening `---`), so it is within 50 lines — BUT the laziness of `{0,50}?` means the engine first tries to match at line 2 (`\n` after `title: Doc`), then line 3 (empty line), then line 4 (`# Introduction\n`). At each step the next token must be `---`; only at line 8 does the body `---` appear. So the new regex WILL match this and strip "---\ntitle: Doc\n\n# Introduction\n\nSome content.\n\n" — stripping more than just the frontmatter.
-
-**Correction**: The behaviour described above means the new regex does NOT fully fix this edge case when the body `---` is within 50 lines of the opening. The 50-line cap only helps when the body `---` is more than 50 lines away. This is an important limitation to document.
-
-**Revised expected outcome for TC-07**: The new regex matches and strips from byte 0 to the first `---` it finds, which here is the body `---` at line 8. This is still wrong — the same as the old regex. The difference is only when the body `---` is more than 50 lines away from the opening `---`.
-
-**Ralph's action**: Update the test to reflect the actual new behaviour. The correct assertion is:
-
-```js
-// The new regex still matches the body "---" if it is within 50 lines of the opening.
-// The fix only helps for deeply-buried closing delimiters (50+ lines away).
-// Document this as a known limitation; a proper fix requires a full YAML parser.
-const result = stripFrontmatter(input);
-// This will still incorrectly strip — include the test as a documentation of the limitation:
-assert.notStrictEqual(result, input); // documents the known remaining issue
-```
-
-### TC-08: Frontmatter with 40 fields — stripped correctly
-
-```js
-const fields = Array.from({ length: 40 }, (_, i) => `key${i}: value${i}`).join("\n");
-const input = `---\n${fields}\n---\nBody\n`;
-const result = stripFrontmatter(input);
-assert.strictEqual(result, "Body\n");
-```
-
-**Expected:** 40-field frontmatter stripped correctly (within the 50-line cap).
-
-### TC-09: Frontmatter with 51 fields — no match (cap exceeded)
-
-```js
-const fields = Array.from({ length: 51 }, (_, i) => `key${i}: value${i}`).join("\n");
-const input = `---\n${fields}\n---\nBody\n`;
-const result = stripFrontmatter(input);
-assert.strictEqual(result, input); // no change — cap exceeded
-```
-
-**Expected:** content returned unchanged. The closing `---` is at line 52, which exceeds the `{0,50}` cap.
-
-## Acceptance criteria
-
-- The `s` flag is removed from the regex.
-- The function body uses `(?:[^\n]*\r?\n){0,50}?` for the inner match.
-- TC-01 through TC-05 and TC-08 all pass.
-- TC-09 passes (51-field frontmatter is not stripped).
-- TC-07 is documented as a known limitation (the fix only applies when the spurious `---` is more than 50 lines from the opening).
-- No change to the function signature or call sites.
-
-## Verification
-
-```sh
-# 1. Confirm the s flag is gone:
-grep -n "stripFrontmatter" scripts/push-to-confluence.mjs
-# Inspect the nearby lines — should not contain /s flag at end of regex
-
-# 2. Quick inline test of the happy path:
-node -e "
-function stripFrontmatter(content) {
-  return content.replace(/^---\r?\n(?:[^\n]*\r?\n){0,50}?---\s*\r?\n/, '');
-}
-const input = '---\ntitle: Test\ndate: 2026\n---\n# Hello\nContent.\n';
-const result = stripFrontmatter(input);
-console.log(JSON.stringify(result));
-// Expected: '\"# Hello\nContent.\n\"'
-"
-
-# 3. Test the 51-field cap:
-node -e "
-function stripFrontmatter(content) {
-  return content.replace(/^---\r?\n(?:[^\n]*\r?\n){0,50}?---\s*\r?\n/, '');
-}
-const fields = Array.from({length:51},(_,i)=>'k'+i+': v'+i).join('\n');
-const input = '---\n'+fields+'\n---\nBody\n';
-const result = stripFrontmatter(input);
-console.log(result === input ? 'PASS: no change for 51-field frontmatter' : 'FAIL: content was changed');
-"
-
-# 4. Smoke test — push a real file with frontmatter to a test page:
-node scripts/push-to-confluence.mjs test-page /tmp/test-with-frontmatter.md
-```
-
-## Out of scope
-
-- Do not add TOML (`+++`) or JSON (`{ }`) frontmatter support.
-- Do not throw an error on malformed frontmatter — the silent no-op (returning content unchanged) is the correct behaviour. Crashing on malformed frontmatter would be surprising.
-- Do not add a `--strict-frontmatter` flag.
-- Do not integrate a full YAML parser (e.g., `js-yaml`) — the function's purpose is to strip the frontmatter block from the raw text, not parse it. YAML parsing is out of scope.
-- Do not change the 50-line cap — it is a defence-in-depth measure and not intended to be configurable.
-
-## Follow-ups
-
-- **TC-07 known limitation**: The regex still incorrectly matches a body `---` that appears within 50 lines of an unclosed frontmatter opening. A robust fix requires either (a) a YAML parser that verifies the block is syntactically valid YAML before stripping, or (b) a stricter pattern that only accepts key–value lines (`/^[A-Za-z_][A-Za-z0-9_-]*:/`) inside the frontmatter block. Option (b) is achievable with a regex change and is worth a follow-up spec.
-- **Spec 09 integration**: Once `stripFrontmatter` is extracted to `scripts/lib/frontmatter.mjs`, the test cases above become automated `node --test` cases. TC-07 should be included as a test documenting the known limitation (with a comment, not an assertion that it is fixed).
-- **Strict key–value line pattern**: Replace `[^\n]*` with `[A-Za-z_][A-Za-z0-9_-]*:[^\n]*` to ensure only valid YAML scalar key lines are matched inside the block. A `---` that follows a blank line or a non-key line inside the "frontmatter" would not be matched, preventing the TC-07 case.
diff --git a/apps/claude-code/unic-confluence/docs/plans/09-tests-for-pure-functions.md b/apps/claude-code/unic-confluence/docs/plans/09-tests-for-pure-functions.md
deleted file mode 100644
index 1e943f3a..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/09-tests-for-pure-functions.md
+++ /dev/null
@@ -1,726 +0,0 @@
-# 09. Tests for Pure Functions
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** M
-**Depends on:** spec 08 (tightened `stripFrontmatter` — extract the updated version, not the old one); spec 05 (if `postProcessHtml` is to be tested here, it needs to be written first; it can be added in a follow-up)
-**Touches:** `scripts/lib/inject.mjs` (new), `scripts/lib/frontmatter.mjs` (new), `scripts/lib/resolve.mjs` (new), `scripts/lib/inject.test.mjs` (new), `scripts/lib/frontmatter.test.mjs` (new), `scripts/lib/resolve.test.mjs` (new), `scripts/push-to-confluence.mjs`, `package.json`
-
-## Context
-
-Three functions in `push-to-confluence.mjs` are pure (or near-pure) and critical to correctness: `stripFrontmatter` (lines 113–117), `injectContent` (lines 126–199), and `resolvePageId` (lines 207–252). They are untested. The P2 refactors planned in specs 12, 13, and 15 will touch `injectContent` directly (adding error on missing markers) and `resolvePageId` (replacing `process.exit` with a `CliError` class). Before those changes land, these functions need a test harness so regressions are caught immediately rather than at manual test time.
-
-This spec extracts the three functions into a `scripts/lib/` module directory and introduces a `node --test` suite covering the documented behaviours. No new test framework (vitest, jest, mocha) is added — `node:test` is available in Node 18+ and is sufficient for synchronous pure-function tests.
-
-The extraction also resolves a secondary concern: `push-to-confluence.mjs` uses `const require = createRequire(import.meta.url)` solely to `require("marked")`. `marked` 17.0.5 ships with an ESM-compatible entry point in its `exports` field. Switching to `import { marked } from "marked"` removes the CommonJS shim entirely, making the file a clean ESM module.
-
-## Current behaviour
-
-All three functions are defined inline in `scripts/push-to-confluence.mjs`:
-
-- `stripFrontmatter` — lines 113–117
-- `injectContent` (plus `TEXT_START_RE`, `TEXT_END_RE`) — lines 121–199
-- `isNumericId`, `resolvePageId` — lines 203–252
-
-The `createRequire` shim is at lines 11, 24–25:
-
-```js
-import { createRequire } from "module";   // line 11 (in the import block)
-// ...
-const require = createRequire(import.meta.url);  // line 24
-const { marked } = require("marked");            // line 25
-```
-
-There is no `scripts/lib/` directory. There is no `test` script in `package.json`.
-
-## Target behaviour
-
-After this spec:
-
-1. `scripts/lib/inject.mjs` exports `injectContent`, `TEXT_START_RE`, `TEXT_END_RE`.
-2. `scripts/lib/frontmatter.mjs` exports `stripFrontmatter`.
-3. `scripts/lib/resolve.mjs` exports `resolvePageId`, `isNumericId`.
-4. `scripts/push-to-confluence.mjs` imports from `./lib/*.mjs` and removes the inline definitions.
-5. The `createRequire` / `require("marked")` shim is replaced with `import { marked } from "marked"`.
-6. Three `*.test.mjs` files exist under `scripts/lib/` and run with `node --test`.
-7. `package.json` has a `"test"` script that runs all three test files.
-8. All tests pass with `pnpm test` (or `npm test`).
-
-## Implementation steps
-
-### Step 1 — Verify `marked` v17 ESM export
-
-Before removing `createRequire`, confirm that `marked` 17.0.5 exports an ESM entry point:
-
-```sh
-node -e "import('marked').then(m => console.log(Object.keys(m)))"
-```
-
-Expected output includes `"marked"`. If the import resolves correctly, proceed. If it throws, `marked` in this version does not have an ESM entry and `createRequire` must be kept.
-
-Check the package exports field:
-
-```sh
-node -e "const fs=require('fs');const pkg=JSON.parse(fs.readFileSync('node_modules/marked/package.json','utf8'));console.log(JSON.stringify(pkg.exports,null,2))"
-```
-
-Look for an `"import"` condition under `"."` or `"./src/marked.js"`. `marked` has had ESM exports since v4 — v17 should be fine.
-
-### Step 2 — Create `scripts/lib/` directory and module files
-
-The `scripts/lib/` directory does not need to be explicitly created — writing the files is sufficient. Proceed directly to writing the files.
-
-#### `scripts/lib/frontmatter.mjs`
-
-Copy from `push-to-confluence.mjs`. Use the **updated** version from spec 08 (with the `{0,50}` cap), not the original with the `s` flag.
-
-```js
-// SPDX-License-Identifier: LGPL-3.0-or-later
-
-/**
- * Strips YAML frontmatter from the start of a Markdown string.
- *
- * The opening --- must be at byte 0. Frontmatter is capped at 50 lines to
- * prevent runaway matches on unclosed blocks. See spec 08 for full rationale.
- *
- * @param {string} content
- * @returns {string}
- */
-export function stripFrontmatter(content) {
-	return content.replace(
-		/^---\r?\n(?:[^\n]*\r?\n){0,50}?---\s*\r?\n/,
-		"",
-	);
-}
-```
-
-#### `scripts/lib/inject.mjs`
-
-Copy `TEXT_START_RE`, `TEXT_END_RE`, and `injectContent` verbatim from `push-to-confluence.mjs` lines 121–199. Add the `export` keyword before each declaration.
-
-```js
-// SPDX-License-Identifier: LGPL-3.0-or-later
-
-export const TEXT_START_RE =
-	/(?:<p>\s*)?\[AUTO_INSERT_START:\s*([^\]]+?)\s*\](?:\s*<\/p>)?/;
-export const TEXT_END_RE =
-	/(?:<p>\s*)?\[AUTO_INSERT_END:\s*([^\]]+?)\s*\](?:\s*<\/p>)?/;
-
-/**
- * Injects newHtml into existingBody using one of three strategies:
- *   1. Plain-text markers ([AUTO_INSERT_START: label] / [AUTO_INSERT_END: label])
- *   2. Anchor macros (md-start / md-end)
- *   3. Append (no markers — deprecated; spec 03 will change this to an error)
- *
- * Calls process.exit(1) on mismatched or unpaired markers.
- *
- * @param {string} existingBody — Confluence storage-format HTML
- * @param {string} newHtml — HTML to inject
- * @param {string} title — page title, used in error messages only
- * @returns {string}
- */
-export function injectContent(existingBody, newHtml, title) {
-	const hasStart = TEXT_START_RE.test(existingBody);
-	const hasEnd = TEXT_END_RE.test(existingBody);
-
-	// Strategy 1: plain-text markers
-	if (hasStart || hasEnd) {
-		if (hasStart !== hasEnd) {
-			console.error(
-				`Found [AUTO_INSERT_START] without a matching [AUTO_INSERT_END] on page "${title}" — fix the Confluence page before publishing`,
-			);
-			process.exit(1);
-		}
-
-		const startMatch = TEXT_START_RE.exec(existingBody);
-		const startLabel = startMatch[1].trim();
-		const afterStart = startMatch.index + startMatch[0].length;
-
-		const escapedLabel = startLabel.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
-		const labelEndRe = new RegExp(
-			`(?:<p>\\s*)?\\[AUTO_INSERT_END:\\s*${escapedLabel}\\s*\\](?:\\s*<\\/p>)?`,
-		);
-		const endMatch = labelEndRe.exec(existingBody.slice(afterStart));
-
-		if (!endMatch) {
-			console.error(
-				`Marker label mismatch on page "${title}": [AUTO_INSERT_START:${startLabel}] has no matching [AUTO_INSERT_END:${startLabel}] — fix the Confluence page before publishing`,
-			);
-			process.exit(1);
-		}
-
-		return (
-			existingBody.slice(0, afterStart) +
-			"\n" +
-			newHtml +
-			"\n" +
-			existingBody.slice(afterStart + endMatch.index)
-		);
-	}
-
-	// Strategy 2: anchor macros (legacy fallback)
-	const anchorStartRe =
-		/<ac:structured-macro[^>]*ac:name="anchor"[^>]*>\s*<ac:parameter[^>]*>md-start<\/ac:parameter>\s*<\/ac:structured-macro>/;
-	const anchorEndRe =
-		/<ac:structured-macro[^>]*ac:name="anchor"[^>]*>\s*<ac:parameter[^>]*>md-end<\/ac:parameter>\s*<\/ac:structured-macro>/;
-	const hasAnchorStart = anchorStartRe.test(existingBody);
-	const hasAnchorEnd = anchorEndRe.test(existingBody);
-
-	if (hasAnchorStart || hasAnchorEnd) {
-		if (hasAnchorStart !== hasAnchorEnd) {
-			console.error(
-				`Found md-start anchor without md-end (or vice versa) on page "${title}" — fix the Confluence page before publishing`,
-			);
-			process.exit(1);
-		}
-
-		const startMatch = anchorStartRe.exec(existingBody);
-		const endMatch = anchorEndRe.exec(existingBody);
-
-		return (
-			existingBody.slice(0, startMatch.index + startMatch[0].length) +
-			"\n" +
-			newHtml +
-			"\n" +
-			existingBody.slice(endMatch.index)
-		);
-	}
-
-	// Strategy 3: append (no markers found)
-	return existingBody + "\n" + newHtml;
-}
-```
-
-#### `scripts/lib/resolve.mjs`
-
-Copy `isNumericId` and `resolvePageId` from lines 203–252. The function reads `confluence-pages.json` from `process.cwd()` — this coupling to cwd is a known issue (spec 15 will address it). Copy verbatim for now.
-
-```js
-// SPDX-License-Identifier: LGPL-3.0-or-later
-import { existsSync, readFileSync } from "fs";
-import path from "path";
-
-/**
- * Returns true if arg is a non-empty string of decimal digits.
- * @param {string} arg
- * @returns {boolean}
- */
-export function isNumericId(arg) {
-	return /^\d+$/.test(arg);
-}
-
-/**
- * Resolves a page ID from either a numeric string or a key in confluence-pages.json.
- *
- * Calls process.exit(1) on all error conditions. Spec 15 will replace these
- * with CliError throws.
- *
- * @param {string} arg — numeric page ID or key name from confluence-pages.json
- * @returns {number} — positive integer page ID
- */
-export function resolvePageId(arg) {
-	if (isNumericId(arg)) {
-		const id = parseInt(arg, 10);
-		if (!Number.isInteger(id) || id <= 0) {
-			console.error(`Invalid page ID: ${arg}`);
-			process.exit(1);
-		}
-		return id;
-	}
-
-	const pagesPath = path.join(process.cwd(), "confluence-pages.json");
-	if (!existsSync(pagesPath)) {
-		console.error(
-			"confluence-pages.json not found — create it or pass a page ID directly",
-		);
-		process.exit(1);
-	}
-
-	let pages;
-	try {
-		pages = JSON.parse(readFileSync(pagesPath, "utf8"));
-	} catch {
-		console.error("invalid JSON in confluence-pages.json — check syntax");
-		process.exit(1);
-	}
-
-	if (!(arg in pages)) {
-		const keys = Object.keys(pages)
-			.filter((k) => k !== "_comment")
-			.join(", ");
-		console.error(
-			`'${arg}' not found in confluence-pages.json — available keys: ${keys}`,
-		);
-		process.exit(1);
-	}
-
-	const id = pages[arg];
-	if (!Number.isInteger(id) || id <= 0) {
-		console.error(
-			`Invalid page ID for key '${arg}': ${id} — must be a positive integer`,
-		);
-		process.exit(1);
-	}
-
-	return id;
-}
-```
-
-### Step 3 — Update `push-to-confluence.mjs`
-
-Three changes to `push-to-confluence.mjs`:
-
-#### 3a — Replace `createRequire` shim with ESM import
-
-**Before** (lines 11, 24–25):
-
-```js
-import { createRequire } from "module";
-// ... (other imports)
-const require = createRequire(import.meta.url);
-const { marked } = require("marked");
-```
-
-**After:**
-
-```js
-import { marked } from "marked";
-// ... (other imports)
-// (remove both createRequire lines entirely)
-```
-
-Remove `createRequire` from the `import { createRequire } from "module"` line. If `createRequire` was the only named import from `"module"`, remove the entire import statement. If other names were imported from `"module"`, retain the import and remove only `createRequire`.
-
-#### 3b — Remove inline function definitions
-
-Remove the following from `push-to-confluence.mjs` (the functions are now in `scripts/lib/`):
-
-- Lines 113–117: `function stripFrontmatter(...)` — remove entirely.
-- Lines 121–125: `const TEXT_START_RE = ...` and `const TEXT_END_RE = ...` — remove.
-- Lines 126–199: `function injectContent(...)` — remove entirely.
-- Lines 203–205: `function isNumericId(...)` — remove entirely.
-- Lines 207–252: `function resolvePageId(...)` — remove entirely.
-
-Also remove the section comment headers that no longer have function bodies:
-- `// ── YAML frontmatter stripping ────...` (before `stripFrontmatter`)
-- `// ── Content injection strategies ─────...` (before `TEXT_START_RE`)
-- `// ── Page ID resolution ──────...` (before `isNumericId`)
-
-#### 3c — Add imports at the top of `push-to-confluence.mjs`
-
-After the existing import block, add:
-
-```js
-import { injectContent } from "./lib/inject.mjs";
-import { stripFrontmatter } from "./lib/frontmatter.mjs";
-import { resolvePageId } from "./lib/resolve.mjs";
-```
-
-The `isNumericId` function is used only inside `resolvePageId` in the original code — if it is not called directly in `push-to-confluence.mjs`, do not import it.
-
-### Step 4 — Create test files
-
-#### `scripts/lib/frontmatter.test.mjs`
-
-```js
-import { describe, it } from "node:test";
-import assert from "node:assert/strict";
-import { stripFrontmatter } from "./frontmatter.mjs";
-
-describe("stripFrontmatter", () => {
-	it("returns content unchanged when no frontmatter", () => {
-		const input = "# Hello\n\nSome content.\n";
-		assert.strictEqual(stripFrontmatter(input), input);
-	});
-
-	it("strips a minimal 1-field frontmatter block", () => {
-		const input = "---\ntitle: Foo\n---\nBody\n";
-		assert.strictEqual(stripFrontmatter(input), "Body\n");
-	});
-
-	it("strips a 2-field frontmatter block", () => {
-		const input = "---\ntitle: My Doc\ndate: 2026-01-01\n---\n# Hello\n\nContent.\n";
-		assert.strictEqual(stripFrontmatter(input), "# Hello\n\nContent.\n");
-	});
-
-	it("strips CRLF frontmatter", () => {
-		const input = "---\r\ntitle: Doc\r\ndate: 2026\r\n---\r\nBody\n";
-		assert.strictEqual(stripFrontmatter(input), "Body\n");
-	});
-
-	it("preserves --- HR in body after valid frontmatter", () => {
-		const input = "---\ntitle: Doc\n---\n# Section 1\n\nText.\n\n---\n\n# Section 2\n";
-		assert.strictEqual(
-			stripFrontmatter(input),
-			"# Section 1\n\nText.\n\n---\n\n# Section 2\n",
-		);
-	});
-
-	it("does not strip when frontmatter has no closing ---", () => {
-		// No --- anywhere after the opening
-		const input = "---\ntitle: Doc\ndate: 2026\n\n# Body\n\nContent here.\n";
-		assert.strictEqual(stripFrontmatter(input), input);
-	});
-
-	it("does not strip when frontmatter exceeds 50 fields (cap exceeded)", () => {
-		const fields = Array.from({ length: 51 }, (_, i) => `key${i}: value${i}`).join("\n");
-		const input = `---\n${fields}\n---\nBody\n`;
-		assert.strictEqual(stripFrontmatter(input), input);
-	});
-
-	it("strips frontmatter with exactly 50 fields", () => {
-		const fields = Array.from({ length: 50 }, (_, i) => `key${i}: value${i}`).join("\n");
-		const input = `---\n${fields}\n---\nBody\n`;
-		assert.strictEqual(stripFrontmatter(input), "Body\n");
-	});
-
-	it("returns empty string when frontmatter covers whole content", () => {
-		const input = "---\ntitle: Only Frontmatter\n---\n";
-		assert.strictEqual(stripFrontmatter(input), "");
-	});
-});
-```
-
-#### `scripts/lib/inject.test.mjs`
-
-```js
-import { describe, it } from "node:test";
-import assert from "node:assert/strict";
-import { injectContent } from "./inject.mjs";
-
-describe("injectContent — strategy 1: plain-text markers", () => {
-	it("replaces content between matching markers", () => {
-		const body =
-			"<p>[AUTO_INSERT_START: docs]</p>\n<p>old content</p>\n<p>[AUTO_INSERT_END: docs]</p>";
-		const result = injectContent(body, "<p>new content</p>", "Test Page");
-		assert.ok(result.includes("<p>new content</p>"), "new content present");
-		assert.ok(!result.includes("old content"), "old content removed");
-		assert.ok(result.includes("[AUTO_INSERT_START: docs]"), "start marker preserved");
-		assert.ok(result.includes("[AUTO_INSERT_END: docs]"), "end marker preserved");
-	});
-
-	it("preserves content outside the markers", () => {
-		const body =
-			"<p>before</p>\n<p>[AUTO_INSERT_START: docs]</p>\n<p>old</p>\n<p>[AUTO_INSERT_END: docs]</p>\n<p>after</p>";
-		const result = injectContent(body, "<p>new</p>", "Test");
-		assert.ok(result.includes("<p>before</p>"), "content before marker preserved");
-		assert.ok(result.includes("<p>after</p>"), "content after marker preserved");
-	});
-
-	it("handles markers without surrounding <p> tags", () => {
-		const body = "[AUTO_INSERT_START: raw]\nold\n[AUTO_INSERT_END: raw]";
-		const result = injectContent(body, "new", "Test");
-		assert.ok(result.includes("new"), "new content present");
-		assert.ok(!result.includes("old"), "old content removed");
-	});
-
-	// Mismatched label test is skipped until spec 15 replaces process.exit with CliError.
-	// Once CliError is in place, rewrite as:
-	//   assert.throws(() => injectContent(body, html, title), /label mismatch/);
-	it.skip("exits 1 on mismatched marker labels — enable after spec 15 CliError", () => {});
-});
-
-describe("injectContent — strategy 2: anchor macros", () => {
-	const anchorStart =
-		'<ac:structured-macro ac:name="anchor"><ac:parameter ac:name="">md-start</ac:parameter></ac:structured-macro>';
-	const anchorEnd =
-		'<ac:structured-macro ac:name="anchor"><ac:parameter ac:name="">md-end</ac:parameter></ac:structured-macro>';
-
-	it("replaces content between anchor macros", () => {
-		const body = `<p>before</p>${anchorStart}<p>old</p>${anchorEnd}<p>after</p>`;
-		const result = injectContent(body, "<p>new</p>", "Test");
-		assert.ok(result.includes("<p>new</p>"), "new content present");
-		assert.ok(!result.includes("<p>old</p>"), "old content removed");
-		assert.ok(result.includes("<p>before</p>"), "before preserved");
-		assert.ok(result.includes("<p>after</p>"), "after preserved");
-	});
-
-	it("preserves the anchor macros themselves", () => {
-		const body = `${anchorStart}<p>old</p>${anchorEnd}`;
-		const result = injectContent(body, "<p>new</p>", "Test");
-		assert.ok(result.includes("md-start"), "start anchor macro preserved");
-		assert.ok(result.includes("md-end"), "end anchor macro preserved");
-	});
-});
-
-describe("injectContent — strategy 3: append fallback", () => {
-	it("appends new content when no markers are present", () => {
-		const body = "<p>existing</p>";
-		const result = injectContent(body, "<p>new</p>", "Test");
-		assert.ok(result.includes("<p>existing</p>"), "existing content preserved");
-		assert.ok(result.includes("<p>new</p>"), "new content appended");
-		// Verify order: existing before new
-		assert.ok(
-			result.indexOf("<p>existing</p>") < result.indexOf("<p>new</p>"),
-			"existing content comes before new content",
-		);
-	});
-
-	it("appends to empty body", () => {
-		const result = injectContent("", "<p>only</p>", "Test");
-		assert.ok(result.includes("<p>only</p>"), "content present in empty body");
-	});
-});
-```
-
-#### `scripts/lib/resolve.test.mjs`
-
-Testing `resolvePageId` error paths is complicated by its `process.exit` calls. For those paths, write a small wrapper script to disk and invoke it with `node --input-type=module` via Node's `child_process.spawnSync`. This is the standard pattern for testing `process.exit` in `node:test` without a mock framework.
-
-**Note on `spawnSync` safety**: The test uses `spawnSync` with a hardcoded command (`process.execPath`, i.e., the Node binary), not `exec`. No user input is passed to a shell — the `input` option pipes a hardcoded inline script. There is no shell interpolation and no injection surface. `spawnSync` with an explicit argv array is safe here.
-
-```js
-import { describe, it, before, after } from "node:test";
-import assert from "node:assert/strict";
-import { writeFileSync, unlinkSync, existsSync, readFileSync } from "fs";
-import path from "path";
-import { spawnSync } from "child_process";
-import { fileURLToPath } from "url";
-import { resolvePageId, isNumericId } from "./resolve.mjs";
-
-const __dirname = path.dirname(fileURLToPath(import.meta.url));
-const testPagesPath = path.join(process.cwd(), "confluence-pages.json");
-const testPagesBackup = path.join(process.cwd(), "confluence-pages.json.bak");
-
-describe("isNumericId", () => {
-	it("returns true for a numeric string", () => {
-		assert.strictEqual(isNumericId("12345"), true);
-	});
-
-	it("returns false for a key name", () => {
-		assert.strictEqual(isNumericId("my-page"), false);
-	});
-
-	it("returns false for empty string", () => {
-		assert.strictEqual(isNumericId(""), false);
-	});
-
-	it("returns false for a string with letters", () => {
-		assert.strictEqual(isNumericId("123abc"), false);
-	});
-});
-
-describe("resolvePageId — numeric input", () => {
-	it("returns a positive integer for a valid numeric string", () => {
-		const result = resolvePageId("98765");
-		assert.strictEqual(result, 98765);
-	});
-});
-
-describe("resolvePageId — key lookup", () => {
-	// Set up a temporary confluence-pages.json in cwd for these tests.
-	// These tests must run from the repo root (where pnpm test is invoked from).
-
-	before(() => {
-		if (existsSync(testPagesPath)) {
-			writeFileSync(testPagesBackup, readFileSync(testPagesPath, "utf8"));
-		}
-		writeFileSync(
-			testPagesPath,
-			JSON.stringify({ "_comment": "test", "my-page": 11111, "other": 22222 }, null, 2),
-			"utf8",
-		);
-	});
-
-	after(() => {
-		unlinkSync(testPagesPath);
-		if (existsSync(testPagesBackup)) {
-			writeFileSync(testPagesPath, readFileSync(testPagesBackup, "utf8"));
-			unlinkSync(testPagesBackup);
-		}
-	});
-
-	it("returns the page ID for a valid key", () => {
-		const result = resolvePageId("my-page");
-		assert.strictEqual(result, 11111);
-	});
-
-	it("returns the page ID for another valid key", () => {
-		const result = resolvePageId("other");
-		assert.strictEqual(result, 22222);
-	});
-});
-
-describe("resolvePageId — error paths (via subprocess)", () => {
-	// These error paths call process.exit(1). We test them by spawning a child
-	// Node process with a hardcoded inline script via --input-type=module.
-	// spawnSync is used (not exec) — the command is always process.execPath
-	// (the Node binary) with a hardcoded argv array; there is no shell and no
-	// injection surface.
-
-	it("exits 1 for missing confluence-pages.json when using a key", () => {
-		// Run in /tmp which has no confluence-pages.json
-		const resolvePath = path.join(__dirname, "resolve.mjs");
-		const script = `import {resolvePageId} from ${JSON.stringify(resolvePath)}; resolvePageId("missing-key")`;
-		const result = spawnSync(process.execPath, ["--input-type=module"], {
-			input: script,
-			cwd: "/tmp",
-			encoding: "utf8",
-		});
-		assert.strictEqual(result.status, 1, `expected exit 1, got ${result.status}\n${result.stderr}`);
-		assert.ok(
-			result.stderr.includes("confluence-pages.json not found"),
-			`expected error message in stderr:\n${result.stderr}`,
-		);
-	});
-
-	it("exits 1 with available keys message for unknown key", () => {
-		const tmpDir = "/tmp";
-		const tmpPages = path.join(tmpDir, "confluence-pages.json");
-		writeFileSync(tmpPages, JSON.stringify({ "known-page": 99999 }, null, 2));
-
-		const resolvePath = path.join(__dirname, "resolve.mjs");
-		const script = `import {resolvePageId} from ${JSON.stringify(resolvePath)}; resolvePageId("unknown-key")`;
-		const result = spawnSync(process.execPath, ["--input-type=module"], {
-			input: script,
-			cwd: tmpDir,
-			encoding: "utf8",
-		});
-
-		unlinkSync(tmpPages);
-
-		assert.strictEqual(result.status, 1, `expected exit 1, got ${result.status}\n${result.stderr}`);
-		assert.ok(
-			result.stderr.includes("not found in confluence-pages.json"),
-			`expected not-found message:\n${result.stderr}`,
-		);
-		assert.ok(
-			result.stderr.includes("known-page"),
-			`expected available keys listed in error message:\n${result.stderr}`,
-		);
-	});
-});
-```
-
-### Step 5 — Add `test` script to `package.json`
-
-**Before** (the `scripts` block after spec 06 lands, or add `"test"` to the existing object):
-
-```json
-"scripts": {
-  "confluence": "node scripts/push-to-confluence.mjs",
-  "sync-version": "node scripts/sync-version.mjs",
-  "release": "..."
-}
-```
-
-**After:**
-
-```json
-"scripts": {
-  "confluence": "node scripts/push-to-confluence.mjs",
-  "sync-version": "node scripts/sync-version.mjs",
-  "release": "...",
-  "test": "node --test scripts/lib/frontmatter.test.mjs scripts/lib/inject.test.mjs scripts/lib/resolve.test.mjs"
-}
-```
-
-Use explicit file paths rather than a glob for the `test` script. Node's `--test` accepts glob patterns since Node 21 via the `--test-glob` flag, but explicit paths are portable across Node 18+.
-
-## Test cases
-
-### TC-01: All tests pass
-
-```sh
-pnpm test
-# Expected: all tests pass, exit code 0
-# Expected output format (node --test TAP output):
-# ok 1 - stripFrontmatter > returns content unchanged when no frontmatter
-# ok 2 - stripFrontmatter > strips a minimal 1-field frontmatter block
-# ...
-# # tests N
-# # pass N
-# # fail 0
-```
-
-### TC-02: Import smoke test — push-to-confluence.mjs still loads
-
-```sh
-node --check scripts/push-to-confluence.mjs
-# Expected: no syntax errors, exits 0
-```
-
-### TC-03: ESM import of marked works
-
-```sh
-node -e "import('marked').then(m => console.log(typeof m.marked))"
-# Expected: "function"
-```
-
-### TC-04: No inline definitions remain in push-to-confluence.mjs
-
-```sh
-grep -n "function stripFrontmatter\|function injectContent\|function resolvePageId\|function isNumericId" scripts/push-to-confluence.mjs
-# Expected: no output (all definitions removed)
-```
-
-### TC-05: createRequire is gone
-
-```sh
-grep -n "createRequire" scripts/push-to-confluence.mjs
-# Expected: no output (or only comments, not live code)
-```
-
-### TC-06: Regression — publish still works end-to-end
-
-```sh
-node scripts/push-to-confluence.mjs test-page /tmp/test.md
-# Expected: publishes without error, exits 0
-```
-
-## Acceptance criteria
-
-- `scripts/lib/frontmatter.mjs`, `scripts/lib/inject.mjs`, and `scripts/lib/resolve.mjs` exist and export the named functions.
-- `pnpm test` exits 0 with all tests passing.
-- No test uses `vitest`, `jest`, or any third-party test library — only `node:test` and `node:assert/strict`.
-- `push-to-confluence.mjs` no longer contains inline definitions of the three extracted functions.
-- `push-to-confluence.mjs` no longer uses `createRequire` or `require("marked")`.
-- The publish flow (`node scripts/push-to-confluence.mjs pageId file.md`) still works correctly after the extraction.
-- `--setup`, `--verify`, and `--check-auth` (if spec 07 landed) still work correctly.
-
-## Verification
-
-```sh
-# Run the test suite:
-pnpm test
-
-# Confirm no inline function definitions remain:
-grep -n "^function stripFrontmatter\|^function injectContent\|^function resolvePageId" scripts/push-to-confluence.mjs
-
-# Confirm imports are present:
-grep -n "from.*lib/" scripts/push-to-confluence.mjs
-
-# Confirm marked is imported via ESM:
-grep -n 'import.*marked.*from.*"marked"' scripts/push-to-confluence.mjs
-
-# Confirm createRequire is gone:
-grep -n "createRequire" scripts/push-to-confluence.mjs
-
-# Syntax check all new files:
-node --check scripts/lib/frontmatter.mjs
-node --check scripts/lib/inject.mjs
-node --check scripts/lib/resolve.mjs
-node --check scripts/lib/frontmatter.test.mjs
-node --check scripts/lib/inject.test.mjs
-node --check scripts/lib/resolve.test.mjs
-```
-
-## Out of scope
-
-- Do not test `httpsRequest`, `runSetup`, `runVerify`, or `main()` — these require network or interactive I/O and belong in integration tests or e2e tests.
-- Do not add vitest, jest, mocha, or any other test framework.
-- Do not add code coverage tooling.
-- Do not test `loadCredentials` — it reads from the filesystem and environment, making it an integration concern.
-- Do not add `postProcessHtml` (spec 05) to this test suite — add it in a follow-up once spec 05 is implemented.
-- Do not change the runtime behaviour of any extracted function — copy verbatim, export, and import.
-
-## Deviations
-
-- **`injectContent` signature**: The spec template shows a simple `injectContent(existingBody, newHtml, title)` signature (strategy 3 = append fallback). The actual code has evolved through specs 03 and 04 to `injectContent(existingBody, newHtml, title, { replaceAll, dryRun, pageId, version })` where strategy 3 is now an error (no markers) or a full replace (`replaceAll=true`). The lib module exports the actual current implementation verbatim. The strategy-3 tests in `inject.test.mjs` are updated to test the `replaceAll=true, dryRun=true` path instead of the old append path.
-
-## Follow-ups
-
-- **`postProcessHtml` tests**: Once spec 05 lands, extract `postProcessHtml` to `scripts/lib/html.mjs` and add `scripts/lib/html.test.mjs` with the TC-01 through TC-06 cases from spec 05.
-- **Spec 15 — `CliError`**: When `process.exit` calls are replaced with `CliError` throws, the skipped test in `inject.test.mjs` (mismatched labels) and the subprocess tests in `resolve.test.mjs` can be rewritten as direct `assert.throws` calls.
-- **`resolvePageId` cwd coupling**: The function reads `confluence-pages.json` from `process.cwd()`. This makes tests fragile (they depend on cwd or must write temp files). Spec 15 should add an optional `pagesFilePath` parameter so tests can pass an explicit path without cwd manipulation.
-- **CI integration**: Add `pnpm test` to the GitHub Actions workflow (if one exists) so tests run on every push.
diff --git a/apps/claude-code/unic-confluence/docs/plans/10-biome-lint-format-and-ci.md b/apps/claude-code/unic-confluence/docs/plans/10-biome-lint-format-and-ci.md
deleted file mode 100644
index cb3707a5..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/10-biome-lint-format-and-ci.md
+++ /dev/null
@@ -1,328 +0,0 @@
-# 10. Biome Lint, Format, and CI
-
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** M
-**Depends on:** spec 09 (test files must exist for `pnpm test` to run in CI)
-**Touches:** `pnpm-workspace.yaml`, `package.json`, `biome.json` (new), `.github/workflows/ci.yml` (new), `scripts/push-to-confluence.mjs`
-
-## Context
-
-No linting or formatting enforcement exists today. The repo relies on the `.editorconfig` file for editor hints (tabs, LF, UTF-8) but nothing catches style drift in CI, and there is no automated way to verify that a contributor has not introduced bad patterns (unused variables, implicit any, loose equality) before a PR lands. Biome is chosen as a single tool covering both lint and format, avoiding the need to maintain both ESLint and Prettier configs. It understands ESM natively, runs fast enough to use as a pre-push gate, and its formatter is opinionated enough that it eliminates diffs caused by inconsistent quoting or comma styles. This spec also adds a GitHub Actions CI workflow that runs `biome ci` (non-destructive check) and `pnpm test` on Node 22 and 24 so regressions are caught on every push and pull request.
-
-## Current behaviour
-
-- `package.json` has no `"scripts"` field.
-- `pnpm-workspace.yaml` exists (added in spec 00) with a `catalog:` entry for `marked` but no dev-dependency entries.
-- There is no `biome.json`, no ESLint config, no Prettier config.
-- There is no `.github/workflows/` directory.
-- Running `pnpm lint` or `pnpm format` fails with `Missing script: lint` / `Missing script: format`.
-- Indentation in `scripts/push-to-confluence.mjs` is already tabs. Quote style is already double-quote throughout. Trailing commas are present on multi-line argument lists.
-
-**`pnpm-workspace.yaml` current content** (after spec 00):
-```yaml
-packages: []
-catalog:
-  marked: "catalog:"
-```
-
-Wait — spec 00 sets `marked: "17.0.5"` in the catalog section. Exact content after spec 00:
-```yaml
-packages: []
-catalog:
-  marked: "17.0.5"
-```
-
-**`package.json` current content** (after spec 00):
-```json
-{
-  "name": "unic-confluence",
-  "private": true,
-  "license": "LGPL-3.0-or-later",
-  "type": "module",
-  "packageManager": "pnpm@10.x.x",
-  "engines": { "node": ">=22" },
-  "scripts": {
-    "confluence": "node scripts/push-to-confluence.mjs",
-    "sync-version": "node scripts/sync-version.mjs"
-  },
-  "dependencies": {
-    "marked": "catalog:"
-  }
-}
-```
-
-(Exact versions of `packageManager` and other spec-00 additions may differ; what matters here is that no `devDependencies` key exists yet and the `scripts` object does not include `lint`, `format`, `check`, `ci:check`, or `test`.)
-
-## Target behaviour
-
-After this spec:
-
-1. `pnpm lint` runs `biome lint .` — exits 0 on clean source, exits 1 with diagnostics on violations.
-2. `pnpm format` runs `biome format --write .` — rewrites files in-place to match biome style.
-3. `pnpm check` runs `biome check .` — combines lint + format check in one pass (does NOT write).
-4. `pnpm ci:check` runs `biome ci .` — same as `check` but produces machine-readable output and always exits non-zero on any violation. Used in CI only.
-5. `pnpm test` runs all test files via Node's built-in test runner (the exact glob comes from spec 09; this spec wires the script).
-6. `.github/workflows/ci.yml` runs on every push to `main` and on every pull request. It installs dependencies, runs `pnpm ci:check`, and runs `pnpm test`, on both Node 22 and 24.
-7. Existing source files are reformatted by `pnpm check --write` before committing — tabs preserved, double quotes enforced, trailing commas enforced, line width ≤ 100.
-
-Biome version is pinned to the latest stable Biome 1.x at time of implementation. Do not upgrade mid-spec; pin once and move on.
-
-## Implementation steps
-
-### Step 1 — Add Biome to the catalog and devDependencies
-
-Edit `pnpm-workspace.yaml` to add `@biomejs/biome` to the catalog:
-
-```yaml
-# BEFORE:
-packages: []
-catalog:
-  marked: "17.0.5"
-
-# AFTER:
-packages: []
-catalog:
-  marked: "17.0.5"
-  "@biomejs/biome": "1.9.4"   # Ralph: replace with latest stable Biome 1.x from https://biomejs.dev/guides/getting-started/
-```
-
-Edit `package.json` to add `devDependencies`:
-
-```json
-// BEFORE (relevant excerpt):
-{
-  "dependencies": {
-    "marked": "catalog:"
-  }
-}
-
-// AFTER:
-{
-  "dependencies": {
-    "marked": "catalog:"
-  },
-  "devDependencies": {
-    "@biomejs/biome": "catalog:"
-  }
-}
-```
-
-Run `pnpm install` to generate the updated `pnpm-lock.yaml`.
-
-### Step 2 — Add scripts to package.json
-
-The full `"scripts"` block after this spec (merges with whatever spec 00 and spec 09 introduced):
-
-```json
-"scripts": {
-  "confluence": "node scripts/push-to-confluence.mjs",
-  "sync-version": "node scripts/sync-version.mjs",
-  "lint": "biome lint .",
-  "format": "biome format --write .",
-  "check": "biome check .",
-  "ci:check": "biome ci .",
-  "test": "node --test scripts/lib/inject.test.mjs scripts/lib/frontmatter.test.mjs scripts/lib/resolve.test.mjs"
-}
-```
-
-Note: the `test` command lists spec-09 test files explicitly. If spec 09 uses a glob-based invocation instead (e.g. `node --test 'scripts/lib/*.test.mjs'`), defer to spec 09's choice — just ensure this spec's `test` script matches exactly what spec 09 defines.
-
-### Step 3 — Create biome.json
-
-Create `/Users/oriol.torrent/Sites/UNIC/unic-claude-code-confluence/biome.json`:
-
-```json
-{
-  "$schema": "https://biomejs.dev/schemas/1.9.4/schema.json",
-  "organizeImports": {
-    "enabled": true
-  },
-  "linter": {
-    "enabled": true,
-    "rules": {
-      "recommended": true
-    }
-  },
-  "formatter": {
-    "enabled": true,
-    "indentStyle": "tab",
-    "lineEnding": "lf",
-    "lineWidth": 100
-  },
-  "javascript": {
-    "formatter": {
-      "quoteStyle": "double",
-      "semicolons": "always",
-      "trailingCommas": "all"
-    }
-  },
-  "files": {
-    "ignore": ["node_modules", ".history", "pnpm-lock.yaml"]
-  }
-}
-```
-
-Key decisions:
-- `"indentStyle": "tab"` matches `.editorconfig` and existing source.
-- `"quoteStyle": "double"` matches existing code — all string literals in `push-to-confluence.mjs` use double quotes.
-- `"trailingCommas": "all"` matches the existing style on multi-line expressions.
-- `"lineWidth": 100` gives enough room for the long error message strings in `handleHttpError` without wrapping them, while still catching runaway long lines.
-- `"$schema"` pin: update this version string to match whatever Biome version you installed in step 1.
-
-### Step 4 — Format all existing files in one pass
-
-Run:
-```sh
-pnpm check --write
-```
-
-Review the diff carefully. This may reformat `push-to-confluence.mjs` (line endings, trailing commas, import ordering). Accept all formatting-only changes. Reject any change that alters logic (Biome's `--write` only applies safe formatter rewrites, but verify anyway).
-
-Common changes to expect in `push-to-confluence.mjs`:
-- Import block re-ordered alphabetically by `organizeImports`
-- Trailing comma added to the last item in multi-line import lists
-- Any `'single-quoted'` string literals converted to `"double-quoted"` (there should be none, but verify)
-
-After the format pass, commit all changes to source files as part of this spec's PR, not as a separate "chore: format" commit. Keep them together so history is clean.
-
-### Step 5 — Create the GitHub Actions CI workflow
-
-Create `/Users/oriol.torrent/Sites/UNIC/unic-claude-code-confluence/.github/workflows/ci.yml`:
-
-```yaml
-name: CI
-
-on:
-  push:
-    branches: [main]
-  pull_request:
-
-jobs:
-  ci:
-    name: Lint, format, test
-    runs-on: ubuntu-latest
-    strategy:
-      matrix:
-        node-version: ["22", "24"]
-    steps:
-      - uses: actions/checkout@v4
-      - uses: pnpm/action-setup@v4
-        with:
-          version: 10
-      - uses: actions/setup-node@v4
-        with:
-          node-version: ${{ matrix.node-version }}
-          cache: pnpm
-      - run: pnpm install --frozen-lockfile
-      - run: pnpm ci:check
-      - run: pnpm test
-```
-
-Notes:
-- `pnpm/action-setup@v4` reads `packageManager` from `package.json` (set by spec 00) to match the pnpm version. The `version: 10` here is a fallback; if spec 00 sets `"packageManager": "pnpm@10.x.x"` the action uses that.
-- `--frozen-lockfile` ensures CI fails fast if `pnpm-lock.yaml` is out of sync with `package.json`, which would indicate a missing `pnpm install` commit.
-- `cache: pnpm` uses GitHub's built-in pnpm cache integration on the `setup-node` action, keyed to `pnpm-lock.yaml`. This shaves 30–60 s from warm CI runs.
-- Matrix on `["22", "24"]` covers the current LTS and the upcoming LTS. Node 22 is the `engines` minimum defined in spec 00.
-
-### Step 6 — Verify the workflow locally (optional but recommended)
-
-If `act` is installed:
-```sh
-act push --job ci
-```
-
-Otherwise, push a branch and watch the Actions tab.
-
-## Test cases
-
-### TC-01: clean repo passes CI check
-```sh
-pnpm ci:check
-# Expected: exit 0, no diagnostics
-```
-
-### TC-02: lint violation causes failure
-Temporarily add an unused variable to `scripts/push-to-confluence.mjs`:
-```js
-const unused = 42;
-```
-```sh
-pnpm lint
-# Expected: exit 1
-# Expected output includes: noUnusedVariables or similar diagnostic
-```
-Remove the line afterward.
-
-### TC-03: format violation causes failure
-Temporarily change a double-quoted string to single-quoted in `push-to-confluence.mjs`:
-```js
-const x = 'hello';
-```
-```sh
-pnpm ci:check
-# Expected: exit 1
-# Expected output: "expected double quotes"
-```
-Run `pnpm format` to fix, verify `pnpm ci:check` exits 0 again.
-
-### TC-04: tests pass
-```sh
-pnpm test
-# Expected: exit 0, all test files report "pass"
-```
-
-### TC-05: GitHub Actions matrix
-Push the branch to GitHub. Confirm both Node 22 and Node 24 jobs turn green in the Actions tab.
-
-## Acceptance criteria
-
-- `pnpm lint` exits 0 on the current codebase with no violations
-- `pnpm format` rewrites files to match biome config without logic changes
-- `pnpm ci:check` exits 0 on a clean repo, exits 1 when a lint or format violation is introduced
-- `pnpm test` exits 0 (depends on spec 09 tests passing)
-- `biome.json` exists at repo root with `indentStyle: tab`, `quoteStyle: double`, `trailingCommas: all`
-- `.github/workflows/ci.yml` runs on push to `main` and on PRs
-- CI matrix covers Node 22 and Node 24
-- No `eslint`, `prettier`, or `@types/eslint` packages are introduced
-- `pnpm-lock.yaml` is committed and up to date after `pnpm install`
-
-## Verification
-
-```sh
-# Install dependencies (generates/updates pnpm-lock.yaml)
-pnpm install
-
-# Check all files pass biome (lint + format)
-pnpm ci:check
-
-# Run tests
-pnpm test
-
-# Confirm biome.json is valid JSON
-node -e "JSON.parse(require('fs').readFileSync('biome.json', 'utf8')); console.log('biome.json OK')"
-
-# Confirm CI workflow file parses as YAML
-node -e "
-  const fs = require('fs');
-  // Just check it exists and is non-empty
-  const s = fs.statSync('.github/workflows/ci.yml');
-  console.log('.github/workflows/ci.yml exists, size:', s.size);
-"
-```
-
-## Out of scope
-
-- No TypeScript (`tsc`) in this spec — that is spec 14.
-- No pre-commit hooks (Husky / lint-staged) — optional follow-up, not part of this spec.
-- No `biome migrate` step — we are creating a fresh config, not migrating from ESLint.
-- Do not change the Biome version after pinning. Pin once during implementation and leave it until a dedicated "upgrade Biome" task.
-- Do not add `@typescript-eslint` or any ESLint package.
-- Do not modify the logic in `push-to-confluence.mjs` — this spec only touches formatting.
-
-## Follow-ups
-
-- Add a pre-commit hook (Husky + lint-staged) that runs `biome check --write` on staged files before commit.
-- Add `biome.json` schema version auto-update when Biome is upgraded (can be a Renovate/Dependabot rule).
-- Add `"release"` script once spec 07 (release automation) is implemented.
diff --git a/apps/claude-code/unic-confluence/docs/plans/11-readme-troubleshooting-section.md b/apps/claude-code/unic-confluence/docs/plans/11-readme-troubleshooting-section.md
deleted file mode 100644
index 6d179b4c..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/11-readme-troubleshooting-section.md
+++ /dev/null
@@ -1,186 +0,0 @@
-# 11. README Troubleshooting Section
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** S
-**Depends on:** none (documentation-only change)
-**Touches:** `README.md`
-
-## Context
-
-`scripts/push-to-confluence.mjs` contains well-crafted, specific error messages for the most common failure modes: 401 token expiry, 403 permission issues, 404 missing page ID, 409 version conflict, network errors, empty HTML, and missing `confluence-pages.json`. These messages are good enough to unblock an experienced developer who reads them carefully, but they appear only at runtime — a user who cannot get the command to run at all (e.g. because they have not set up VPN, or they have not created `confluence-pages.json`) will see nothing useful. The README currently has no troubleshooting section. This spec adds one, positioned after "Per-repo setup → Verify page IDs", so it is the first place a stuck user looks after the setup instructions. Each entry is keyed to a recognisable symptom rather than to an internal mechanism, making it searchable and scannable.
-
-## Current behaviour
-
-`README.md` sections (in order):
-
-1. Usage
-2. Per-repo setup
-   - Install plugin
-   - Install as npm package
-   - Create confluence-pages.json
-   - Configure credentials
-   - Verify page IDs
-3. Updating
-4. Naming convention
-5. License
-
-There is no troubleshooting section. No error symptom reference. No VPN mention. No mention that API tokens expire.
-
-The error messages in source that this section mirrors:
-
-| Source location | Error text |
-|---|---|
-| `push-to-confluence.mjs:99` | `"API token rejected — generate a new one at https://id.atlassian.com → Security → API tokens (note: tokens created before 2025 may have expired)"` |
-| `push-to-confluence.mjs:100` | `"Access denied — check that your API token has permission to read and write this page"` |
-| `push-to-confluence.mjs:101` | `"Page ID not found — check confluence-pages.json or verify the page still exists"` |
-| `push-to-confluence.mjs:102` | `` `Page was updated by someone else. Retry with: npm run confluence -- ${process.argv[2]} ${process.argv[3]}` `` |
-| `push-to-confluence.mjs:88` | `"Cannot reach Confluence — check VPN/network connectivity"` |
-| `push-to-confluence.mjs:425-428` | `"Markdown converted to empty HTML — check the source file is not empty"` |
-| `push-to-confluence.mjs:218-221` | `"confluence-pages.json not found — create it or pass a page ID directly"` |
-| `push-to-confluence.mjs:54` | `` "Run `npm run confluence -- --setup` to configure credentials" `` |
-
-## Target behaviour
-
-`README.md` gains a "## Troubleshooting" section immediately before "## Updating". The section contains one subsection per common failure mode. Each subsection:
-
-- Starts with a `###` heading phrased as the user sees it (the HTTP status code or the symptom keyword).
-- Gives one or two sentences explaining the most likely cause.
-- Gives the exact corrective action (command to run, URL to visit, person to contact).
-- Does not duplicate prose from the setup section — links back by heading anchor where relevant.
-
-After this spec the section reads exactly as shown in the Implementation steps below. Do not paraphrase.
-
-## Implementation steps
-
-### Step 1 — Locate the insertion point in README.md
-
-Find the line in `README.md` that reads `## Updating` (the exact heading for the Updating section). Insert the new section **immediately before** that line, with one blank line separating the new section from the Updating heading.
-
-### Step 2 — Insert the Troubleshooting section
-
-The exact Markdown to insert (copy verbatim, preserving blank lines between subsections):
-
-````markdown
-## Troubleshooting
-
-### 401 API token rejected
-
-Generate a new token at <https://id.atlassian.com> → Security → API tokens.
-Tokens created before 2025 may have expired even if the Atlassian interface shows them as active.
-After generating, re-run setup:
-
-```sh
-pnpm confluence --setup
-# or, if running the script directly:
-node scripts/push-to-confluence.mjs --setup
-```
-
-### 403 Access denied
-
-Your API token does not have Edit permission for this Confluence page or space.
-Ask a Confluence space admin to grant your account Edit access, then retry.
-
-### 404 Page ID not found
-
-The page ID in `confluence-pages.json` is incorrect, or the page has been deleted or moved.
-Verify the ID by opening the page in Confluence and checking the URL — the numeric ID appears after `/pages/`.
-Run `pnpm confluence --verify` to test all IDs in `confluence-pages.json` at once.
-
-### 409 Conflict — page updated by someone else
-
-Someone edited the Confluence page between the time this script read the current version (GET) and tried to write the new version (PUT).
-Re-run the same command — the script will fetch the current version and retry cleanly.
-
-### Network error / Cannot reach Confluence
-
-Make sure you are connected to the Unic VPN.
-The Confluence instance (`https://uniccom.atlassian.net`) is not reachable from the public internet.
-
-### Markdown converts to empty HTML
-
-The Markdown source file may be empty, or it may contain only a YAML frontmatter block with no body content below the closing `---`.
-Check the file path you passed as the second argument.
-
-### confluence-pages.json not found
-
-Either create `confluence-pages.json` at your repository root (see [Per-repo setup → Create confluence-pages.json](#create-confluence-pagesjson) above), or pass the numeric Confluence page ID directly as the first argument instead of a key name.
-
-### Credentials not configured
-
-Run `pnpm confluence --setup` (or `node scripts/push-to-confluence.mjs --setup`) to store your Confluence URL, email, and API token in `~/.unic-confluence.json`.
-Alternatively, set the environment variables `CONFLUENCE_URL`, `CONFLUENCE_USER`, and `CONFLUENCE_TOKEN`.
-````
-
-### Step 3 — Verify section heading anchor
-
-The `confluence-pages.json` subsection links to `#create-confluence-pagesjson`. Check that the existing README has a heading that renders to that anchor. In standard GitHub Markdown, `## Create confluence-pages.json` becomes `#create-confluence-pagesjson`. If the existing heading is worded differently, adjust the anchor link in the troubleshooting entry to match the actual heading, or leave it as a plain reference without a link.
-
-### Step 4 — Update the 409 retry command reference
-
-The troubleshooting entry for 409 says "Re-run the same command". Once spec 12 lands, the runtime error message will also print the exact command. These two are consistent — no change needed to keep them in sync.
-
-## Test cases
-
-### TC-01: Section renders correctly on GitHub
-Push the change and open `README.md` on GitHub. Each `###` heading should appear as a level-3 heading in the rendered view. The fenced code block inside "401 API token rejected" should render as a code block with `sh` syntax highlighting.
-
-### TC-02: Anchor link resolves
-Click the link `#create-confluence-pagesjson` in the "confluence-pages.json not found" subsection when viewing the README on GitHub. It should scroll to the correct setup subsection. If it does not, fix the anchor as described in Step 3.
-
-### TC-03: Section position
-The `## Troubleshooting` heading must appear after the last Per-repo setup subsection and before `## Updating`. Verify by reading the raw Markdown section order.
-
-### TC-04: No content duplication
-No setup instruction is repeated verbatim in the troubleshooting section — it only contains cross-references (links or short mentions). Verify by reading both sections.
-
-### TC-05: All 8 error categories covered
-Count the `###` subsections: 401, 403, 404, 409, Network, Empty HTML, confluence-pages.json, Credentials. Exactly 8 subsections present.
-
-## Acceptance criteria
-
-- `README.md` contains a `## Troubleshooting` section
-- Section is placed between "Per-repo setup" (or its last subsection) and `## Updating`
-- Exactly 8 `###` subsections, one per common failure mode listed in "Current behaviour"
-- The 401 subsection includes the `https://id.atlassian.com` URL
-- The 401 subsection includes a code block showing how to re-run setup with `pnpm`
-- The Network subsection explicitly mentions VPN
-- The Network subsection includes the Confluence base URL `https://uniccom.atlassian.net`
-- The credentials subsection mentions all three env vars: `CONFLUENCE_URL`, `CONFLUENCE_USER`, `CONFLUENCE_TOKEN`
-- No `handleHttpError`, `push-to-confluence.mjs`, or internal function names are mentioned in the README (keep user-facing language)
-- No changes to any `.mjs` file in this spec
-
-## Verification
-
-```sh
-# Confirm section exists
-grep -n "## Troubleshooting" README.md
-
-# Confirm all 8 subsections are present
-grep -c "^### " README.md
-# Expected: 8 (or more if other subsections existed before — adjust count accordingly)
-
-# Confirm VPN mention
-grep -i "vpn" README.md
-
-# Confirm credential env vars are mentioned
-grep "CONFLUENCE_URL" README.md
-grep "CONFLUENCE_USER" README.md
-grep "CONFLUENCE_TOKEN" README.md
-
-# Confirm section is before Updating
-awk '/^## Troubleshooting/{t=NR} /^## Updating/{u=NR} END{if(t && u && t<u) print "OK: Troubleshooting before Updating"; else print "FAIL"}' README.md
-```
-
-## Out of scope
-
-- Do not modify `handleHttpError` in `push-to-confluence.mjs` — the error messages in the source file are separate from the README documentation.
-- Do not add a "debug mode" flag or `--verbose` flag — that is a separate feature.
-- Do not add screenshots or GIFs to the README.
-- Do not rewrite existing README sections — only insert the new Troubleshooting section.
-- Do not change the heading hierarchy of existing sections.
-
-## Follow-ups
-
-- Once spec 12 lands (fix `handleHttpError` argv leak), update the 409 troubleshooting entry to show the exact `pnpm confluence <pageArg> <filePath>` retry syntax instead of the generic "re-run the same command".
-- Consider adding a "Known limitations" section covering the 5 MB file size limit and single-page-at-a-time constraint.
diff --git a/apps/claude-code/unic-confluence/docs/plans/12-fix-handlehttperror-argv-leak.md b/apps/claude-code/unic-confluence/docs/plans/12-fix-handlehttperror-argv-leak.md
deleted file mode 100644
index 76066fd5..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/12-fix-handlehttperror-argv-leak.md
+++ /dev/null
@@ -1,294 +0,0 @@
-# 12. Fix handleHttpError argv Leak
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** S
-**Depends on:** none (self-contained change to one function and its two call sites)
-**Touches:** `scripts/push-to-confluence.mjs`
-
-## Context
-
-`handleHttpError` is the central HTTP error handler. It currently reads `process.argv[2]` and `process.argv[3]` directly to build the 409 conflict retry hint (line 102). This is fragile: `process.argv[2]` is always whatever the first CLI token after `node script.mjs` was, and `process.argv[3]` is the second. Today the call is `node scripts/push-to-confluence.mjs <pageArg> <filePath>`, so `argv[2]` and `argv[3]` are the right values — but that is only true as long as no flags are inserted before the positional arguments. Specs 03 and 04 introduce flags like `--dry-run` and `--replace-all`, which may appear as `argv[2]`, pushing the page argument to `argv[3]` or later and making the retry hint in the 409 message print the wrong values. The fix is to pass the resolved page argument and file path explicitly from the call sites into `handleHttpError`, decoupling the error formatter from the global argv state.
-
-A secondary issue: the message currently reads `npm run confluence --` which is inconsistent with the rest of the codebase moving to `pnpm confluence` (spec 00). The fix updates this reference too.
-
-## Current behaviour
-
-**File:** `scripts/push-to-confluence.mjs`
-
-**`handleHttpError` definition (lines 97–109):**
-```js
-function handleHttpError(status, title) {
-	const messages = {
-		401: "API token rejected — generate a new one at https://id.atlassian.com → Security → API tokens (note: tokens created before 2025 may have expired)",
-		403: "Access denied — check that your API token has permission to read and write this page",
-		404: "Page ID not found — check confluence-pages.json or verify the page still exists",
-		409: `Page was updated by someone else. Retry with: npm run confluence -- ${process.argv[2]} ${process.argv[3]}`,
-	};
-	const msg =
-		messages[status] ??
-		`Unexpected response from Confluence (HTTP ${status}) — check VPN/network and retry`;
-	console.error(msg);
-	process.exit(1);
-}
-```
-
-**Problems on line 102:**
-1. `process.argv[2]` and `process.argv[3]` are global state — they break when flags appear before positional args.
-2. If either is `undefined` (e.g. the user only passed one argument), the message reads `"Retry with: pnpm confluence undefined undefined"`.
-3. `"npm run confluence --"` should be `"pnpm confluence"` after spec 00.
-
-**Call sites:**
-
-Call site A — GET response error path (line 443):
-```js
-if (getRes.status < 200 || getRes.status >= 300) {
-	handleHttpError(getRes.status, "");
-}
-```
-
-Call site B — PUT response error path (line 487–488):
-```js
-if (putRes.status < 200 || putRes.status >= 300) {
-	handleHttpError(putRes.status, title);
-}
-```
-
-Both call sites are inside `main()`, which already has `pageArg` and `filePath` in scope (line 405):
-```js
-const [pageArg, filePath] = args;
-```
-
-**`runVerify` (lines 256–310):** calls `httpsRequest` directly and uses `console.error` for failures — does NOT call `handleHttpError`. No change needed there.
-
-## Target behaviour
-
-`handleHttpError` accepts an optional third argument `{ pageArg, filePath }`. The 409 message is built from those values. If `pageArg` is absent or empty, the message degrades gracefully to a version-agnostic retry instruction. Both call sites in `main()` pass `{ pageArg, filePath }`.
-
-The 409 message changes from:
-```
-Page was updated by someone else. Retry with: npm run confluence -- my-page docs/spec.md
-```
-to:
-```
-Page was updated by someone else. Retry with: pnpm confluence my-page docs/spec.md
-```
-
-Or, if no pageArg is available:
-```
-Page was updated by someone else. Re-run the command with the current page version.
-```
-
-All other messages (401, 403, 404, the catch-all) are unchanged.
-
-## Implementation steps
-
-### Step 1 — Update the function signature and 409 message
-
-**File:** `scripts/push-to-confluence.mjs`, lines 97–109
-
-```js
-// BEFORE:
-function handleHttpError(status, title) {
-	const messages = {
-		401: "API token rejected — generate a new one at https://id.atlassian.com → Security → API tokens (note: tokens created before 2025 may have expired)",
-		403: "Access denied — check that your API token has permission to read and write this page",
-		404: "Page ID not found — check confluence-pages.json or verify the page still exists",
-		409: `Page was updated by someone else. Retry with: npm run confluence -- ${process.argv[2]} ${process.argv[3]}`,
-	};
-	const msg =
-		messages[status] ??
-		`Unexpected response from Confluence (HTTP ${status}) — check VPN/network and retry`;
-	console.error(msg);
-	process.exit(1);
-}
-
-// AFTER:
-function handleHttpError(status, title, { pageArg = "", filePath = "" } = {}) {
-	const retryHint = pageArg
-		? `Retry with: pnpm confluence ${pageArg} ${filePath}`
-		: "Re-run the command with the current page version.";
-	const messages = {
-		401: "API token rejected — generate a new one at https://id.atlassian.com → Security → API tokens (note: tokens created before 2025 may have expired)",
-		403: "Access denied — check that your API token has permission to read and write this page",
-		404: "Page ID not found — check confluence-pages.json or verify the page still exists",
-		409: `Page was updated by someone else. ${retryHint}`,
-	};
-	const msg =
-		messages[status] ??
-		`Unexpected response from Confluence (HTTP ${status}) — check VPN/network and retry`;
-	console.error(msg);
-	process.exit(1);
-}
-```
-
-The `retryHint` variable is computed before the `messages` object so the template literal on line 409 can reference it cleanly.
-
-### Step 2 — Update call site A (GET error path)
-
-**File:** `scripts/push-to-confluence.mjs`, around line 443
-
-```js
-// BEFORE:
-if (getRes.status < 200 || getRes.status >= 300) {
-	handleHttpError(getRes.status, "");
-}
-
-// AFTER:
-if (getRes.status < 200 || getRes.status >= 300) {
-	handleHttpError(getRes.status, "", { pageArg, filePath });
-}
-```
-
-`pageArg` and `filePath` are already in scope — they are destructured from `args` at line 405:
-```js
-const [pageArg, filePath] = args;
-```
-
-### Step 3 — Update call site B (PUT error path)
-
-**File:** `scripts/push-to-confluence.mjs`, around line 487–488
-
-```js
-// BEFORE:
-if (putRes.status < 200 || putRes.status >= 300) {
-	handleHttpError(putRes.status, title);
-}
-
-// AFTER:
-if (putRes.status < 200 || putRes.status >= 300) {
-	handleHttpError(putRes.status, title, { pageArg, filePath });
-}
-```
-
-`title` is in scope (derived from `pageData?.title ?? ""` at line 464). `pageArg` and `filePath` are also in scope from line 405.
-
-### Step 4 — Confirm no other call sites exist
-
-Search for all occurrences of `handleHttpError` in the file:
-
-```sh
-grep -n "handleHttpError" scripts/push-to-confluence.mjs
-```
-
-Expected output (exact line numbers may shift after prior specs):
-```
-97:function handleHttpError(status, title, { pageArg = "", filePath = "" } = {}) {
-443:	handleHttpError(getRes.status, "", { pageArg, filePath });
-488:	handleHttpError(putRes.status, title, { pageArg, filePath });
-```
-
-If any additional call site appears (e.g. inside `runVerify`), examine it. `runVerify` should NOT call `handleHttpError` — it uses `console.error` directly. If a call site exists in `runVerify`, do not pass `pageArg`/`filePath` (they would not be in scope there); instead add a separate guard so the 409 message falls back to the no-arg path.
-
-### Step 5 — Note on pageArg vs pageId
-
-Pass `pageArg` (the raw string the user typed, e.g. `"my-page"` or `"123456789"`), NOT `pageId` (the resolved numeric integer). The retry hint is for the user to re-type on their terminal. A human-readable key name like `"home-overview"` is more useful in the retry hint than the resolved numeric `"987654321"`.
-
-`pageArg` is the pre-resolution string at line 405:
-```js
-const [pageArg, filePath] = args;      // raw user input: "my-page", "docs/spec.md"
-const pageId = resolvePageId(pageArg); // resolved: 987654321
-```
-
-## Test cases
-
-### TC-01: 409 with both pageArg and filePath provided
-Simulate a 409 response with `pageArg = "home-overview"` and `filePath = "docs/home.md"`:
-
-Expected message printed to stderr:
-```
-Page was updated by someone else. Retry with: pnpm confluence home-overview docs/home.md
-```
-
-Expected exit code: `1`
-
-### TC-02: 409 with numeric pageArg
-Simulate a 409 with `pageArg = "987654321"` and `filePath = "docs/spec.md"`:
-
-Expected message:
-```
-Page was updated by someone else. Retry with: pnpm confluence 987654321 docs/spec.md
-```
-
-### TC-03: 409 with empty pageArg (degraded path)
-Call `handleHttpError(409, "My Page", {})` directly (or with `pageArg = ""`):
-
-Expected message:
-```
-Page was updated by someone else. Re-run the command with the current page version.
-```
-
-Expected exit code: `1`
-
-### TC-04: 409 with no third argument (default parameter path)
-Call `handleHttpError(409, "My Page")` with no third argument:
-
-Expected message:
-```
-Page was updated by someone else. Re-run the command with the current page version.
-```
-
-Expected exit code: `1`
-
-This verifies the `= {}` default prevents a destructuring TypeError.
-
-### TC-05: Other status codes unchanged
-- `handleHttpError(401, "")` → `"API token rejected — …"`
-- `handleHttpError(403, "")` → `"Access denied — …"`
-- `handleHttpError(404, "")` → `"Page ID not found — …"`
-- `handleHttpError(500, "")` → `"Unexpected response from Confluence (HTTP 500) — …"`
-
-All should still work without providing `pageArg`/`filePath`.
-
-### TC-06: No `process.argv` reference inside handleHttpError
-```sh
-grep -n "process\.argv" scripts/push-to-confluence.mjs
-```
-Expected: `process.argv` appears only in `main()` (line 389, `process.argv.slice(2)`), NOT inside `handleHttpError`.
-
-## Acceptance criteria
-
-- `handleHttpError` signature is `(status, title, { pageArg = "", filePath = "" } = {})`
-- No `process.argv` reference exists inside the body of `handleHttpError`
-- The 409 message includes `"pnpm confluence"` (not `"npm run confluence --"`) when `pageArg` is non-empty
-- The 409 message falls back gracefully when `pageArg` is empty or not provided
-- Both call sites in `main()` pass `{ pageArg, filePath }` as the third argument
-- `runVerify` does not call `handleHttpError` — verify with `grep`
-- All other error messages (401, 403, 404, catch-all) are unchanged in text
-- `pnpm test` passes (spec 09 tests cover `handleHttpError` indirectly via integration paths, or spec 09 adds a direct unit test — either way, no test regression)
-
-## Verification
-
-```sh
-# No argv reference inside handleHttpError body
-# (The function spans ~15 lines starting at line 97; argv should only appear in main())
-grep -n "process\.argv" scripts/push-to-confluence.mjs
-
-# Confirm new signature is present
-grep -n "pageArg" scripts/push-to-confluence.mjs
-
-# Confirm pnpm reference in 409 message
-grep -n "pnpm confluence" scripts/push-to-confluence.mjs
-
-# Confirm no "npm run confluence" reference remains in the 409 message
-grep -n "npm run confluence" scripts/push-to-confluence.mjs
-# Expected: 0 matches (the Usage error at line ~401 may still say "npm run confluence" — that is
-# a separate issue; only the 409 message is in scope for this spec)
-
-# Confirm both call sites pass the third argument
-grep -A2 "handleHttpError" scripts/push-to-confluence.mjs
-```
-
-## Out of scope
-
-- Do not change the 401, 403, or 404 messages.
-- Do not change `runVerify` — it uses `console.error` directly and does not call `handleHttpError`.
-- Do not add a `--retry` flag or automatic retry logic — the message is informational only.
-- Do not update the `Usage:` error message at the top of `main()` (line ~401 — `"Usage: npm run confluence -- {pageId} {file.md}"`). That is a separate cleanup item.
-- Do not add tests for `handleHttpError` in this spec — tests are spec 09's domain.
-
-## Follow-ups
-
-- Update the `Usage:` error message at line ~401 from `"npm run confluence --"` to `"pnpm confluence"` (small follow-up, can be bundled with spec 00 cleanup or spec 12 itself if convenient).
-- Once spec 09 (lib extraction) lands, add a unit test for `handleHttpError` that verifies the 409 message text programmatically rather than relying on manual grep.
-- Update the README troubleshooting entry for 409 (spec 11) to show the exact `pnpm confluence <pageArg> <filePath>` form once this fix is shipped.
diff --git a/apps/claude-code/unic-confluence/docs/plans/13-normalise-p-wrapping-markers.md b/apps/claude-code/unic-confluence/docs/plans/13-normalise-p-wrapping-markers.md
deleted file mode 100644
index 70073135..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/13-normalise-p-wrapping-markers.md
+++ /dev/null
@@ -1,317 +0,0 @@
-# 13. Normalise `<p>`-Wrapping on Injection Markers
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** M
-**Depends on:** none (self-contained change to `injectContent` and its regexes)
-**Touches:** `scripts/push-to-confluence.mjs`
-
-## Context
-
-`injectContent` (Strategy 1, lines 121–165) handles `[AUTO_INSERT_START:label]` and `[AUTO_INSERT_END:label]` markers that users place in Confluence page body HTML. Confluence's rich-text editor wraps free-standing text nodes in `<p>` elements, so a marker typed into the Confluence editor often arrives in storage format as `<p>[AUTO_INSERT_START:label]</p>` rather than the bare `[AUTO_INSERT_START:label]`. The current regexes handle this by making both the leading `<p>` and the trailing `</p>` optional on each match:
-
-```js
-const TEXT_START_RE =
-    /(?:<p>\s*)?\[AUTO_INSERT_START:\s*([^\]]+?)\s*\](?:\s*<\/p>)?/;
-```
-
-This has a subtle structural flaw. When the marker is `<p>`-wrapped, the slice `existingBody.slice(afterStart + endMatch.index)` begins at the `[AUTO_INSERT_END:…]` text, not at the `<p>` that precedes it. That `<p>` was inside the "content to replace" zone and gets discarded, but the paired `</p>` after the marker text is preserved — producing a dangling `</p>` in the output. Confluence re-renders this as invalid storage HTML, occasionally producing a validation warning or extra whitespace on the page.
-
-The fix is to use separate regex patterns for the `<p>`-wrapped and bare cases, detecting which form is present and applying the correct one consistently across both START and END markers, so the full `<p>marker</p>` block is consumed symmetrically.
-
-## Current behaviour
-
-**File:** `scripts/push-to-confluence.mjs`
-
-**Lines 121–124 — marker regexes:**
-```js
-const TEXT_START_RE =
-	/(?:<p>\s*)?\[AUTO_INSERT_START:\s*([^\]]+?)\s*\](?:\s*<\/p>)?/;
-const TEXT_END_RE =
-	/(?:<p>\s*)?\[AUTO_INSERT_END:\s*([^\]]+?)\s*\](?:\s*<\/p>)?/;
-```
-
-Both regexes make `<p>` and `</p>` independently optional via `(?:…)?`. The optional groups are NOT paired: the regex can match `<p>[AUTO_INSERT_START:label]` without a trailing `</p>`, or `[AUTO_INSERT_START:label]</p>` without a leading `<p>`. In practice Confluence always emits matched pairs, but the regex does not enforce this — and the slice logic that follows does not account for the wrapper being part of the match.
-
-**Lines 159–165 — the defective return slice:**
-```js
-return (
-	existingBody.slice(0, afterStart) +
-	"\n" +
-	newHtml +
-	"\n" +
-	existingBody.slice(afterStart + endMatch.index)
-);
-```
-
-When the END marker is `<p>[AUTO_INSERT_END:label]</p>`, `endMatch.index` points to `[AUTO_INSERT_END:…]`, not to the `<p>` before it. That `<p>` is included in the discarded content zone, but the `</p>` after the marker text is preserved, producing:
-
-```html
-<p>[AUTO_INSERT_START:label]</p>
-[injected HTML]
-[AUTO_INSERT_END:label]</p>   ← dangling </p>
-```
-
-**Line 148 — the label-specific END regex also uses optional groups:**
-```js
-const labelEndRe = new RegExp(
-	`(?:<p>\\s*)?\\[AUTO_INSERT_END:\\s*${escapedLabel}\\s*\\](?:\\s*<\\/p>)?`,
-);
-```
-Same problem: `(?:<p>\\s*)?` and `(?:\\s*<\\/p>)?` are both optional and independently matchable.
-
-## Target behaviour
-
-Strategy 1 must consume markers and their `<p>` wrappers symmetrically:
-
-1. Detect whether markers are `<p>`-wrapped by testing `_P_RE` patterns first.
-2. If `<p>`-wrapped: use regex patterns that require the full `<p>…</p>` pair (no optional groups). Both START and END patterns consume the entire `<p>marker</p>` block.
-3. If bare (no `<p>`): use patterns that match only the bare marker text.
-4. Wrapping style must be consistent between START and END on the same page. Mismatch → log error + `process.exit(1)`.
-5. The return value preserves both markers verbatim and replaces only the content between them.
-6. Strategy 2 (anchor macros, lines 168–195) and Strategy 3 (append, lines 197–199) are unchanged.
-
-## Implementation steps
-
-### Step 1 — Replace the two module-level regex constants
-
-**File:** `scripts/push-to-confluence.mjs`, lines 121–124
-
-Before:
-```js
-const TEXT_START_RE =
-	/(?:<p>\s*)?\[AUTO_INSERT_START:\s*([^\]]+?)\s*\](?:\s*<\/p>)?/;
-const TEXT_END_RE =
-	/(?:<p>\s*)?\[AUTO_INSERT_END:\s*([^\]]+?)\s*\](?:\s*<\/p>)?/;
-```
-
-After (4 named constants replacing 2):
-```js
-// Bare markers — no surrounding <p> tags
-const TEXT_START_BARE_RE = /\[AUTO_INSERT_START:\s*([^\]]+?)\s*\]/;
-const TEXT_END_BARE_RE   = /\[AUTO_INSERT_END:\s*([^\]]+?)\s*\]/;
-// Full <p>-wrapped markers — both opening and closing tags required, no optional groups
-const TEXT_START_P_RE    = /<p>\s*\[AUTO_INSERT_START:\s*([^\]]+?)\s*\]\s*<\/p>/;
-const TEXT_END_P_RE      = /<p>\s*\[AUTO_INSERT_END:\s*([^\]]+?)\s*\]\s*<\/p>/;
-```
-
-Design notes:
-- `BARE_RE`: no `<p>` at all, match only the bracket syntax.
-- `P_RE`: both `<p>` and `</p>` are required. No `?` quantifier on either wrapper.
-- `\s*` inside `<p>\s*[…` and `…]\s*<\/p>` allows for whitespace between the tag and the bracket (Confluence occasionally adds a non-breaking space or newline inside the `<p>`).
-- The capture group `([^\]]+?)` extracts the label in all four patterns.
-
-### Step 2 — Rewrite the Strategy 1 block inside `injectContent`
-
-Replace lines 126–165 (the opening of `injectContent` through the end of the Strategy 1 return). Strategy 2 (lines 168–195) and Strategy 3 (lines 197–199) are NOT changed.
-
-New function opening and Strategy 1 block:
-
-```js
-function injectContent(existingBody, newHtml, title) {
-	// ── Strategy 1: plain-text markers ────────────────────────────────────────
-
-	const hasPWrappedStart = TEXT_START_P_RE.test(existingBody);
-	const hasBareStart     = TEXT_START_BARE_RE.test(existingBody);
-	const hasPWrappedEnd   = TEXT_END_P_RE.test(existingBody);
-	const hasBareEnd       = TEXT_END_BARE_RE.test(existingBody);
-
-	const hasStart = hasPWrappedStart || hasBareStart;
-	const hasEnd   = hasPWrappedEnd   || hasBareEnd;
-
-	if (hasStart || hasEnd) {
-		// Must have both or neither
-		if (hasStart !== hasEnd) {
-			console.error(
-				`Found [AUTO_INSERT_START] without a matching [AUTO_INSERT_END] on page "${title}" — fix the Confluence page before publishing`,
-			);
-			process.exit(1);
-		}
-
-		// Wrapping style must be consistent across START and END
-		const startIsWrapped = hasPWrappedStart;
-		const endIsWrapped   = hasPWrappedEnd;
-
-		if (startIsWrapped !== endIsWrapped) {
-			console.error(
-				`Marker wrapping mismatch on page "${title}": START is ${startIsWrapped ? "<p>-wrapped" : "bare"} but END is ${endIsWrapped ? "<p>-wrapped" : "bare"} — fix the Confluence page so both markers use the same format`,
-			);
-			process.exit(1);
-		}
-
-		const START_RE = startIsWrapped ? TEXT_START_P_RE : TEXT_START_BARE_RE;
-
-		const startMatch = START_RE.exec(existingBody);
-		const startLabel = startMatch[1].trim();
-
-		// Build a label-specific END regex with the same wrapping style as START.
-		// Escape special regex characters in the label before interpolating.
-		const escapedLabel = startLabel.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
-		const labelEndRe = endIsWrapped
-			? new RegExp(`<p>\\s*\\[AUTO_INSERT_END:\\s*${escapedLabel}\\s*\\]\\s*<\\/p>`)
-			: new RegExp(`\\[AUTO_INSERT_END:\\s*${escapedLabel}\\s*\\]`);
-
-		// Search for the END marker only in the content AFTER the START marker ends.
-		// This avoids false label-mismatch errors when the page body contains
-		// documentation examples of the marker syntax with different labels.
-		const afterStart = startMatch.index + startMatch[0].length;
-		const endMatch = labelEndRe.exec(existingBody.slice(afterStart));
-
-		if (!endMatch) {
-			console.error(
-				`Marker label mismatch on page "${title}": [AUTO_INSERT_START:${startLabel}] has no matching [AUTO_INSERT_END:${startLabel}] — fix the Confluence page before publishing`,
-			);
-			process.exit(1);
-		}
-
-		// Slice construction:
-		//   prefix   — everything up to the start of the START marker (incl. its <p> if wrapped)
-		//   suffix   — everything from the end of the END marker (incl. its </p> if wrapped)
-		// The old content between markers is discarded. Both markers are re-inserted verbatim.
-		const prefixEnd   = startMatch.index;                                 // start of START marker
-		const suffixStart = afterStart + endMatch.index + endMatch[0].length; // end of END marker
-
-		return (
-			existingBody.slice(0, prefixEnd) +
-			startMatch[0] +       // preserve START marker verbatim (with <p> wrapper if present)
-			"\n" +
-			newHtml +
-			"\n" +
-			endMatch[0] +         // preserve END marker verbatim (with <p> wrapper if present)
-			existingBody.slice(suffixStart)
-		);
-	}
-
-	// Strategy 2 and Strategy 3 follow here (unchanged) …
-```
-
-Key differences from the original:
-- `prefixEnd = startMatch.index` (start of START marker, not end)
-- `startMatch[0]` is explicitly re-inserted to preserve the START marker
-- `endMatch[0]` is explicitly re-inserted to preserve the END marker
-- `suffixStart = afterStart + endMatch.index + endMatch[0].length` (end of END marker including `</p>` if wrapped)
-- No dangling `<p>` or `</p>` can appear because the full match — including wrapper tags — is accounted for in `startMatch[0]` and `endMatch[0]`
-
-### Step 3 — Verify no remaining references to old regex names
-
-```sh
-grep -n "TEXT_START_RE\b\|TEXT_END_RE\b" scripts/push-to-confluence.mjs
-```
-Expected: 0 results. The two old names are fully replaced by the four new ones.
-
-## Test cases
-
-All test cases belong in `scripts/lib/inject.test.mjs` (created by spec 09).
-
-### TC-01: bare markers — no `<p>` wrapping
-```js
-const result = injectContent(
-  "<h1>Intro</h1>\n[AUTO_INSERT_START:overview]\n<p>old</p>\n[AUTO_INSERT_END:overview]\n<p>Footer</p>",
-  "<p>new</p>",
-  "Test Page",
-);
-// Expected:
-// "<h1>Intro</h1>\n[AUTO_INSERT_START:overview]\n<p>new</p>\n[AUTO_INSERT_END:overview]\n<p>Footer</p>"
-```
-No dangling tags. Both markers preserved. Old content replaced.
-
-### TC-02: `<p>`-wrapped markers — full pair consumed
-```js
-const result = injectContent(
-  "<h1>Intro</h1>\n<p>[AUTO_INSERT_START:overview]</p>\n<p>old</p>\n<p>[AUTO_INSERT_END:overview]</p>\n<p>Footer</p>",
-  "<p>new</p>",
-  "Test Page",
-);
-// Expected:
-// "<h1>Intro</h1>\n<p>[AUTO_INSERT_START:overview]</p>\n<p>new</p>\n<p>[AUTO_INSERT_END:overview]</p>\n<p>Footer</p>"
-```
-Both `<p>…</p>` wrappers preserved intact. No dangling tags.
-
-### TC-03: `<p>`-wrapped with internal whitespace
-```js
-const result = injectContent(
-  "<p>  [AUTO_INSERT_START:overview]  </p>\n<p>old</p>\n<p>  [AUTO_INSERT_END:overview]  </p>",
-  "<p>new</p>",
-  "Test Page",
-);
-// Expected: markers matched (whitespace-tolerant), full <p>  …  </p> blocks preserved verbatim
-```
-
-### TC-04: mismatch — START bare, END `<p>`-wrapped → error + exit 1
-```
-existingBody: "[AUTO_INSERT_START:overview]\n<p>old</p>\n<p>[AUTO_INSERT_END:overview]</p>"
-Expected: process.exit(1), stderr contains "Marker wrapping mismatch"
-```
-
-### TC-05: mismatch — START `<p>`-wrapped, END bare → error + exit 1
-```
-existingBody: "<p>[AUTO_INSERT_START:overview]</p>\n<p>old</p>\n[AUTO_INSERT_END:overview]"
-Expected: process.exit(1), stderr contains "Marker wrapping mismatch"
-```
-
-### TC-06: START without END → existing error path unchanged
-```
-existingBody: "[AUTO_INSERT_START:overview]\n<p>content</p>"
-Expected: process.exit(1), stderr contains "Found [AUTO_INSERT_START] without a matching [AUTO_INSERT_END]"
-```
-
-### TC-07: label mismatch → existing error path unchanged
-```
-existingBody: "[AUTO_INSERT_START:overview]\n<p>content</p>\n[AUTO_INSERT_END:summary]"
-Expected: process.exit(1), stderr contains "Marker label mismatch" and "overview"
-```
-
-### TC-08: no markers → Strategy 3 append (unchanged)
-```js
-injectContent("<p>existing</p>", "<p>appended</p>", "Test Page")
-// Expected: "<p>existing</p>\n<p>appended</p>"
-```
-
-### TC-09: Strategy 2 anchor macros — no regression
-Input `existingBody` contains `<ac:structured-macro ac:name="anchor">…md-start…</ac:structured-macro>` and `…md-end…`.
-Expected: Strategy 2 path triggered, content replaced between anchors, no crash.
-
-## Acceptance criteria
-
-- `TEXT_START_RE` and `TEXT_END_RE` no longer exist in the file
-- Four new constants defined: `TEXT_START_BARE_RE`, `TEXT_END_BARE_RE`, `TEXT_START_P_RE`, `TEXT_END_P_RE`
-- `P_RE` patterns contain no `?` quantifier on the `<p>` or `</p>` portions (verified by inspection)
-- TC-01 (bare): output has no dangling tags, both markers preserved
-- TC-02 (`<p>`-wrapped): full `<p>marker</p>` blocks preserved, no dangling tags
-- TC-03 (whitespace tolerance): `\s*` in P_RE patterns works
-- TC-04 and TC-05 (wrapping mismatch): exit 1 with `"Marker wrapping mismatch"` in stderr
-- TC-06 (START without END): exit 1, existing error message text unchanged
-- TC-07 (label mismatch): exit 1, existing error message text unchanged
-- TC-08 (Strategy 3 append): no regression
-- TC-09 (Strategy 2 anchors): no regression
-- `pnpm test` passes all cases
-
-## Verification
-
-```sh
-# Confirm old regex names are gone
-grep -c "TEXT_START_RE\b\|TEXT_END_RE\b" scripts/push-to-confluence.mjs
-# Expected: 0
-
-# Confirm four new names are defined
-grep -n "TEXT_START_BARE_RE\|TEXT_END_BARE_RE\|TEXT_START_P_RE\|TEXT_END_P_RE" scripts/push-to-confluence.mjs
-# Expected: 4 definition lines + usage lines inside injectContent
-
-# Run tests
-pnpm test
-```
-
-## Out of scope
-
-- Do not touch Strategy 2 (anchor macros, lines 168–195).
-- Do not touch Strategy 3 (append, lines 197–199).
-- Do not change the `injectContent` function signature.
-- Do not change the label-mismatch error message text (TC-07 path).
-- Do not add support for multiple marker pairs on the same page.
-- Do not change how `marked` generates HTML.
-- Do not normalise Confluence storage format on fetch (a larger refactor deferred to post-spec-09).
-
-## Follow-ups
-
-- Add a `--validate-markers` flag to fetch a page and report marker wrapping style without pushing content.
-- Post-spec-09: normalise markers on fetch by stripping `<p>` wrappers from bare text markers, so the script always writes bare markers back and never has to handle the `<p>`-wrapped case at runtime.
diff --git a/apps/claude-code/unic-confluence/docs/plans/14-jsdoc-types-ts-check.md b/apps/claude-code/unic-confluence/docs/plans/14-jsdoc-types-ts-check.md
deleted file mode 100644
index c3fc9d25..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/14-jsdoc-types-ts-check.md
+++ /dev/null
@@ -1,436 +0,0 @@
-# 14. JSDoc Types and `--checkJs` Type-Checking
-**Status: done — 2026-04-24**
-
-**Priority:** P2
-**Effort:** M
-**Depends on:** spec 09 (lib extraction — the per-file modules must exist before annotating them), spec 10 (Biome — format pass should run before JSDoc is added so the two diffs do not tangle)
-**Touches:** `scripts/push-to-confluence.mjs`, `scripts/sync-version.mjs`, `scripts/lib/inject.mjs`, `scripts/lib/frontmatter.mjs`, `scripts/lib/resolve.mjs`, `scripts/lib/types.mjs` (new), `package.json`, `pnpm-workspace.yaml`
-
-## Context
-
-The codebase is plain JavaScript (ESM). There is no build step and no TypeScript transpilation. Adding `// @ts-check` at the top of each source file and JSDoc type annotations on public functions gives editor-level type safety (VS Code underlines type errors in real time) and catches an entire class of bugs — wrong argument types, missing properties, `undefined` values used as strings — without introducing a TypeScript build step. The `tsc --checkJs` flag extends this to CI: type errors block merges. This is the lightest possible path to type safety in a no-build ESM project. The choice to NOT convert to `.ts` files is deliberate — it keeps the repo usable as a copy-paste snippet or a Claude plugin without a compile step.
-
-## Current behaviour
-
-- No `// @ts-check` directive in any `.mjs` file.
-- No JSDoc annotations on any function.
-- No `tsconfig.json`.
-- No `typescript` dev dependency.
-- `pnpm typecheck` script does not exist.
-- VS Code shows no type errors (because `checkJs` is off) even when, for example, `pageData?.version?.number` is used as an `integer` parameter without a guard.
-
-Known latent type issues in `push-to-confluence.mjs`:
-- Line 456: `const version = pageData?.version?.number` — type is `number | undefined`, not `number`. Used on line 477 as `version + 1` and passed to `Number.isInteger(version)`. Requires a guard.
-- Line 405: `const [pageArg, filePath] = args` where `args = process.argv.slice(2)` — type is `string[]`, so destructuring gives `string | undefined`. Passing to `resolvePageId(pageArg)` without a guard is a latent type error when `args` is empty (but `main()` checks `args.length < 2` before this line, so it is safe at runtime; tsc needs the narrowing to be explicit).
-- `JSON.parse` return type is `any` — assignments like `const pageData = JSON.parse(getRes.body)` give `any` which propagates and suppresses all downstream checks.
-- `process.argv[2]` is `string | undefined` — already guarded in some places, not in others.
-
-## Target behaviour
-
-After this spec:
-
-1. Every `.mjs` source file starts with `// @ts-check` on the second line (after the SPDX comment or shebang).
-2. `scripts/lib/types.mjs` defines shared type shapes as JSDoc `@typedef` blocks (no runtime exports).
-3. Key functions are annotated with `@param` and `@returns` JSDoc tags.
-4. `pnpm typecheck` runs `tsc --noEmit --allowJs --checkJs --strict --target ES2022 --module NodeNext --moduleResolution NodeNext scripts/**/*.mjs` and exits 0.
-5. VS Code (with the TypeScript language server) shows 0 errors in the Problems panel for all `.mjs` files.
-6. No `.ts` files are created. No transpilation step is introduced.
-
-## Implementation steps
-
-### Step 1 — Add `typescript` to the catalog and devDependencies
-
-Edit `pnpm-workspace.yaml`:
-
-```yaml
-# BEFORE:
-catalog:
-  marked: "17.0.5"
-  "@biomejs/biome": "1.9.4"
-
-# AFTER:
-catalog:
-  marked: "17.0.5"
-  "@biomejs/biome": "1.9.4"
-  typescript: "5.8.3"   # Ralph: replace with latest stable TypeScript 5.x from https://www.typescriptlang.org/
-```
-
-Edit `package.json` `devDependencies`:
-
-```json
-"devDependencies": {
-  "@biomejs/biome": "catalog:",
-  "typescript": "catalog:"
-}
-```
-
-Run `pnpm install`.
-
-### Step 2 — Add `pnpm typecheck` script to package.json
-
-Add to the `"scripts"` block:
-
-```json
-"typecheck": "tsc --noEmit --allowJs --checkJs --strict --target ES2022 --module NodeNext --moduleResolution NodeNext scripts/**/*.mjs"
-```
-
-The `--strict` flag enables `strictNullChecks`, `strictFunctionTypes`, `noImplicitAny`, and others. This may surface errors in step 5 that need fixing. It is intentional — better to find them now than later.
-
-Do NOT create a `tsconfig.json` file. The inline flags are sufficient for a repo with one directory of JS files. If the command line grows unwieldy in a follow-up, a `tsconfig.json` can be added then.
-
-### Step 3 — Create `scripts/lib/types.mjs`
-
-Create a new file at `scripts/lib/types.mjs`. This file has NO runtime exports — it exists only to define JSDoc types that can be imported with `@import` directives in other files.
-
-```js
-// @ts-check
-// SPDX-License-Identifier: LGPL-3.0-or-later
-// Copyright © 2026 Unic
-/**
- * Shared JSDoc type definitions for push-to-confluence.mjs and its lib modules.
- * No runtime exports — import types only via JSDoc @import directives.
- */
-
-/**
- * Credentials loaded from env vars or ~/.unic-confluence.json.
- *
- * @typedef {{ url: string, username: string, token: string }} Credentials
- */
-
-/**
- * Raw HTTP response from httpsRequest.
- *
- * @typedef {{ status: number, body: string }} HttpResponse
- */
-
-/**
- * Confluence page version object embedded in the page GET response.
- *
- * @typedef {{ number: number }} PageVersion
- */
-
-/**
- * Confluence page data as returned by GET /wiki/api/v2/pages/:id?body-format=storage.
- *
- * @typedef {{
- *   title: string,
- *   version: PageVersion,
- *   body: { storage: { value: string } }
- * }} PageData
- */
-
-/**
- * Options for injectContent.
- *
- * @typedef {{
- *   replaceAll?: boolean,
- *   pageId?: number,
- *   version?: number
- * }} InjectOptions
- */
-
-export {};
-```
-
-The `export {}` at the end makes this a valid ESM module (required for `--moduleResolution NodeNext`). The `@typedef` blocks are visible to any file that references this module via JSDoc `@import`.
-
-### Step 4 — Add `// @ts-check` and JSDoc to `scripts/lib/frontmatter.mjs`
-
-This file is created by spec 09. It exports `stripFrontmatter`.
-
-Add as line 1 (before the SPDX comment, or as line 2 after it):
-```js
-// @ts-check
-```
-
-Annotate `stripFrontmatter`:
-```js
-/**
- * Strips a YAML frontmatter block from the start of a Markdown string.
- * The opening `---` must be at byte 0; a CommonMark HR in the body is not affected.
- *
- * @param {string} content - Raw file content including optional frontmatter.
- * @returns {string} Content with the frontmatter block removed, or the original string if none.
- */
-function stripFrontmatter(content) {
-```
-
-### Step 5 — Add `// @ts-check` and JSDoc to `scripts/lib/resolve.mjs`
-
-This file is created by spec 09. It exports `resolvePageId`.
-
-Add `// @ts-check` at the top.
-
-Annotate `resolvePageId`:
-```js
-/**
- * Resolves a page argument to a numeric Confluence page ID.
- * Accepts either a bare numeric string ("987654321") or a key name looked up
- * in `confluence-pages.json` at the current working directory.
- *
- * @param {string} arg - Raw user input: numeric ID or confluence-pages.json key.
- * @returns {number} Positive integer Confluence page ID.
- */
-function resolvePageId(arg) {
-```
-
-### Step 6 — Add `// @ts-check` and JSDoc to `scripts/lib/inject.mjs`
-
-This file is created by spec 09. It exports `injectContent`.
-
-Add at the top:
-```js
-// @ts-check
-/** @import { InjectOptions } from './types.mjs' */
-```
-
-Annotate `injectContent`:
-```js
-/**
- * Injects `newHtml` into `existingBody` using one of three strategies:
- * 1. Plain-text markers ([AUTO_INSERT_START:label] / [AUTO_INSERT_END:label])
- * 2. Confluence anchor macros (md-start / md-end)
- * 3. Append (no markers found)
- *
- * @param {string} existingBody - Current Confluence page storage-format HTML.
- * @param {string} newHtml - New HTML to inject (output of marked()).
- * @param {string} title - Page title, used in error messages only.
- * @param {InjectOptions} [opts] - Reserved for future flags (unused in current implementation).
- * @returns {string} Updated storage-format HTML with new content injected.
- */
-function injectContent(existingBody, newHtml, title, opts) {
-```
-
-### Step 7 — Add `// @ts-check` and JSDoc to `scripts/push-to-confluence.mjs`
-
-Add `// @ts-check` as line 3 (after `#!/usr/bin/env node` and the SPDX comment):
-
-```js
-#!/usr/bin/env node
-// SPDX-License-Identifier: LGPL-3.0-or-later
-// @ts-check
-// Copyright © 2026 Unic
-```
-
-Add import directives after the copyright comment (before the `import` statements):
-```js
-/** @import { Credentials, HttpResponse, PageData } from './lib/types.mjs' */
-```
-
-Annotate key functions:
-
-**`loadCredentials`:**
-```js
-/**
- * Loads Confluence credentials from environment variables or ~/.unic-confluence.json.
- * Exits with code 1 if credentials are not configured.
- *
- * @returns {Credentials}
- */
-function loadCredentials() {
-```
-
-**`makeBasicAuth`:**
-```js
-/**
- * @param {string} username
- * @param {string} token
- * @returns {string} Base64-encoded Basic auth header value.
- */
-function makeBasicAuth(username, token) {
-```
-
-**`httpsRequest`:**
-```js
-/**
- * Makes an HTTPS request and resolves with the status code and response body.
- * Rejects if the network is unreachable.
- *
- * @param {string} method - HTTP method ("GET" or "PUT").
- * @param {string} urlStr - Full HTTPS URL.
- * @param {string} authHeader - Authorization header value.
- * @param {object} [bodyObj] - Request body object, JSON-serialised if provided.
- * @returns {Promise<HttpResponse>}
- */
-function httpsRequest(method, urlStr, authHeader, bodyObj) {
-```
-
-**`handleHttpError`** (after spec 12 has updated the signature):
-```js
-/**
- * Prints a human-readable error for a non-2xx HTTP response and exits with code 1.
- *
- * @param {number} status - HTTP status code.
- * @param {string} title - Page title, used in error messages.
- * @param {{ pageArg?: string, filePath?: string }} [opts] - Used to build the 409 retry hint.
- * @returns {never}
- */
-function handleHttpError(status, title, { pageArg = "", filePath = "" } = {}) {
-```
-
-Note `@returns {never}` — `handleHttpError` always calls `process.exit(1)`, so its return type is `never`. This allows tsc to understand that code after a `handleHttpError` call is unreachable.
-
-### Step 8 — Fix type errors surfaced by `pnpm typecheck`
-
-Run `pnpm typecheck` and address each error. Common issues and their fixes:
-
-**Issue 1: `JSON.parse` returns `any`**
-
-On lines where `JSON.parse` result is assigned, add a cast comment:
-
-```js
-// BEFORE:
-const pageData = JSON.parse(getRes.body);
-
-// AFTER:
-/** @type {PageData} */
-const pageData = /** @type {PageData} */ (JSON.parse(getRes.body));
-```
-
-Similarly for `confluence-pages.json` parsing in `resolvePageId`:
-```js
-/** @type {Record<string, unknown>} */
-const pages = /** @type {Record<string, unknown>} */ (JSON.parse(readFileSync(pagesPath, "utf8")));
-```
-
-**Issue 2: `process.argv[n]` is `string | undefined`**
-
-In `main()`, `process.argv.slice(2)` returns `string[]`. Destructuring:
-```js
-const [pageArg, filePath] = args;
-```
-gives `pageArg: string | undefined` and `filePath: string | undefined`. The `args.length < 2` guard before this line narrows the array length but tsc may not narrow the destructured variables. Fix by adding a post-guard assertion or using non-null assertion style:
-
-```js
-// BEFORE:
-const [pageArg, filePath] = args;
-
-// AFTER:
-const pageArg  = args[0] ?? "";
-const filePath = args[1] ?? "";
-```
-
-This is safe because the `args.length < 2` guard above already exits if either is missing. The `?? ""` provides a fallback that satisfies the type checker without adding a second runtime check.
-
-**Issue 3: `pageData?.version?.number` is `number | undefined`**
-
-```js
-// BEFORE:
-const version = pageData?.version?.number;
-if (!Number.isInteger(version) || version <= 0) {
-
-// AFTER — same runtime behaviour, tsc-friendly:
-const version = pageData?.version?.number;
-if (typeof version !== "number" || !Number.isInteger(version) || version <= 0) {
-```
-
-The `typeof version !== "number"` check narrows the type to `number` for the rest of the block, satisfying `strictNullChecks`. `Number.isInteger` alone does not narrow `number | undefined` to `number` in tsc's type system.
-
-**Issue 4: optional chaining on typed objects**
-
-After the `pageData` cast, `pageData.title` and `pageData.body.storage.value` may still need optional chaining guards because `PageData` is defined with all fields required but the actual Confluence response could omit them if the API changes. Keep the `?? ""` fallbacks — they are correct.
-
-### Step 9 — Add `// @ts-check` to `scripts/sync-version.mjs`
-
-This file is created by spec 06 (version sync). Add `// @ts-check` at the top and annotate its functions following the same pattern. The exact annotations depend on spec 06's implementation. At minimum:
-- Any function that reads JSON files should cast the `JSON.parse` result.
-- Any function that writes files should have `@param` annotations.
-
-### Step 10 — Verify zero type errors
-
-```sh
-pnpm typecheck
-# Expected: exit 0, no output
-```
-
-If errors remain after step 8, address them one by one. Do NOT disable `@ts-check` or add `@ts-ignore` directives to work around errors — fix the underlying type issue. The only acceptable suppression is `// @ts-expect-error` with a comment explaining why, on a line where tsc is provably wrong.
-
-## Test cases
-
-### TC-01: `pnpm typecheck` exits 0 on clean codebase
-```sh
-pnpm typecheck
-# Expected: exit 0
-```
-
-### TC-02: Introducing a type error causes `pnpm typecheck` to exit 1
-Temporarily add to `push-to-confluence.mjs`:
-```js
-const x = loadCredentials();
-const bad = x.nonExistentField; // 'nonExistentField' does not exist on type 'Credentials'
-```
-```sh
-pnpm typecheck
-# Expected: exit 1, output mentions 'nonExistentField' and 'Credentials'
-```
-Remove the line afterward.
-
-### TC-03: `handleHttpError` return type is `never`
-After annotating `@returns {never}`, tsc should not complain about unreachable code after a call to `handleHttpError`. Verify by adding code after a `handleHttpError` call and checking that tsc reports it as unreachable (not as an error, but as a warning, depending on tsc flags).
-
-### TC-04: `JSON.parse` casts suppress `any` propagation
-After adding `/** @type {PageData} */` casts, `pageData.title` should be typed as `string` (not `any`) in the IDE. Verify in VS Code by hovering over `title`.
-
-### TC-05: No `.ts` files created
-```sh
-find scripts -name "*.ts" | wc -l
-# Expected: 0
-```
-
-### TC-06: `pnpm test` still passes
-```sh
-pnpm test
-# Expected: exit 0 — JSDoc annotations must not break any existing tests
-```
-
-## Acceptance criteria
-
-- `// @ts-check` appears at the top of all `.mjs` files in `scripts/` (including lib files from spec 09)
-- `scripts/lib/types.mjs` exists with `Credentials`, `HttpResponse`, `PageVersion`, `PageData`, and `InjectOptions` typedefs
-- `loadCredentials`, `httpsRequest`, `handleHttpError`, `stripFrontmatter`, `injectContent`, and `resolvePageId` all have `@param` and `@returns` JSDoc annotations
-- `pnpm typecheck` exits 0 with zero errors
-- No `@ts-ignore` directives (only `@ts-expect-error` with comments if truly needed)
-- No `.ts` files created
-- No `tsconfig.json` created (inline tsc flags are sufficient)
-- `pnpm test` still passes
-- `pnpm ci:check` still passes (Biome does not lint JSDoc beyond basic formatting)
-
-## Verification
-
-```sh
-# Confirm @ts-check in all .mjs source files
-grep -rL "@ts-check" scripts/
-# Expected: empty output (all files have it)
-
-# Confirm types.mjs exists
-ls scripts/lib/types.mjs
-
-# Run type check
-pnpm typecheck
-
-# Confirm no .ts files
-find scripts -name "*.ts"
-# Expected: empty
-
-# Confirm no @ts-ignore
-grep -rn "@ts-ignore" scripts/
-# Expected: 0 results
-
-# Run tests
-pnpm test
-```
-
-## Out of scope
-
-- Do not convert any `.mjs` file to `.ts`.
-- Do not create a `tsconfig.json` (inline flags are sufficient for this repo size).
-- Do not add `strict: false` to work around errors — fix the underlying issues.
-- Do not add `@ts-ignore` directives.
-- Do not annotate test files (`*.test.mjs`) — they are not type-checked by the command in step 2.
-- Do not annotate the Biome config (`biome.json`) or GitHub Actions workflow — they are not JS files.
-
-## Follow-ups
-
-- Create a `tsconfig.json` if the inline `tsc` flags line grows longer than ~120 chars or if a second `checkJs` target (e.g. test files) is added.
-- Add `typecheck` to the CI workflow (`.github/workflows/ci.yml`) as a fourth step after `pnpm test` — deferred because spec 10 (CI) ships before spec 14, so this requires updating the workflow file.
-- Once all type errors are resolved and the codebase is stable under `--strict`, consider tightening further with `--noUncheckedIndexedAccess` to catch array index access (`args[0]` returning `string | undefined`).
diff --git a/apps/claude-code/unic-confluence/docs/plans/15-clierror-class.md b/apps/claude-code/unic-confluence/docs/plans/15-clierror-class.md
deleted file mode 100644
index 426001db..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/15-clierror-class.md
+++ /dev/null
@@ -1,443 +0,0 @@
-# 15. Introduce `CliError` Class — Replace `console.error` + `process.exit` Pairs
-**Status: done — 2026-04-24**
-
-**Priority:** P2
-**Effort:** M
-**Depends on:** 09 (unit tests must exist to catch regressions when error-path behaviour changes)
-**Touches:** `scripts/push-to-confluence.mjs`, `scripts/lib/errors.mjs` (new file)
-
-## Context
-
-`scripts/push-to-confluence.mjs` contains approximately 20 sites where error handling is expressed as `console.error(msg); process.exit(1)`. This pattern has two problems. First, it is untestable: any test that calls a function containing `process.exit` will kill the test process unless the caller mocks `process.exit` globally — a brittle technique that leaks state between tests. Second, it scatters output and exit-code logic across the file, making it impossible to change the exit-code strategy (e.g. add structured error output, or return exit code 2 for configuration errors vs. 1 for runtime errors) without touching every site. Replacing the pattern with a `CliError` class centralises both concerns: callers `throw new CliError(msg)` and a single `try/catch` in `main()` handles `console.error` and `process.exit`. Tests can now assert `throws CliError` without mocking the process.
-
-The new class lives in `scripts/lib/errors.mjs` to keep the main file clean. The `httpsRequest` helper's `reject(new Error(...))` pattern is intentionally left unchanged — those are network-layer errors that propagate to callers via promise rejection and are already handled by dedicated `catch` blocks.
-
-## Current behaviour
-
-All error paths in `scripts/push-to-confluence.mjs` call `console.error` then `process.exit(1)` directly. Full list of sites (file line numbers from current source):
-
-| Line(s) | Context | Message |
-|---|---|---|
-| 54–55 | `loadCredentials()` | `"Run \`npm run confluence -- --setup\` to configure credentials"` |
-| 107–108 | `handleHttpError()` | Various HTTP error messages, then `process.exit(1)` |
-| 133–136 | `injectContent()` strategy 1 | `"Found [AUTO_INSERT_START] without a matching [AUTO_INSERT_END]…"` |
-| 153–156 | `injectContent()` strategy 1 label mismatch | `"Marker label mismatch on page…"` |
-| 179–181 | `injectContent()` strategy 2 | `"Found md-start anchor without md-end (or vice versa)…"` |
-| 210–211 | `resolvePageId()` numeric path | `"Invalid page ID: ${arg}"` |
-| 219–221 | `resolvePageId()` key path — file missing | `"confluence-pages.json not found…"` |
-| 229–230 | `resolvePageId()` key path — bad JSON | `"invalid JSON in confluence-pages.json…"` |
-| 237–239 | `resolvePageId()` key path — key absent | `"'${arg}' not found in confluence-pages.json…"` |
-| 244–247 | `resolvePageId()` key path — invalid ID | `"Invalid page ID for key '${arg}'…"` |
-| 258–260 | `runVerify()` — file missing | `"confluence-pages.json not found…"` |
-| 263–265 | `runVerify()` — bad JSON | `"invalid JSON in confluence-pages.json…"` |
-| 306 | `runVerify()` exit | `process.exit(hasErrors ? 1 : 0)` |
-| 313–315 | `runSetup()` — not a TTY | `"--setup requires an interactive terminal"` |
-| 358–360 | `runSetup()` — missing fields | `"Username and API token are required."` |
-| 401–402 | `main()` — insufficient args | `"Usage: npm run confluence -- {pageId} {file.md}"` |
-| 410–411 | `main()` — file not found | `"File not found: ${filePath}"` |
-| 415–416 | `main()` — file too large | `"File too large for Confluence API…"` |
-| 425–427 | `main()` — empty HTML | `"Markdown converted to empty HTML…"` |
-| 439–440 | `main()` GET catch | `err.message` |
-| 450–453 | `main()` — JSON parse failure | `"Unexpected response from Confluence…"` |
-| 458–460 | `main()` — no version | `"Could not read page version…"` |
-| 481–482 | `main()` PUT catch | `err.message` |
-
-None of these are catchable by callers without a `process.exit` mock.
-
-## Target behaviour
-
-After this change:
-
-1. A new file `scripts/lib/errors.mjs` exports `CliError extends Error` with `exitCode`.
-2. Every `console.error(msg); process.exit(1)` pair in `push-to-confluence.mjs` becomes `throw new CliError(msg)`.
-3. `handleHttpError` throws `new CliError(msg)` instead of calling `process.exit(1)`.
-4. `runVerify()`'s terminal `process.exit(hasErrors ? 1 : 0)` becomes `throw new CliError("One or more pages failed verification", hasErrors ? 1 : 0)` when `hasErrors` is true, and simply `return` (exit 0) when all pages passed.
-5. `main()` wraps its body in a `try/catch` that catches `CliError`, calls `console.error(err.message)`, and calls `process.exit(err.exitCode)`. All other errors are re-thrown so they become unhandled-exception crashes (useful for unexpected bugs).
-6. `httpsRequest`'s `reject(new Error(...))` is unchanged — it is a promise rejection, not a direct exit, and callers already handle it in their own `catch` blocks.
-7. Tests written under spec 09 can now use `assert.throws(() => fn(), CliError)` without mocking the process.
-
-The external observable behaviour is identical: error messages still go to stderr, the process still exits 1 on error, exit 0 on success.
-
-## Implementation steps
-
-### Step 1 — Create `scripts/lib/errors.mjs`
-
-Create a new file at `scripts/lib/errors.mjs` (create the `scripts/lib/` directory if it does not exist):
-
-```js
-// SPDX-License-Identifier: LGPL-3.0-or-later
-// Copyright © 2026 Unic
-
-/**
- * CliError — thrown by any function in push-to-confluence.mjs when user-facing
- * error conditions are encountered. Caught once in main() which handles
- * console.error output and process.exit.
- *
- * Using a dedicated class (instead of process.exit) makes all error paths
- * testable: callers can assert.throws(fn, CliError) without mocking process.exit.
- */
-export class CliError extends Error {
-	/** @param {string} message @param {number} [exitCode=1] */
-	constructor(message, exitCode = 1) {
-		super(message);
-		this.name = "CliError";
-		this.exitCode = exitCode;
-	}
-}
-```
-
-### Step 2 — Import `CliError` in `scripts/push-to-confluence.mjs`
-
-At the top of `scripts/push-to-confluence.mjs`, after the existing `import` statements (approximately line 23), add:
-
-```js
-import { CliError } from "./lib/errors.mjs";
-```
-
-### Step 3 — Replace error sites in `loadCredentials()` (lines 54–55)
-
-Before:
-```js
-	console.error("Run `npm run confluence -- --setup` to configure credentials");
-	process.exit(1);
-```
-
-After:
-```js
-	throw new CliError("Run `npm run confluence -- --setup` to configure credentials");
-```
-
-### Step 4 — Update `handleHttpError()` (lines 97–109)
-
-Before:
-```js
-function handleHttpError(status, title) {
-	const messages = {
-		401: "API token rejected — generate a new one at https://id.atlassian.com → Security → API tokens (note: tokens created before 2025 may have expired)",
-		403: "Access denied — check that your API token has permission to read and write this page",
-		404: "Page ID not found — check confluence-pages.json or verify the page still exists",
-		409: `Page was updated by someone else. Retry with: npm run confluence -- ${process.argv[2]} ${process.argv[3]}`,
-	};
-	const msg =
-		messages[status] ??
-		`Unexpected response from Confluence (HTTP ${status}) — check VPN/network and retry`;
-	console.error(msg);
-	process.exit(1);
-}
-```
-
-After:
-```js
-function handleHttpError(status, title) {
-	const messages = {
-		401: "API token rejected — generate a new one at https://id.atlassian.com → Security → API tokens (note: tokens created before 2025 may have expired)",
-		403: "Access denied — check that your API token has permission to read and write this page",
-		404: "Page ID not found — check confluence-pages.json or verify the page still exists",
-		409: `Page was updated by someone else. Retry with: npm run confluence -- ${process.argv[2]} ${process.argv[3]}`,
-	};
-	const msg =
-		messages[status] ??
-		`Unexpected response from Confluence (HTTP ${status}) — check VPN/network and retry`;
-	throw new CliError(msg);
-}
-```
-
-### Step 5 — Replace error sites in `injectContent()` (lines 133–136, 153–156, 179–181)
-
-Three `console.error(...); process.exit(1)` pairs in `injectContent`. Replace each:
-
-**Lines 133–136** (unmatched start marker):
-```js
-// Before
-console.error(
-	`Found [AUTO_INSERT_START] without a matching [AUTO_INSERT_END] on page "${title}" — fix the Confluence page before publishing`,
-);
-process.exit(1);
-
-// After
-throw new CliError(
-	`Found [AUTO_INSERT_START] without a matching [AUTO_INSERT_END] on page "${title}" — fix the Confluence page before publishing`,
-);
-```
-
-**Lines 153–156** (label mismatch):
-```js
-// Before
-console.error(
-	`Marker label mismatch on page "${title}": [AUTO_INSERT_START:${startLabel}] has no matching [AUTO_INSERT_END:${startLabel}] — fix the Confluence page before publishing`,
-);
-process.exit(1);
-
-// After
-throw new CliError(
-	`Marker label mismatch on page "${title}": [AUTO_INSERT_START:${startLabel}] has no matching [AUTO_INSERT_END:${startLabel}] — fix the Confluence page before publishing`,
-);
-```
-
-**Lines 179–181** (unmatched anchor macro):
-```js
-// Before
-console.error(
-	`Found md-start anchor without md-end (or vice versa) on page "${title}" — fix the Confluence page before publishing`,
-);
-process.exit(1);
-
-// After
-throw new CliError(
-	`Found md-start anchor without md-end (or vice versa) on page "${title}" — fix the Confluence page before publishing`,
-);
-```
-
-### Step 6 — Replace error sites in `resolvePageId()` (lines 210–249)
-
-Five sites. Replace each `console.error(msg); process.exit(1)` pair with `throw new CliError(msg)`:
-
-```js
-// Line 210-211 — invalid numeric ID
-throw new CliError(`Invalid page ID: ${arg}`);
-
-// Line 219-221 — file missing
-throw new CliError(
-	"confluence-pages.json not found — create it or pass a page ID directly",
-);
-
-// Line 229-230 — bad JSON
-throw new CliError("invalid JSON in confluence-pages.json — check syntax");
-
-// Line 237-239 — key not found
-throw new CliError(
-	`'${arg}' not found in confluence-pages.json — available keys: ${keys}`,
-);
-
-// Line 244-247 — invalid ID for key
-throw new CliError(
-	`Invalid page ID for key '${arg}': ${id} — must be a positive integer`,
-);
-```
-
-### Step 7 — Replace error sites in `runVerify()` (lines 258–306)
-
-Two `console.error + process.exit` pairs at the top of the function, plus the terminal `process.exit` at line 306.
-
-**Lines 258–260** (file missing):
-```js
-throw new CliError("confluence-pages.json not found — create it first");
-```
-
-**Lines 263–265** (bad JSON):
-```js
-throw new CliError("invalid JSON in confluence-pages.json — check syntax");
-```
-
-**Line 306** (terminal exit — only the failure case needs a throw; success case returns normally):
-```js
-// Before
-process.exit(hasErrors ? 1 : 0);
-
-// After
-if (hasErrors) {
-	throw new CliError("One or more pages failed verification — see errors above", 1);
-}
-// success: fall through; main() handles exit 0 after await runVerify()
-```
-
-### Step 8 — Replace error sites in `runSetup()` (lines 313–315, 358–360)
-
-**Lines 313–315** (not a TTY):
-```js
-throw new CliError("--setup requires an interactive terminal");
-```
-
-**Lines 358–360** (missing fields):
-```js
-throw new CliError("Username and API token are required.");
-```
-
-Note: `runSetup()` also calls `process.exit(0)` on user abort (line 339) — that is a normal early exit, not an error. Leave it as `process.exit(0)`.
-
-### Step 9 — Replace error sites in `main()` (lines 401–482)
-
-Replace eight sites in the main publish flow:
-
-```js
-// Line 401-402 — insufficient args
-throw new CliError("Usage: npm run confluence -- {pageId} {file.md}");
-
-// Line 410-411 — file not found
-throw new CliError(`File not found: ${filePath}`);
-
-// Line 415-416 — file too large
-throw new CliError("File too large for Confluence API — split the document");
-
-// Line 425-427 — empty HTML
-throw new CliError(
-	"Markdown converted to empty HTML — check the source file is not empty",
-);
-
-// Line 439-440 — GET catch block (err.message from httpsRequest rejection)
-// Before:
-catch (err) {
-	console.error(err.message);
-	process.exit(1);
-}
-// After:
-catch (err) {
-	throw new CliError(err.message);
-}
-
-// Line 450-453 — JSON parse failure on GET body
-throw new CliError(
-	"Unexpected response from Confluence — check VPN/network and retry",
-);
-
-// Line 458-460 — no version in response
-throw new CliError(
-	"Could not read page version from Confluence response — retry or contact support",
-);
-
-// Line 481-482 — PUT catch block
-// Before:
-catch (err) {
-	console.error(err.message);
-	process.exit(1);
-}
-// After:
-catch (err) {
-	throw new CliError(err.message);
-}
-```
-
-### Step 10 — Wrap `main()` body in `try/catch`
-
-File: `scripts/push-to-confluence.mjs`, `main()` starting at line 388.
-
-Before (the current overall structure):
-```js
-async function main() {
-	const args = process.argv.slice(2);
-	// ... all logic ...
-	console.log(`✓ Published "${title}" to Confluence (version ${version + 1})`);
-}
-
-main();
-```
-
-After:
-```js
-async function main() {
-	try {
-		const args = process.argv.slice(2);
-		// ... all logic (unchanged) ...
-		console.log(`✓ Published "${title}" to Confluence (version ${version + 1})`);
-	} catch (err) {
-		if (err instanceof CliError) {
-			console.error(err.message);
-			process.exit(err.exitCode);
-		}
-		throw err; // re-throw unexpected errors as unhandled exceptions
-	}
-}
-
-main();
-```
-
-The `try` wraps the entire existing body of `main()`. The `catch` only handles `CliError`; anything else (e.g. a `TypeError` from a bug) propagates as an unhandled rejection and prints a full stack trace — which is correct debugging behaviour.
-
-### Step 11 — Update tests (spec 09 test files)
-
-In any test files under `scripts/lib/*.test.mjs`, replace `process.exit` spy patterns with direct assertion:
-
-```js
-// Before (mock-based, brittle)
-const exitSpy = sinon.stub(process, "exit");
-resolvePageId("-1");
-assert.ok(exitSpy.calledWith(1));
-exitSpy.restore();
-
-// After (clean, no mocking)
-import { CliError } from "./errors.mjs";
-assert.throws(() => resolvePageId("-1"), CliError);
-```
-
-> Ralph: if spec 09 tests have not yet landed when this spec is implemented, skip this step and note it in the follow-ups.
-
-## Test cases
-
-| Input | Function under test | Expected result |
-|---|---|---|
-| No credentials file, no env vars | `loadCredentials()` | Throws `CliError` with message `"Run \`npm run confluence -- --setup\` to configure credentials"` |
-| `resolvePageId("0")` | `resolvePageId()` | Throws `CliError` with message containing `"Invalid page ID: 0"` |
-| `resolvePageId("my-key")` without `confluence-pages.json` | `resolvePageId()` | Throws `CliError` with message containing `"confluence-pages.json not found"` |
-| `resolvePageId("missing-key")` with valid `confluence-pages.json` | `resolvePageId()` | Throws `CliError` with message containing `"not found in confluence-pages.json"` |
-| `injectContent` with `[AUTO_INSERT_START: x]` but no `[AUTO_INSERT_END: x]` | `injectContent()` | Throws `CliError` with message containing `"without a matching [AUTO_INSERT_END]"` |
-| `main()` receives a `CliError` thrown from deep in the call stack | `main()` `catch` block | `console.error` called once with `err.message`; `process.exit` called once with `err.exitCode` (1) |
-| `main()` receives an unexpected `TypeError` (bug) | `main()` `catch` block | Error is re-thrown; Node prints stack trace; process exits with non-zero code |
-| `runVerify()` with one failing page | `runVerify()` | Throws `CliError` with `exitCode: 1`; caught by `main()`'s try/catch |
-| `runVerify()` with all pages passing | `runVerify()` | Returns normally; `main()` falls through to exit 0 |
-
-## Acceptance criteria
-
-- [ ] `scripts/lib/errors.mjs` exists and exports `CliError extends Error` with `name`, `message`, and `exitCode` properties.
-- [ ] `CliError` constructor defaults `exitCode` to `1`.
-- [ ] Every `console.error(msg); process.exit(1)` pair in `push-to-confluence.mjs` is replaced with `throw new CliError(msg)`.
-- [ ] `handleHttpError` throws `new CliError(msg)` instead of calling `process.exit(1)`.
-- [ ] `runVerify()`'s terminal `process.exit(hasErrors ? 1 : 0)` is replaced: success path returns normally; failure path throws `CliError`.
-- [ ] `main()` has a single top-level `try/catch` that handles `CliError` with `console.error` + `process.exit`, and re-throws all other errors.
-- [ ] `httpsRequest`'s `reject(new Error(...))` is unchanged.
-- [ ] `runSetup()`'s `process.exit(0)` on user abort (line 339) is unchanged.
-- [ ] Observable behaviour is identical: same messages to stderr, same exit codes.
-- [ ] `grep 'process\.exit(1)' scripts/push-to-confluence.mjs` returns zero matches (all replaced).
-- [ ] `grep 'process\.exit' scripts/push-to-confluence.mjs` returns only the two legitimate exits: the `process.exit(0)` in `runSetup()` abort, and the one inside `main()`'s `catch` block.
-- [ ] Lint passes (`pnpm run lint` if configured).
-
-## Verification
-
-```sh
-# 1. Confirm CliError file was created
-ls scripts/lib/errors.mjs
-
-# 2. Confirm CliError is imported in the main script
-grep "CliError" scripts/push-to-confluence.mjs | head -3
-
-# 3. Confirm no bare process.exit(1) calls remain (should be 0 matches)
-grep -n 'process\.exit(1)' scripts/push-to-confluence.mjs
-# Expected: no output
-
-# 4. Confirm only expected process.exit calls remain
-grep -n 'process\.exit' scripts/push-to-confluence.mjs
-# Expected: two lines — the abort in runSetup() (exit 0) and main()'s catch (err.exitCode)
-
-# 5. Confirm CliError is thrown in resolvePageId
-grep -n 'throw new CliError' scripts/push-to-confluence.mjs | wc -l
-# Expected: ~20 lines
-
-# 6. Confirm main() has a try/catch wrapping its body
-grep -n 'instanceof CliError' scripts/push-to-confluence.mjs
-# Expected: 1 match in main()
-
-# 7. Smoke test — no credentials
-unset CONFLUENCE_URL CONFLUENCE_USER CONFLUENCE_TOKEN
-node scripts/push-to-confluence.mjs some-key docs/plans/15-clierror-class.md; echo "Exit: $?"
-# Expected: stderr "Run `npm run confluence -- --setup`...", exit 1
-
-# 8. Smoke test — insufficient args
-node scripts/push-to-confluence.mjs; echo "Exit: $?"
-# Expected: stderr "Usage: npm run confluence...", exit 1
-```
-
-## Out of scope
-
-- Do not add `exitCode: 2` or other non-1 exit codes (beyond the `runVerify` false case). A single `exitCode` field is added for structural correctness but the values remain 1 in all new paths.
-- Do not change `httpsRequest`'s `reject(new Error(...))` pattern.
-- Do not add structured JSON error output.
-- Do not create a top-level `errors.mjs` at the repo root — the file goes in `scripts/lib/`.
-- Do not modify the slash command (`commands/unic-confluence.md`).
-- No version bump required for this change.
-
-## Follow-ups
-
-- If spec 09 (unit tests) has not yet landed: once it does, update the tests to use `assert.throws(fn, CliError)` as described in Step 11.
-- Consider adding `exitCode: 2` for configuration errors (missing credentials, missing `confluence-pages.json`) vs. `exitCode: 1` for runtime errors — deferred until there is a demonstrated need.
-- `handleHttpError` currently uses `process.argv[2]` / `process.argv[3]` to construct the 409 retry hint. After this change the function signature could accept the page arg directly — deferred.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/16-claude-plugin-readme.md b/apps/claude-code/unic-confluence/docs/plans/16-claude-plugin-readme.md
deleted file mode 100644
index 6cf319ee..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/16-claude-plugin-readme.md
+++ /dev/null
@@ -1,239 +0,0 @@
-# 16. Add `.claude-plugin/README.md` for Marketplace Documentation
-
-**Status: done — 2026-04-24**
-
-**Priority:** P2
-**Effort:** S
-**Depends on:** none
-**Touches:** `.claude-plugin/README.md` (new file)
-
-## Context
-
-The Claude Code plugin marketplace renders `.claude-plugin/README.md` as the primary documentation surface visible to users browsing or installing the plugin. This surface is separate from the repo-root `README.md`, which is aimed at npm consumers who install the package into their own repos. Currently `.claude-plugin/README.md` does not exist, so marketplace users see no documentation when they inspect the `unic-confluence` plugin — no setup instructions, no usage examples, no marker syntax explanation. A user who discovers the plugin through the marketplace has no in-product path to configure it correctly. This spec creates that file. The content is intentionally scoped to plugin consumers (people using `/unic-confluence` in Claude Code) and does not duplicate the npm-consumer instructions in the root README.
-
-## Current behaviour
-
-```sh
-ls .claude-plugin/
-# plugin.json  marketplace.json
-# (no README.md)
-```
-
-Marketplace users who run `claude plugins info unic-confluence` or browse the marketplace listing see no documentation beyond the `description` field in `marketplace.json` (`"Publish markdown files to Confluence pages via the Confluence v2 API."`).
-
-## Target behaviour
-
-`.claude-plugin/README.md` exists and is shown in the marketplace UI. It covers:
-
-1. What the plugin does (one line).
-2. How to install it via the marketplace (two commands: marketplace add, then install).
-3. One-time setup: run `--setup`, create `confluence-pages.json`, add markers to the Confluence page.
-4. How to use the slash command (`/unic-confluence`).
-5. How to verify page IDs (`--verify`).
-6. How to update the plugin.
-
-The file avoids npm-consumer instructions (no `npm install`, no `"confluence"` script, no `package.json` snippets).
-
-## Implementation steps
-
-### Step 1 — Create `.claude-plugin/README.md`
-
-Create the file at `.claude-plugin/README.md` with the following content:
-
-```markdown
-# unic-confluence
-
-Publish local Markdown files to Confluence pages directly from Claude Code.
-
-## Install
-
-```sh
-# Register the Unic plugin marketplace (once per machine)
-claude plugins marketplace add unic https://raw.githubusercontent.com/unic/unic-claude-code-confluence/main/.claude-plugin/marketplace.json
-
-# Install the plugin
-claude plugins install unic-confluence@unic
-```
-
-## Setup (once per repo)
-
-### 1. Configure credentials
-
-Run the interactive setup wizard. It stores credentials in `~/.unic-confluence.json` (chmod 600).
-
-```sh
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --setup
-```
-
-You will be prompted for:
-
-- **Confluence URL** — e.g. `https://yourcompany.atlassian.net` (default pre-filled for Unic: `https://uniccom.atlassian.net`)
-- **Email** — the email address associated with your Atlassian account
-- **API token** — generate one at https://id.atlassian.com → Security → API tokens
-
-> Note: API tokens created before 2025 may have expired. Generate a new one if setup fails with a 401 error.
-
-Credentials can also be provided as environment variables per-run:
-`CONFLUENCE_URL`, `CONFLUENCE_USER`, `CONFLUENCE_TOKEN`
-
-### 2. Create `confluence-pages.json` at your repo root
-
-Map human-readable keys to Confluence page IDs. The page ID is the number in the Confluence URL: `.../pages/804848595/My+Page`.
-
-```json
-{
-  "_comment": "Map human-readable keys to Confluence page IDs",
-  "my-spec": 804848595,
-  "architecture": 804848601
-}
-```
-
-### 3. Add markers to your Confluence page
-
-In the Confluence page editor, place these markers where the Markdown content should be injected. Everything between the markers is replaced on each publish.
-
-```
-[AUTO_INSERT_START: my-spec]
-
-(Claude Code will inject content here)
-
-[AUTO_INSERT_END: my-spec]
-```
-
-**Rules:**
-- The label (e.g. `my-spec`) must match exactly between START and END — case-sensitive, no extra spaces.
-- The label does not have to match the key in `confluence-pages.json`, but using the same value is recommended for clarity.
-- Each page should have exactly one START/END pair.
-
-If no markers are found, the publish command will refuse to proceed and instruct you to add them. (This is intentional — without markers, repeated publishes would duplicate content.)
-
-## Usage
-
-In Claude Code, invoke the slash command:
-
-```
-/unic-confluence <page-key-or-id> <markdown-file>
-```
-
-**Examples:**
-
-```sh
-# Use a key from confluence-pages.json
-/unic-confluence my-spec docs/features/spec.md
-
-# Use a raw numeric page ID directly (no confluence-pages.json required)
-/unic-confluence 804848595 docs/features/spec.md
-```
-
-The command will:
-1. Probe credentials (lightweight single GET).
-2. Read the Markdown file and convert it to HTML.
-3. Fetch the current page body from Confluence.
-4. Inject the new HTML between the markers.
-5. Publish the updated body via the Confluence v2 API.
-
-## Verify page IDs
-
-Check that all pages in `confluence-pages.json` still exist and are accessible:
-
-```sh
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --verify
-```
-
-Prints a `✅` or `❌` line for each entry. Exits 0 if all pages pass, 1 if any fail.
-
-Useful for CI checks or after a Confluence reorganisation.
-
-## Update
-
-```sh
-claude plugins marketplace update unic
-claude plugins install unic-confluence@unic
-```
-
-## Troubleshooting
-
-| Error | Cause | Fix |
-|---|---|---|
-| `Run --setup to configure credentials` | No credentials file and no env vars | Run `node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --setup` |
-| `API token rejected` | Token expired or wrong | Generate a new token at https://id.atlassian.com |
-| `Page ID not found` | Wrong ID in `confluence-pages.json` | Run `--verify` to audit all IDs |
-| `No [AUTO_INSERT_START:label] markers found` | Markers missing from Confluence page | Add `[AUTO_INSERT_START: label]` / `[AUTO_INSERT_END: label]` to the page in Confluence |
-| `Cannot reach Confluence — check VPN/network` | Not on VPN or network down | Connect to VPN, then retry |
-| `Marker label mismatch` | START and END labels differ | Edit the Confluence page to make labels match exactly |
-```
-
-> Ralph: the marketplace registration URL (`https://raw.githubusercontent.com/...`) is a placeholder. Confirm the correct URL from the repo's `marketplace.json` or the Claude Code plugin documentation before shipping.
-
-### Step 2 — Verify the file is picked up by the plugin manifest
-
-Open `.claude-plugin/plugin.json` and confirm there is no `readme` or `docs` field that explicitly points elsewhere. The default convention for Claude Code plugins is `.claude-plugin/README.md` — no manifest change is required if the convention is followed.
-
-If `plugin.json` has a `readme` key pointing to a different path, update it to `"./README.md"` (relative to `.claude-plugin/`).
-
-## Test cases
-
-| Scenario | Expected result |
-|---|---|
-| `ls .claude-plugin/` after this change | `marketplace.json  plugin.json  README.md` |
-| `cat .claude-plugin/README.md` contains "Install" section | Yes |
-| `cat .claude-plugin/README.md` contains `--setup` instructions | Yes |
-| `cat .claude-plugin/README.md` contains `confluence-pages.json` snippet | Yes |
-| `cat .claude-plugin/README.md` contains `[AUTO_INSERT_START` example | Yes |
-| `cat .claude-plugin/README.md` contains `/unic-confluence` usage | Yes |
-| `cat .claude-plugin/README.md` contains `--verify` | Yes |
-| Root `README.md` is unchanged | Yes — this spec does not touch it |
-| `.claude-plugin/plugin.json` is unchanged | Yes — no manifest change needed |
-
-## Acceptance criteria
-
-- [ ] `.claude-plugin/README.md` exists.
-- [ ] File contains a top-level `# unic-confluence` heading.
-- [ ] File contains an Install section with both `marketplace add` and `plugins install` commands.
-- [ ] File contains Setup instructions covering credentials (`--setup`), `confluence-pages.json`, and markers.
-- [ ] File contains a Usage section with the `/unic-confluence` slash command syntax and at least one example.
-- [ ] File contains a `--verify` section.
-- [ ] File contains an Update section.
-- [ ] File does NOT contain `npm install` or `package.json` script instructions (those belong in the root README).
-- [ ] Root `README.md` is unchanged.
-- [ ] `.claude-plugin/plugin.json` is unchanged (unless it had an explicit `readme` key pointing elsewhere).
-
-## Verification
-
-```sh
-# 1. Confirm file exists
-ls -la .claude-plugin/README.md
-
-# 2. Confirm key sections are present
-grep -c "## Install\|## Setup\|## Usage\|## Verify\|## Update\|## Troubleshooting" .claude-plugin/README.md
-# Expected: 6
-
-# 3. Confirm no npm-consumer content leaked in
-grep "npm install\|package\.json" .claude-plugin/README.md
-# Expected: no output
-
-# 4. Confirm markers example is present
-grep "AUTO_INSERT_START" .claude-plugin/README.md
-# Expected: at least one match
-
-# 5. Confirm root README is untouched
-git diff README.md
-# Expected: no changes
-```
-
-## Out of scope
-
-- Do not modify the root `README.md`.
-- Do not add `npm install` or consuming-repo setup instructions to this file — those belong in the root README.
-- Do not duplicate the full marker syntax specification here; a working example is enough.
-- Do not add a `CHANGELOG` or release notes to this file.
-- Do not add badges (CI status, npm version, etc.).
-- No version bump to `plugin.json` or `marketplace.json` for this documentation-only change.
-
-## Follow-ups
-
-- Once the marketplace registration URL is confirmed, replace the placeholder URL in the Install section.
-- If Claude Code's marketplace adds support for screenshots or video previews in plugin READMEs, consider adding a screenshot of a successful publish.
-- After spec 03 (`--replace-all`) lands, add a note to the Troubleshooting table for the "No markers found" error that mentions `--replace-all`.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/17-claude-md-do-not-add-section.md b/apps/claude-code/unic-confluence/docs/plans/17-claude-md-do-not-add-section.md
deleted file mode 100644
index 2812dba2..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/17-claude-md-do-not-add-section.md
+++ /dev/null
@@ -1,139 +0,0 @@
-# 17. Add "Do Not Add" Section to `CLAUDE.md`
-
-**Status: done — 2026-04-24**
-
-**Priority:** P2
-**Effort:** S
-**Depends on:** none
-**Touches:** `CLAUDE.md`
-
-## Context
-
-The repo's `CLAUDE.md` accurately documents what the codebase does and how it works, but says nothing about what is intentionally out of scope. Without an explicit boundary, AI contributors (and new human contributors) are likely to propose or implement features that sound reasonable on the surface but would undermine the plugin's deliberate simplicity. Examples of this drift that have already appeared in other Confluence tool projects: image upload pipelines, create-page support, multi-instance config schemas, MCP server wrappers. Each of these is a plausible extension but each also significantly expands the attack surface, the configuration schema, and the maintenance burden. A `## Do not add` section in `CLAUDE.md` is the correct place to document these decisions because it is the first file Claude reads when starting a session in this repo. The section converts implicit architectural decisions into explicit, machine-readable policy.
-
-## Current behaviour
-
-`CLAUDE.md` ends after the `## Naming convention` section:
-
-```markdown
-## Naming convention
-
-| Surface           | Value                        |
-| ----------------- | ---------------------------- |
-| GitHub repo       | `unic-claude-code-<service>` |
-| Plugin identifier | `unic-<service>`             |
-| npm package name  | `unic-<service>`             |
-```
-
-(End of file — no "Do not add" or "Out of scope" section exists.)
-
-To verify:
-```sh
-tail -15 CLAUDE.md
-# Shows naming convention table, then EOF
-```
-
-## Target behaviour
-
-`CLAUDE.md` has a new `## Do not add` section appended after `## Naming convention`. The section lists eight categories of features that are explicitly out of scope, each with a concise rationale. The final paragraph gives a decision rule for anything not on the list.
-
-The existing content of `CLAUDE.md` is untouched.
-
-## Implementation steps
-
-### Step 1 — Append section to `CLAUDE.md`
-
-Open `CLAUDE.md` and append the following block after the last line of the `## Naming convention` table:
-
-```markdown
-
-## Do not add
-
-The following are explicitly out of scope for this plugin. Do not implement them without first opening a GitHub issue and getting explicit sign-off from the maintainer:
-
-- **Image upload / attachments** — walking the Markdown AST to find local image references and uploading them via `/wiki/rest/api/content/{id}/child/attachment` is significant work that requires a new CLI subcommand, a new content-negotiation path, and multi-part form handling. Defer until a user explicitly requests it with a concrete use case.
-
-- **Create-page support** — the script only updates existing pages. Adding `POST /wiki/api/v2/pages` with `spaceId` + `parentId` requires a schema change to `confluence-pages.json` (value becomes an object, not just a page ID integer), complicating every code path that reads the file. Defer unless there is real demand.
-
-- **Multi-space or cross-instance publishing** — `confluence-pages.json` maps keys to page IDs, and page IDs are unique per Confluence instance. Supporting multiple Confluence instances would require a different config schema (e.g. a `baseUrl` field per entry) and credential routing logic. Out of scope.
-
-- **MCP server** — there is no benefit to wrapping this plugin's functionality in a Model Context Protocol server. The slash command is the correct and sufficient surface. An MCP server would add a process lifecycle, a transport layer, and a versioned protocol schema for zero user-visible benefit.
-
-- **Agents or sub-agents** — the publish task is a deterministic one-shot sequence: read file → convert Markdown → GET page → inject → PUT page. There is no branching, no tool-selection, and no iteration. Agent autonomy adds complexity without value here.
-
-- **Recursive directory publishing** — publishing all Markdown files under a directory tree in one command (e.g. `node push-to-confluence.mjs docs/`) requires mapping every file to a page ID, handling partial failures, and defining rollback semantics. The complexity grows faster than the value. Publish one file at a time.
-
-- **Changesets or release-please** — the `sync-version.mjs` script (spec 06) is sufficient for one package with two version fields. Do not add a release-management framework (Changesets, release-please, semantic-release) — the overhead exceeds the benefit for a single-file plugin.
-
-- **Watch mode / file-watcher** — a flag like `--watch` that re-publishes on file change is not appropriate for a Confluence publishing tool. Confluence is not a live preview target; each publish increments the page version and creates a revision in Confluence's history. Accidental rapid publishes would pollute the revision history.
-
-When in doubt: if the feature is not in the existing `scripts/push-to-confluence.mjs` command set and is not listed in an open `docs/plans/` spec, it is out of scope. Open a GitHub issue before starting implementation.
-```
-
-No other changes to `CLAUDE.md`.
-
-## Test cases
-
-| Check | Expected result |
-|---|---|
-| `grep "Do not add" CLAUDE.md` | At least one match |
-| `grep "Image upload" CLAUDE.md` | One match |
-| `grep "Create-page" CLAUDE.md` | One match |
-| `grep "MCP server" CLAUDE.md` | One match |
-| `grep "Agents" CLAUDE.md` | One match |
-| `grep "Recursive directory" CLAUDE.md` | One match |
-| `grep "Changesets" CLAUDE.md` | One match |
-| `grep "Watch mode" CLAUDE.md` | One match |
-| Existing sections (`## What this repo is`, `## Key files`, etc.) still present | Yes |
-| Naming convention table still present | Yes |
-| No new files created | Yes |
-
-## Acceptance criteria
-
-- [ ] `CLAUDE.md` contains a `## Do not add` section.
-- [ ] The section appears after `## Naming convention` (at the end of the file).
-- [ ] The section covers all eight categories: image upload, create-page, multi-instance, MCP server, agents, recursive directory publishing, changesets, and watch mode.
-- [ ] Each category has a one-sentence rationale (not just a label).
-- [ ] The section ends with a decision rule for unlisted features.
-- [ ] The existing `CLAUDE.md` sections are unchanged.
-- [ ] No new files are created by this spec.
-
-## Verification
-
-```sh
-# 1. Confirm "Do not add" section exists
-grep -n "## Do not add" CLAUDE.md
-# Expected: one match near the end of the file
-
-# 2. Confirm all eight categories are present
-grep -c "Image upload\|Create-page\|Multi-space\|MCP server\|Agents\|Recursive directory\|Changesets\|Watch mode" CLAUDE.md
-# Expected: 8
-
-# 3. Confirm existing sections are still present
-grep "## What this repo is\|## Key files\|## Running\|## Content injection\|## Credentials\|## Plugin versioning\|## Naming convention" CLAUDE.md | wc -l
-# Expected: 7
-
-# 4. Confirm "Do not add" is the last section
-tail -30 CLAUDE.md | grep "^## "
-# Expected: "## Do not add" is the only heading in the last 30 lines
-
-# 5. No uncommitted changes to other files
-git diff --name-only
-# Expected: only CLAUDE.md
-```
-
-## Out of scope
-
-- Do not modify any existing `CLAUDE.md` sections.
-- Do not move implementation guidance from `docs/plans/` into `CLAUDE.md` — specs belong in `docs/plans/`.
-- Do not add a "Do add" or "Planned features" section — that information lives in `docs/plans/`.
-- Do not add links to GitHub issues (none exist yet for these items).
-- No changes to `scripts/`, `commands/`, `.claude-plugin/`, or any other file.
-
-## Follow-ups
-
-- After spec 06 (version sync script) lands, remove the `sync-version.mjs` forward-reference from the Changesets bullet and replace with a direct description of what that script does.
-- If any of the eight "Do not add" items is eventually accepted and specced out, remove it from this section and add a note: "This item was deferred — see spec NN."
-- Consider adding a `## Canonical patterns` section to `CLAUDE.md` documenting the positive analogs to the "do not add" list (e.g. "add new CLI subcommands as `if (args[0] === "--foo")` branches in `main()`").
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/18-parallelise-runverify.md b/apps/claude-code/unic-confluence/docs/plans/18-parallelise-runverify.md
deleted file mode 100644
index bc77c01a..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/18-parallelise-runverify.md
+++ /dev/null
@@ -1,272 +0,0 @@
-# 18. Parallelise `runVerify()` with `Promise.all`
-
-**Status: done — 2026-04-24**
-
-**Priority:** P2
-**Effort:** S
-**Depends on:** spec 01 outcome — if `--verify` was fully deleted by spec 01, this spec is superseded. Spec 15 (`CliError`) is a soft dependency: if it has landed, replace `process.exit(1)` calls below with `throw new CliError(msg)`.
-**Touches:** `scripts/push-to-confluence.mjs`
-
-## Context
-
-`runVerify()` (lines 256–306 of `scripts/push-to-confluence.mjs`) performs one HTTP GET per entry in `confluence-pages.json` using a sequential `for...of` loop. Each GET waits for the previous one to complete before starting. For a team that accumulates 10 mapped pages, a verify run makes 10 serial requests; at ~300ms per request over VPN, that is 3 seconds minimum. At 20 pages it is 6 seconds. The requests are independent — no result depends on another — so there is no reason for the serialisation. Replacing the loop with `Promise.all` cuts verify time to approximately one round-trip regardless of page count. The change also restructures the function slightly: ID validation (synchronous, no I/O) is separated from the HTTP phase so that invalid IDs are reported immediately without firing any network requests.
-
-**Prerequisite check:** Read spec 01 (`docs/plans/01-drop-verify-from-slash-command.md`). Spec 01 removes `--verify` from the slash command's pre-flight step but explicitly states "The existing `--verify` subcommand is left unchanged — it remains useful as an interactive diagnostic tool." If the version of spec 01 that was actually implemented kept `runVerify()` in place (the intended outcome), proceed with this spec. If for some reason `runVerify()` was deleted entirely, mark this spec as superseded by spec 01 and skip it.
-
-## Current behaviour
-
-`scripts/push-to-confluence.mjs`, lines 256–307:
-
-```js
-async function runVerify() {
-	const pagesPath = path.join(process.cwd(), "confluence-pages.json");
-	if (!existsSync(pagesPath)) {
-		console.error("confluence-pages.json not found — create it first");
-		process.exit(1);
-	}
-
-	let pages;
-	try {
-		pages = JSON.parse(readFileSync(pagesPath, "utf8"));
-	} catch {
-		console.error("invalid JSON in confluence-pages.json — check syntax");
-		process.exit(1);
-	}
-
-	const pageKeys = Object.keys(pages).filter((k) => k !== "_comment");
-	console.log(`Verifying ${pageKeys.length} page(s): ${pageKeys.join(", ")}`);
-
-	const { url: baseUrl, username, token } = loadCredentials();
-	const authHeader = makeBasicAuth(username, token);
-
-	let hasErrors = false;
-	for (const [key, id] of Object.entries(pages)) {
-		if (key === "_comment") continue;
-		if (!Number.isInteger(id) || id <= 0) {
-			console.error(
-				`❌ ${key}: Invalid page ID in confluence-pages.json: ${JSON.stringify(id)} — must be a positive integer`,
-			);
-			hasErrors = true;
-			continue;
-		}
-		const getUrl = `${baseUrl.replace(/\/$/, "")}/wiki/api/v2/pages/${id}?body-format=storage`;
-		try {
-			const res = await httpsRequest("GET", getUrl, authHeader);
-			if (res.status === 200) {
-				const pageData = JSON.parse(res.body);
-				console.log(`✅ ${key} (${id}): ${pageData.title}`);
-			} else if (res.status === 404) {
-				console.error(`❌ ${key} (${id}): Page not found — 404`);
-				hasErrors = true;
-			} else {
-				console.error(`⚠️  ${key} (${id}): Unexpected HTTP ${res.status}`);
-				hasErrors = true;
-			}
-		} catch {
-			console.error(`⚠️  ${key} (${id}): Network error`);
-			hasErrors = true;
-		}
-	}
-
-	process.exit(hasErrors ? 1 : 0);
-}
-```
-
-Performance characteristics: N entries → N serial HTTP GETs. Each GET blocks the next.
-
-## Target behaviour
-
-`runVerify()` fires all HTTP GETs concurrently via `Promise.all`. The function:
-
-1. Validates the `confluence-pages.json` file (sync — unchanged).
-2. Validates all page IDs synchronously before making any network requests. If any ID is invalid, prints all invalid-ID errors immediately and exits 1 without firing any GETs.
-3. Fires all GETs in parallel with `Promise.all`.
-4. After all settle, prints results in the original key order (order preserved because `Promise.all` preserves order).
-5. Exits 0 if all passed, 1 if any failed.
-
-The output format is identical — same `✅ / ❌ / ⚠️` lines, same stderr/stdout distinction. The only observable difference is that output lines may arrive in a different order than file order if some requests resolve before others, but since all results are collected before printing (see implementation), the printed order matches the `confluence-pages.json` key order.
-
-If spec 15 (`CliError`) has landed: replace `process.exit(1)` in the early-exit paths with `throw new CliError(msg)`.
-
-## Implementation steps
-
-### Step 1 — Replace the `for...of` loop in `runVerify()`
-
-Locate the `for...of` loop starting at approximately line 278:
-
-```js
-	let hasErrors = false;
-	for (const [key, id] of Object.entries(pages)) {
-		if (key === "_comment") continue;
-		// ... validation and await httpsRequest ...
-	}
-
-	process.exit(hasErrors ? 1 : 0);
-```
-
-Replace everything from `let hasErrors = false;` to the end of the function with:
-
-```js
-	const entries = Object.entries(pages).filter(([k]) => k !== "_comment");
-
-	// Phase 1: synchronous ID validation (no network I/O)
-	// Report all invalid IDs before making any HTTP requests.
-	let hasErrors = false;
-	for (const [key, id] of entries) {
-		if (!Number.isInteger(id) || id <= 0) {
-			console.error(
-				`❌ ${key}: Invalid page ID in confluence-pages.json: ${JSON.stringify(id)} — must be a positive integer`,
-			);
-			hasErrors = true;
-		}
-	}
-	if (hasErrors) {
-		process.exit(1);
-		// If spec 15 has landed, use: throw new CliError("Invalid page IDs found — fix confluence-pages.json", 1);
-	}
-
-	// Phase 2: parallel GET requests
-	const results = await Promise.all(
-		entries.map(async ([key, id]) => {
-			const getUrl = `${baseUrl.replace(/\/$/, "")}/wiki/api/v2/pages/${id}?body-format=storage`;
-			try {
-				const res = await httpsRequest("GET", getUrl, authHeader);
-				if (res.status === 200) {
-					const pageData = JSON.parse(res.body);
-					return { ok: true, key, id, title: pageData.title };
-				}
-				return { ok: false, key, id, status: res.status };
-			} catch {
-				return { ok: false, key, id, networkError: true };
-			}
-		}),
-	);
-
-	// Phase 3: collect results and print in key order
-	for (const r of results) {
-		if (r.ok) {
-			console.log(`✅ ${r.key} (${r.id}): ${r.title}`);
-		} else if (r.networkError) {
-			console.error(`⚠️  ${r.key} (${r.id}): Network error`);
-			hasErrors = true;
-		} else if (r.status === 404) {
-			console.error(`❌ ${r.key} (${r.id}): Page not found — 404`);
-			hasErrors = true;
-		} else {
-			console.error(`⚠️  ${r.key} (${r.id}): Unexpected HTTP ${r.status}`);
-			hasErrors = true;
-		}
-	}
-
-	process.exit(hasErrors ? 1 : 0);
-	// If spec 15 has landed, use:
-	// if (hasErrors) throw new CliError("One or more pages failed verification — see errors above", 1);
-	// (and remove process.exit — main()'s catch handles it)
-}
-```
-
-No other changes to `runVerify()`. The file-loading preamble (lines 257–275: `existsSync`, `JSON.parse`, `loadCredentials`, `makeBasicAuth`, `console.log("Verifying...")`) is unchanged.
-
-### Step 2 — Verify no sequential `await` remains inside the verify loop
-
-After the edit, confirm that `runVerify` contains no `await` inside a `for` loop. The only `await` should be `await Promise.all(...)`.
-
-```sh
-# Should show 0 matches (no await inside a for loop in runVerify)
-awk '/async function runVerify/,/^}/' scripts/push-to-confluence.mjs | grep -c "await httpsRequest"
-# Expected: 0
-```
-
-### Step 3 — Conditional: if spec 15 has already landed
-
-If `CliError` is already imported in the file (check with `grep "CliError" scripts/push-to-confluence.mjs`), replace the two `process.exit(1)` calls added in Step 1 with `throw new CliError(...)` as noted in the comments:
-
-```js
-// Phase 1 early exit
-if (hasErrors) {
-	throw new CliError("Invalid page IDs found — fix confluence-pages.json", 1);
-}
-
-// Phase 3 terminal exit
-if (hasErrors) {
-	throw new CliError("One or more pages failed verification — see errors above", 1);
-}
-// success: return normally; main()'s try/catch handles process.exit(0)
-```
-
-If spec 15 has not landed, leave `process.exit(1)` and `process.exit(hasErrors ? 1 : 0)` as-is.
-
-## Test cases
-
-| Scenario | Input | Expected output | Exit code |
-|---|---|---|---|
-| All pages valid | `confluence-pages.json` with 3 valid numeric IDs, all return HTTP 200 | Three `✅ key (id): Title` lines; printed in key order | 0 |
-| One invalid ID (string, not integer) | `{ "page": "not-a-number" }` in `confluence-pages.json` | `❌ page: Invalid page ID … must be a positive integer` on stderr; no GETs fired | 1 |
-| One invalid ID (zero) | `{ "page": 0 }` | `❌ page: Invalid page ID …` on stderr; no GETs fired | 1 |
-| Multiple invalid IDs | `{ "a": "x", "b": -1 }` | Two `❌` lines on stderr; no GETs fired | 1 |
-| One 404 response | Page ID in file, Confluence returns 404 | `❌ key (id): Page not found — 404` on stderr | 1 |
-| One network error | `httpsRequest` rejects for one entry | `⚠️  key (id): Network error` on stderr | 1 |
-| One unexpected HTTP status (e.g. 503) | Confluence returns 503 | `⚠️  key (id): Unexpected HTTP 503` on stderr | 1 |
-| Mix: one ✅, one ❌ 404 | Two entries, one passes, one 404 | One `✅` line, one `❌` line; both GETs fired in parallel | 1 |
-| Empty `confluence-pages.json` (only `_comment`) | `{ "_comment": "…" }` | `Verifying 0 page(s):` on stdout; exits immediately | 0 |
-| Performance: 5 entries at 200ms each (simulated) | — | Total time ≈ 200ms (one round trip), not ≈ 1000ms (five serial) | 0 |
-
-## Acceptance criteria
-
-- [ ] `runVerify()` contains no `await` inside a `for` loop — ID validation is sync, HTTP calls are concurrent.
-- [ ] `Promise.all` is used for the HTTP phase.
-- [ ] Results are printed after all promises settle (not interleaved with I/O operations).
-- [ ] Output lines are printed in `confluence-pages.json` key order.
-- [ ] Invalid page IDs are reported synchronously, before any HTTP requests are made.
-- [ ] If any IDs are invalid, the function exits 1 without firing any GETs.
-- [ ] The output format (`✅ / ❌ / ⚠️` prefix + `key (id): message`) is identical to the current format.
-- [ ] The `_comment` key is filtered out and never included in verification.
-- [ ] All other functions in the file are unchanged.
-- [ ] Lint passes.
-
-## Verification
-
-```sh
-# 1. Confirm no sequential await inside verify
-awk '/^async function runVerify/,/^\}/' scripts/push-to-confluence.mjs | grep "await httpsRequest"
-# Expected: no output (no direct await httpsRequest inside the function body outside Promise.all)
-
-# 2. Confirm Promise.all is used
-grep -n "Promise.all" scripts/push-to-confluence.mjs
-# Expected: at least one match inside runVerify
-
-# 3. Confirm ID validation phase is separate from HTTP phase
-grep -n "Phase 1\|Phase 2\|Phase 3" scripts/push-to-confluence.mjs
-# Expected: three comment lines (optional — they may be removed, but the structure should exist)
-
-# 4. Confirm output format is unchanged (key strings present)
-grep "✅\|❌\|⚠️" scripts/push-to-confluence.mjs | grep "runVerify\|key.*id\|r\.key"
-# Expected: matches in the results-printing loop
-
-# 5. Smoke test — no confluence-pages.json
-unset CONFLUENCE_URL CONFLUENCE_USER CONFLUENCE_TOKEN
-node scripts/push-to-confluence.mjs --verify 2>&1; echo "Exit: $?"
-# Expected: "confluence-pages.json not found" error, exit 1
-
-# 6. Timing test (optional, requires real credentials and 3+ entries in confluence-pages.json)
-# time node scripts/push-to-confluence.mjs --verify
-# Before this change: ~N × round-trip latency
-# After this change:  ~1 × round-trip latency
-```
-
-## Out of scope
-
-- Do not add rate limiting or a concurrency cap. Confluence's API handles the typical number of mapped pages (< 50) without throttling. If a project has hundreds of entries, a concurrency cap can be added as a follow-up.
-- Do not change the output format (`✅ / ❌ / ⚠️` lines).
-- Do not modify `httpsRequest`, `loadCredentials`, `makeBasicAuth`, or any other helper.
-- Do not add `--verbose` or `--quiet` flags.
-- Do not modify `runSetup()`, `main()`, or any other function.
-- No version bump required for this internal refactor.
-
-## Follow-ups
-
-- If a project accumulates > 50 pages in `confluence-pages.json`, consider adding a `p-limit` concurrency cap (e.g. 10 concurrent requests) to avoid saturating the VPN or hitting Confluence rate limits. This is not needed now.
-- After spec 15 (`CliError`) lands, update the two `process.exit` calls in this function to `throw new CliError(...)` as noted in the Step 3 conditional.
-- After spec 09 (unit tests) lands, add a test that mocks `httpsRequest` to return immediately and asserts that all calls were made before any results were printed.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/19-pnpm-bump-tooling.md b/apps/claude-code/unic-confluence/docs/plans/19-pnpm-bump-tooling.md
deleted file mode 100644
index a2a31d69..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/19-pnpm-bump-tooling.md
+++ /dev/null
@@ -1,373 +0,0 @@
-# 19. `pnpm bump` — atomic version bump tooling
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** M
-**Version impact:** minor
-**Depends on:** spec 15 (`CliError` class in `scripts/lib/errors.mjs`)
-**Touches:** `scripts/bump-version.mjs` (create), `scripts/tag.mjs` (create), `package.json`, `README.md`
-
-## Context
-
-All 18 shipped specs left `plugin.json` at `2.0.0` with `CHANGELOG.md`'s `[Unreleased]` section empty. The root cause: bumping requires three coordinated manual edits (`plugin.json`, `marketplace.json` via `pnpm sync-version`, `CHANGELOG.md`) with no script to orchestrate them. The existing `pnpm release` script stages and commits, but does not bump, does not promote `[Unreleased]`, and conflates "promote CHANGELOG" with "commit" in a way that no longer fits a per-spec cadence.
-
-This spec adds:
-
-- `pnpm bump <patch|minor|major>` — one command that: bumps `plugin.json`, mirrors into `marketplace.json` (via existing `sync-version.mjs`), and promotes `## [Unreleased]` → `## [X.Y.Z] — YYYY-MM-DD` with a fresh empty `[Unreleased]` above it. Refuses to promote an empty `[Unreleased]`.
-- `pnpm tag` — a lightweight replacement for `pnpm release`; syncs `marketplace.json` and creates a local `git tag vX.Y.Z`. Does not commit or push (those remain manual).
-
-After this spec, `pnpm release` is removed (its stage-and-commit behaviour is now the implementer's responsibility per the per-spec workflow, not a script's).
-
-## Current behaviour
-
-`package.json` scripts:
-
-```json
-"sync-version": "node scripts/sync-version.mjs",
-"release": "node scripts/sync-version.mjs && git add .claude-plugin/marketplace.json CHANGELOG.md && git commit -m \"chore: release v$(node -e 'const fs=require(\"fs\");const p=JSON.parse(fs.readFileSync(\".claude-plugin/plugin.json\",\"utf8\"));process.stdout.write(p.version)')\"",
-```
-
-No `bump` or `tag` script exists. Bumping version is fully manual:
-1. Hand-edit `.claude-plugin/plugin.json` `version` field.
-2. Hand-edit `CHANGELOG.md` to promote `[Unreleased]`.
-3. `pnpm sync-version` (mirrors to marketplace.json).
-4. `pnpm release` (stages + commits).
-5. `git tag vX.Y.Z && git push origin vX.Y.Z`.
-
-`scripts/bump-version.mjs` — does not exist.
-`scripts/tag.mjs` — does not exist.
-
-## Target behaviour
-
-### `pnpm bump <patch|minor|major>`
-
-1. Reads current version from `.claude-plugin/plugin.json`.
-2. Validates `## [Unreleased]` in `CHANGELOG.md` contains at least one non-`(none)` bullet. Exits 1 with a clear message if not.
-3. Computes next version per SemVer (minor resets patch to 0; major resets minor and patch).
-4. Writes updated `plugin.json` (same 2-space indent, trailing newline).
-5. Calls `node scripts/sync-version.mjs` via `spawnSync` to mirror into `marketplace.json`.
-6. Rewrites `CHANGELOG.md`: replaces `## [Unreleased]` and its body with a fresh empty `[Unreleased]` block followed by `## [X.Y.Z] — YYYY-MM-DD` (UTC date via `new Date().toISOString().slice(0, 10)`) and the original body.
-7. Logs `bump: 2.0.0 → 2.1.0`.
-
-### `pnpm tag`
-
-1. Reads version from `plugin.json`.
-2. Runs `sync-version.mjs` as a safety net (idempotent if already in sync).
-3. Creates a local `git tag vX.Y.Z`. Fails clearly if tag already exists.
-4. Logs `Tagged v2.1.0. Run: git push --follow-tags`.
-
-### `pnpm release` — removed
-
-The `release` script is deleted from `package.json`. Its auto-commit behaviour is replaced by the standard `git add -A && git commit -m "feat(spec-NN): ..."` workflow described in `PROMPT.md`.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `scripts/bump-version.mjs` | Create |
-| `scripts/tag.mjs` | Create |
-| `package.json` | Add `bump`, add `tag`, remove `release` |
-| `README.md` | Replace "Releasing" section |
-
-## Implementation steps
-
-### Step 1 — Create `scripts/bump-version.mjs`
-
-Create the file with exactly this content (tabs for indentation, `// @ts-check`):
-
-```js
-#!/usr/bin/env node
-// @ts-check
-// SPDX-License-Identifier: LGPL-3.0-or-later
-import { readFileSync, writeFileSync } from "node:fs";
-import { spawnSync } from "node:child_process";
-import path from "node:path";
-import { fileURLToPath } from "node:url";
-import { CliError } from "./lib/errors.mjs";
-
-const TYPES = /** @type {const} */ (["patch", "minor", "major"]);
-const type = process.argv[2];
-
-try {
-	if (!TYPES.includes(/** @type {any} */ (type))) {
-		throw new CliError(`Usage: pnpm bump <patch|minor|major>`);
-	}
-
-	const root = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..");
-	const pluginPath = path.join(root, ".claude-plugin/plugin.json");
-	const changelogPath = path.join(root, "CHANGELOG.md");
-
-	/** @type {{ version: string, [key: string]: unknown }} */
-	let pluginJson;
-	try {
-		pluginJson = /** @type {any} */ (JSON.parse(readFileSync(pluginPath, "utf8")));
-	} catch (err) {
-		throw new CliError(`bump: cannot read ${pluginPath}: ${/** @type {Error} */ (err).message}`);
-	}
-
-	const version = pluginJson.version;
-	if (!version || typeof version !== "string") {
-		throw new CliError(`bump: .version is missing or not a string in ${pluginPath}`);
-	}
-
-	const parts = version.split(".");
-	if (parts.length !== 3 || parts.some((p) => !/^\d+$/.test(p))) {
-		throw new CliError(`bump: cannot parse version "${version}" — expected X.Y.Z`);
-	}
-	let [major, minor, patch] = parts.map(Number);
-	if (type === "major") {
-		major++;
-		minor = 0;
-		patch = 0;
-	} else if (type === "minor") {
-		minor++;
-		patch = 0;
-	} else {
-		patch++;
-	}
-	const nextVersion = `${major}.${minor}.${patch}`;
-
-	let changelog;
-	try {
-		changelog = readFileSync(changelogPath, "utf8");
-	} catch (err) {
-		throw new CliError(`bump: cannot read ${changelogPath}: ${/** @type {Error} */ (err).message}`);
-	}
-
-	const unreleasedMatch = changelog.match(/## \[Unreleased\]([\s\S]*?)(?=\n## \[)/);
-	if (!unreleasedMatch) {
-		throw new CliError(`bump: no ## [Unreleased] section found in CHANGELOG.md`);
-	}
-
-	const unreleasedBody = unreleasedMatch[1];
-	const hasRealEntry = unreleasedBody
-		.split("\n")
-		.filter((l) => l.startsWith("- "))
-		.some((l) => l !== "- (none)");
-	if (!hasRealEntry) {
-		throw new CliError(
-			`bump: [Unreleased] has no entries — add a CHANGELOG entry before bumping`,
-		);
-	}
-
-	pluginJson.version = nextVersion;
-	writeFileSync(pluginPath, `${JSON.stringify(pluginJson, null, 2)}\n`, "utf8");
-
-	const today = new Date().toISOString().slice(0, 10);
-	const emptyUnreleased =
-		`## [Unreleased]\n\n### Breaking\n- (none)\n\n### Added\n- (none)\n\n### Fixed\n- (none)`;
-	const newChangelog = changelog.replace(
-		/## \[Unreleased\]([\s\S]*?)(?=\n## \[)/,
-		`${emptyUnreleased}\n\n## [${nextVersion}] — ${today}$1`,
-	);
-	writeFileSync(changelogPath, newChangelog, "utf8");
-
-	const syncResult = spawnSync("node", [path.join(root, "scripts/sync-version.mjs")], {
-		stdio: "inherit",
-	});
-	if (syncResult.status !== 0) {
-		throw new CliError(`bump: sync-version failed (exit ${syncResult.status ?? "unknown"})`);
-	}
-
-	console.log(`bump: ${version} → ${nextVersion}`);
-} catch (err) {
-	if (err instanceof CliError) {
-		console.error(err.message);
-		process.exit(err.exitCode);
-	}
-	throw err;
-}
-```
-
-### Step 2 — Create `scripts/tag.mjs`
-
-```js
-#!/usr/bin/env node
-// @ts-check
-// SPDX-License-Identifier: LGPL-3.0-or-later
-import { readFileSync } from "node:fs";
-import { spawnSync } from "node:child_process";
-import path from "node:path";
-import { fileURLToPath } from "node:url";
-import { CliError } from "./lib/errors.mjs";
-
-try {
-	const root = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..");
-	const pluginPath = path.join(root, ".claude-plugin/plugin.json");
-
-	/** @type {{ version: string }} */
-	const pluginJson = /** @type {any} */ (JSON.parse(readFileSync(pluginPath, "utf8")));
-	const version = pluginJson.version;
-	if (!version || typeof version !== "string") {
-		throw new CliError(`tag: .version missing in ${pluginPath}`);
-	}
-
-	// Safety sync before tagging (idempotent)
-	spawnSync("node", [path.join(root, "scripts/sync-version.mjs")], { stdio: "inherit" });
-
-	const tagResult = spawnSync("git", ["tag", `v${version}`], { stdio: "inherit" });
-	if (tagResult.status !== 0) {
-		throw new CliError(`tag: git tag failed — v${version} may already exist`);
-	}
-
-	console.log(`Tagged v${version}. Run: git push --follow-tags`);
-} catch (err) {
-	if (err instanceof CliError) {
-		console.error(err.message);
-		process.exit(/** @type {CliError} */ (err).exitCode);
-	}
-	throw err;
-}
-```
-
-### Step 3 — Update `package.json` scripts
-
-In `package.json`, locate the `"scripts"` block.
-
-**Remove** the `"release"` entry entirely.
-
-**Add** two new entries after `"sync-version"`:
-
-```json
-"bump": "node scripts/bump-version.mjs",
-"tag": "node scripts/tag.mjs",
-```
-
-The scripts block should look like:
-
-```json
-"scripts": {
-    "confluence": "node scripts/push-to-confluence.mjs",
-    "sync-version": "node scripts/sync-version.mjs",
-    "bump": "node scripts/bump-version.mjs",
-    "tag": "node scripts/tag.mjs",
-    "test": "node --test scripts/lib/frontmatter.test.mjs scripts/lib/inject.test.mjs scripts/lib/resolve.test.mjs",
-    "lint": "biome lint .",
-    "format": "biome format --write .",
-    "check": "biome check .",
-    "ci:check": "biome ci ."
-},
-```
-
-Note: `"typecheck"` is kept if it exists (it lives outside the block shown above — do not remove it).
-
-### Step 4 — Replace README "Releasing" section
-
-Locate the "Releasing" section in `README.md` (currently lines 223–228). Replace the entire section (from `## Releasing` through the final numbered step) with:
-
-```markdown
-## Releasing
-
-Every `feat(spec-NN)` / `fix(spec-NN)` commit includes its own version bump and CHANGELOG entry via `pnpm bump`. No separate "release commit" is needed.
-
-**Per change (every commit):**
-1. Add one bullet under the matching subsection of `## [Unreleased]` in `CHANGELOG.md` (`### Breaking`, `### Added`, or `### Fixed`).
-2. `pnpm bump <patch|minor|major>` — bumps `plugin.json`, mirrors into `marketplace.json`, promotes `[Unreleased]` → a dated version section.
-3. `git add -A && git commit -m "feat(spec-NN): <description> (vX.Y.Z)"`.
-
-**SemVer policy:**
-| Change type | Bump |
-|---|---|
-| Breaking change to CLI flags, exit codes, or on-disk contracts | `major` |
-| New flag, subcommand, or user-visible feature | `minor` |
-| Bug fix, refactor, docs, internal tooling | `patch` |
-
-**To tag and push a release boundary (optional, periodic):**
-```sh
-pnpm tag                  # creates local git tag vX.Y.Z
-git push --follow-tags    # pushes tag to GitHub
-```
-
-**Optional — enable the local pre-push changelog check:**
-```sh
-git config core.hooksPath .githooks
-```
-```
-
-## Test cases
-
-| Scenario | Action | Expected outcome |
-|---|---|---|
-| Normal minor bump | Add `- foo` under `### Added`, run `pnpm bump minor` | `plugin.json` version incremented (patch reset to 0), `marketplace.json` matches, `CHANGELOG.md` has new dated section |
-| Bump with empty `[Unreleased]` | Run `pnpm bump patch` without editing CHANGELOG | Exits 1: `bump: [Unreleased] has no entries` |
-| Invalid bump type | `pnpm bump hotfix` | Exits 1: `Usage: pnpm bump <patch|minor|major>` |
-| Major bump resets minor + patch | `plugin.json` at `2.3.1`, `pnpm bump major` | Next version is `3.0.0` |
-| Minor bump resets patch | `plugin.json` at `2.3.1`, `pnpm bump minor` | Next version is `2.4.0` |
-| `pnpm tag` creates tag | After a bump, run `pnpm tag` | `git tag -l` shows `v<version>`; log says "Tagged vX.Y.Z" |
-| `pnpm tag` on already-tagged | Run `pnpm tag` twice | Exits 1: "git tag failed — may already exist" |
-
-## Verification
-
-```sh
-# 1. Syntax check
-pnpm typecheck
-# Expected: no errors in new scripts
-
-# 2. Lint
-pnpm lint
-# Expected: passes
-
-# 3. Simulate a bump (undo after)
-# First add a dummy CHANGELOG entry
-ORIG_CHANGELOG=$(cat CHANGELOG.md)
-ORIG_PLUGIN=$(cat .claude-plugin/plugin.json)
-ORIG_MARKET=$(cat .claude-plugin/marketplace.json)
-
-# Add an entry to [Unreleased]
-node -e "
-const fs = require('fs');
-let c = fs.readFileSync('CHANGELOG.md', 'utf8');
-c = c.replace('- (none)\n\n## [2', '- Test entry for spec-19\n\n## [2');
-fs.writeFileSync('CHANGELOG.md', c);
-"
-
-# Run bump
-node scripts/bump-version.mjs minor
-
-# Check plugin.json bumped
-node -e "const p=JSON.parse(require('fs').readFileSync('.claude-plugin/plugin.json','utf8')); console.log('plugin.json version:', p.version)"
-# Expected: 2.1.0 (or whatever is next)
-
-# Check marketplace.json matches
-node -e "const m=JSON.parse(require('fs').readFileSync('.claude-plugin/marketplace.json','utf8')); console.log('marketplace.json version:', m.plugins[0].version)"
-# Expected: same as plugin.json
-
-# Check CHANGELOG has dated section
-grep "## \[2\." CHANGELOG.md
-# Expected: new dated line e.g. "## [2.1.0] — 2026-04-24"
-
-# Check [Unreleased] is fresh and empty
-grep -A6 "## \[Unreleased\]" CHANGELOG.md
-# Expected: empty Breaking/Added/Fixed all "(none)"
-
-# 4. Verify bump rejects empty [Unreleased]
-node scripts/bump-version.mjs patch; echo "Exit: $?"
-# Expected: Exit: 1 with "no entries" message
-
-# 5. Restore original files (UNDO)
-echo "$ORIG_CHANGELOG" > CHANGELOG.md
-echo "$ORIG_PLUGIN" > .claude-plugin/plugin.json
-echo "$ORIG_MARKET" > .claude-plugin/marketplace.json
-```
-
-## Acceptance criteria
-
-- [ ] `pnpm bump minor` on a CHANGELOG with at least one non-`(none)` bullet: bumps `plugin.json`, syncs `marketplace.json`, promotes `[Unreleased]` → `[X.Y.Z] — today`, inserts fresh empty `[Unreleased]`.
-- [ ] `pnpm bump patch` with empty `[Unreleased]` (all `- (none)`): exits 1 with a message pointing at adding a CHANGELOG entry.
-- [ ] `pnpm bump hotfix`: exits 1 with usage message.
-- [ ] Major bump resets minor and patch to 0; minor bump resets patch to 0.
-- [ ] `pnpm tag`: creates local `git tag vX.Y.Z`; exits 1 if tag already exists.
-- [ ] `pnpm release` is removed from `package.json`.
-- [ ] `pnpm typecheck` passes on new scripts.
-- [ ] `pnpm lint` passes.
-- [ ] README "Releasing" section reflects the new per-spec flow.
-
-## Out of scope
-
-- Pushing the tag — remains a manual `git push --follow-tags`.
-- Auto-creating a GitHub Release.
-- `pnpm verify:changelog` (enforcement CI check) — that is spec 20.
-- Updating `PROMPT.md` or `CLAUDE.md` — that is spec 21.
-- Backfilling version bumps for specs 00–18 — history is shipped.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/20-pnpm-verify-changelog.md b/apps/claude-code/unic-confluence/docs/plans/20-pnpm-verify-changelog.md
deleted file mode 100644
index 38e57990..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/20-pnpm-verify-changelog.md
+++ /dev/null
@@ -1,322 +0,0 @@
-# 20. `pnpm verify:changelog` — CI enforcement and pre-push hook
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** M
-**Version impact:** minor
-**Depends on:** spec 19 (`pnpm bump` tooling)
-**Touches:** `scripts/verify-changelog.mjs` (create), `.githooks/pre-push` (create), `package.json`, `.github/workflows/ci.yml`, `README.md`
-
-## Context
-
-Spec 19 adds the `pnpm bump` command that atomically bumps the version and promotes `CHANGELOG.md`. But `pnpm bump` is only useful if developers (and Ralph) actually run it. Without enforcement, the pattern repeats: specs ship without bumps.
-
-This spec adds two enforcement layers:
-
-1. **CI check** (`pnpm verify:changelog`): a GitHub Actions step that fails PRs where source files changed but `plugin.json` version did not bump, or where the new version section in `CHANGELOG.md` is missing or empty. Gated to pull-request events so direct pushes to `main` (post-merge) don't re-fail.
-2. **Local pre-push hook** (optional, opt-in): a committed `.githooks/pre-push` file that runs the same check. Users enable it with `git config core.hooksPath .githooks`.
-
-Both layers run the same `scripts/verify-changelog.mjs` script so there is a single source of logic.
-
-## Current behaviour
-
-`package.json`: no `verify:changelog` script.
-`.github/workflows/ci.yml`: runs `pnpm ci:check` and `pnpm test` only.
-`.githooks/` directory: does not exist.
-`scripts/verify-changelog.mjs`: does not exist.
-
-## Target behaviour
-
-### `scripts/verify-changelog.mjs`
-
-Logic:
-
-1. Determine the diff base:
-   - In CI (env `CI=true`): `origin/main` (set via `git fetch --unshallow` already done by `actions/checkout@v4` with `fetch-depth: 0`).
-   - Locally: `@{upstream}` if a tracking branch exists; fallback to `HEAD~1`.
-2. Run `git diff --name-only <base>...HEAD` to get the list of changed files.
-3. If **no changed files match** the guarded paths (see below), exit 0 silently (nothing to enforce).
-4. **Guarded paths** (any match triggers enforcement):
-   - `scripts/**/*.mjs`
-   - `commands/**/*.md`
-   - `.claude-plugin/plugin.json`
-   - `.claude-plugin/marketplace.json`
-   - `CLAUDE.md`
-   - `README.md`
-5. Read `plugin.json` version on HEAD and on base (via `git show <base>:.claude-plugin/plugin.json`).
-6. Versions must differ. If they are the same, exit 1:
-   ```
-   verify:changelog: version in plugin.json was not bumped
-     current: 2.0.0 (same as base)
-     Run: pnpm bump <patch|minor|major>
-   ```
-7. Check `CHANGELOG.md` HEAD contains a section `## [<HEAD version>] — YYYY-MM-DD` with at least one non-`(none)` bullet. If not, exit 1:
-   ```
-   verify:changelog: CHANGELOG.md has no entry for version 2.1.0
-     Add bullets under [Unreleased] then run: pnpm bump minor
-   ```
-8. Exit 0 with a brief summary: `verify:changelog: ok — version 2.0.0 → 2.1.0`.
-
-No runtime deps: `node:child_process` (`spawnSync`), `node:fs` (`readFileSync`), `node:path`.
-
-### `.githooks/pre-push`
-
-A committed, executable shell script that invokes `pnpm verify:changelog`. Users opt in by running:
-```sh
-git config core.hooksPath .githooks
-```
-
-### `.github/workflows/ci.yml`
-
-Append one new step after the existing `pnpm test` step, gated to PRs:
-
-```yaml
-      - run: pnpm verify:changelog
-        if: github.event_name == 'pull_request'
-```
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `scripts/verify-changelog.mjs` | Create |
-| `.githooks/pre-push` | Create (executable) |
-| `package.json` | Add `verify:changelog` script |
-| `.github/workflows/ci.yml` | Append one step |
-| `README.md` | The hook opt-in note is already added by spec 19; no additional README changes needed here |
-
-## Implementation steps
-
-### Step 1 — Create `scripts/verify-changelog.mjs`
-
-```js
-#!/usr/bin/env node
-// @ts-check
-// SPDX-License-Identifier: LGPL-3.0-or-later
-import { readFileSync } from "node:fs";
-import { spawnSync } from "node:child_process";
-import path from "node:path";
-import { fileURLToPath } from "node:url";
-
-const root = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..");
-
-/** Paths that trigger enforcement when changed */
-const GUARDED = [
-	/^scripts\/.+\.mjs$/,
-	/^commands\/.+\.md$/,
-	/^\.claude-plugin\/plugin\.json$/,
-	/^\.claude-plugin\/marketplace\.json$/,
-	/^CLAUDE\.md$/,
-	/^README\.md$/,
-];
-
-/**
- * @param {string[]} args
- * @returns {{ stdout: string, status: number }}
- */
-function git(...args) {
-	const result = spawnSync("git", args, { encoding: "utf8", cwd: root });
-	return { stdout: result.stdout ?? "", status: result.status ?? 1 };
-}
-
-function fail(/** @type {string} */ msg) {
-	console.error(`verify:changelog: ${msg}`);
-	process.exit(1);
-}
-
-// Determine diff base
-const isCI = process.env.CI === "true";
-let base = "origin/main";
-if (!isCI) {
-	const upstream = git("rev-parse", "--abbrev-ref", "--symbolic-full-name", "@{upstream}");
-	base = upstream.status === 0 ? upstream.stdout.trim() : "HEAD~1";
-}
-
-// List changed files
-const diff = git("diff", "--name-only", `${base}...HEAD`);
-if (diff.status !== 0) {
-	// If diff fails (e.g. on a shallow clone), skip silently
-	console.log("verify:changelog: skipped (git diff unavailable)");
-	process.exit(0);
-}
-
-const changedFiles = diff.stdout.trim().split("\n").filter(Boolean);
-const triggered = changedFiles.some((f) => GUARDED.some((re) => re.test(f)));
-if (!triggered) {
-	console.log("verify:changelog: ok (no guarded paths changed)");
-	process.exit(0);
-}
-
-// Read HEAD version
-const pluginPath = path.join(root, ".claude-plugin/plugin.json");
-/** @type {{ version: string }} */
-let headPlugin;
-try {
-	headPlugin = /** @type {any} */ (JSON.parse(readFileSync(pluginPath, "utf8")));
-} catch {
-	fail(`cannot read ${pluginPath}`);
-	process.exit(1); // unreachable — satisfies TS
-}
-const headVersion = headPlugin.version;
-
-// Read base version
-const basePluginRaw = git("show", `${base}:.claude-plugin/plugin.json`);
-let baseVersion = "";
-if (basePluginRaw.status === 0) {
-	try {
-		baseVersion = /** @type {{ version: string }} */ (JSON.parse(basePluginRaw.stdout)).version;
-	} catch {
-		// base doesn't have plugin.json yet — treat as empty
-	}
-}
-
-if (headVersion === baseVersion) {
-	fail(
-		`version in plugin.json was not bumped\n  current: ${headVersion} (same as base)\n  Run: pnpm bump <patch|minor|major>`,
-	);
-}
-
-// Check CHANGELOG has a section for headVersion with at least one real bullet
-let changelog;
-try {
-	changelog = readFileSync(path.join(root, "CHANGELOG.md"), "utf8");
-} catch {
-	fail("cannot read CHANGELOG.md");
-	process.exit(1);
-}
-
-const sectionMatch = changelog.match(
-	new RegExp(`## \\[${headVersion.replace(/\./g, "\\.")}\\] — \\d{4}-\\d{2}-\\d{2}([\\s\\S]*?)(?=\\n## \\[|$)`),
-);
-if (!sectionMatch) {
-	fail(
-		`CHANGELOG.md has no entry for version ${headVersion}\n  Add bullets under [Unreleased] then run: pnpm bump`,
-	);
-}
-
-const sectionBody = sectionMatch[1];
-const hasRealEntry = sectionBody
-	.split("\n")
-	.filter((l) => l.startsWith("- "))
-	.some((l) => l !== "- (none)");
-if (!hasRealEntry) {
-	fail(
-		`CHANGELOG.md section [${headVersion}] has no entries — only "(none)" placeholders found\n  Add bullets under [Unreleased] then re-run: pnpm bump`,
-	);
-}
-
-console.log(`verify:changelog: ok — version ${baseVersion} → ${headVersion}`);
-```
-
-### Step 2 — Create `.githooks/pre-push` (executable)
-
-Create the directory `.githooks/` in the repo root and create the file `.githooks/pre-push`:
-
-```sh
-#!/bin/sh
-pnpm verify:changelog
-```
-
-Make it executable:
-```sh
-chmod +x .githooks/pre-push
-```
-
-The file must be executable in git's index too:
-```sh
-git update-index --chmod=+x .githooks/pre-push
-```
-
-### Step 3 — Add `verify:changelog` to `package.json`
-
-In the `"scripts"` block, add after `"tag"`:
-
-```json
-"verify:changelog": "node scripts/verify-changelog.mjs",
-```
-
-### Step 4 — Append step to `.github/workflows/ci.yml`
-
-In `.github/workflows/ci.yml`, find the final `- run: pnpm test` step and append immediately after:
-
-```yaml
-      - run: pnpm verify:changelog
-        if: github.event_name == 'pull_request'
-```
-
-**Before** (end of file):
-```yaml
-      - run: pnpm ci:check
-      - run: pnpm test
-```
-
-**After**:
-```yaml
-      - run: pnpm ci:check
-      - run: pnpm test
-      - run: pnpm verify:changelog
-        if: github.event_name == 'pull_request'
-```
-
-## Test cases
-
-| Scenario | Setup | Expected |
-|---|---|---|
-| Source changed + version bumped + CHANGELOG entry | `scripts/*.mjs` in diff, version different, CHANGELOG section present with real bullet | Exit 0: `verify:changelog: ok — version X.Y.Z → X.Y+1.Z` |
-| Source changed, version NOT bumped | `scripts/*.mjs` in diff, same version on HEAD and base | Exit 1: "version in plugin.json was not bumped" |
-| Source changed, version bumped, CHANGELOG missing section | version changed but no `## [X.Y.Z]` in CHANGELOG | Exit 1: "CHANGELOG.md has no entry for version X.Y.Z" |
-| Source changed, version bumped, CHANGELOG section all `(none)` | section exists but all bullets are `- (none)` | Exit 1: "section … has no entries" |
-| Only docs-only changes outside guarded paths | e.g. only `docs/plans/*.md` changed | Exit 0: "no guarded paths changed" |
-| README.md changed | `README.md` in diff, but version not bumped | Exit 1: "version … was not bumped" |
-| No changes at all | empty diff | Exit 0: "no guarded paths changed" |
-
-## Acceptance criteria
-
-- [ ] `pnpm verify:changelog` exits 0 when guarded files changed AND version bumped AND CHANGELOG section with real entry exists.
-- [ ] `pnpm verify:changelog` exits 1 with a clear message when version not bumped.
-- [ ] `pnpm verify:changelog` exits 1 with a clear message when CHANGELOG section missing or all `(none)`.
-- [ ] `pnpm verify:changelog` exits 0 silently when no guarded paths changed.
-- [ ] `.github/workflows/ci.yml` has a new `pnpm verify:changelog` step gated to `pull_request` events.
-- [ ] `.githooks/pre-push` exists, is executable, and runs `pnpm verify:changelog`.
-- [ ] `git ls-files --stage .githooks/pre-push` shows mode `100755`.
-- [ ] `pnpm typecheck` passes on the new script.
-- [ ] `pnpm lint` passes.
-
-## Verification
-
-```sh
-# 1. Typecheck + lint
-pnpm typecheck
-pnpm lint
-
-# 2. Test: no guarded files changed (should pass)
-# Reset to a clean state and run the check
-git stash  # if any uncommitted changes
-pnpm verify:changelog
-# Expected: exit 0, "ok (no guarded paths changed)" or "ok — version ..."
-
-# 3. Test: check that pre-push hook file is executable
-ls -la .githooks/pre-push
-# Expected: -rwxr-xr-x
-
-# 4. Test: hook executable bit in git index
-git ls-files --stage .githooks/pre-push
-# Expected: 100755 ...
-
-# 5. Confirm CI yaml has the new step
-grep "verify:changelog" .github/workflows/ci.yml
-# Expected: one match
-
-# 6. Confirm verify:changelog is in package.json
-grep "verify:changelog" package.json
-# Expected: one match
-```
-
-## Out of scope
-
-- Updating `PROMPT.md`, `CLAUDE.md`, or `docs/plans/README.md` to document this tool for Ralph — that is spec 21.
-- Enforcing commit message format (conventional commits are a convention, not enforced by this script).
-- Detecting force-pushes or rebased commits (the simple three-dot diff is sufficient).
-- Adding `fetch-depth: 0` to the CI workflow — `actions/checkout@v4` with default settings does a shallow clone; the three-dot diff `origin/main...HEAD` works because CI fetches the PR ref with enough history. If it fails on a shallow clone, the script exits 0 silently (see Step 1 logic).
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/21-versioning-docs-and-prompt.md b/apps/claude-code/unic-confluence/docs/plans/21-versioning-docs-and-prompt.md
deleted file mode 100644
index c8081ad3..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/21-versioning-docs-and-prompt.md
+++ /dev/null
@@ -1,271 +0,0 @@
-# 21. Versioning docs and Ralph integration
-**Status: done — 2026-04-24**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec 19 (`pnpm bump`), spec 20 (`pnpm verify:changelog`)
-**Touches:** `PROMPT.md`, `docs/plans/README.md`, `CLAUDE.md`
-
-## Context
-
-Specs 19 and 20 add the tooling (`pnpm bump`, `pnpm verify:changelog`, `.githooks/pre-push`). This spec wires them into the human- and Ralph-facing documentation so the workflow is actually followed going forward.
-
-The three gaps to close:
-1. **`PROMPT.md`** — Ralph's brief says nothing about CHANGELOG or version bumps. Ralph completed specs 14–18 without touching either. A new Step 4.5 tells Ralph exactly what to do between "Verify" and "Commit".
-2. **`docs/plans/README.md`** — The "Versioning policy" section has a stale "Bump plan" table (it was never followed). The "Release flow" code block describes a now-deleted `pnpm release`. It needs to reflect the per-spec cadence and new scripts. Each future spec should start with a `**Version impact:**` line.
-3. **`CLAUDE.md`** — The "Plugin versioning" section says "Keep them in sync manually when releasing" — the opposite of what specs 19–20 enable. It needs to describe `pnpm bump` and CI enforcement.
-
-No code changes in this spec — docs only.
-
-## Current behaviour
-
-### `PROMPT.md` — Step 3 ground rules (relevant excerpt)
-
-```
-- **Never hand-edit** `.claude-plugin/marketplace.json` version — use `pnpm release` (available after spec 06)
-```
-
-There is no step about CHANGELOG or version bumps anywhere in `PROMPT.md`.
-
-### `docs/plans/README.md` — Versioning policy section (relevant excerpt)
-
-```markdown
-**Bump plan across this roadmap:**
-
-| Spec(s) | Target version | Reason |
-|---|---|---|
-| `00`–`02` | `1.0.3` (optional patch) | Tooling + docs only; can skip and batch |
-| `03` | **`2.0.0`** | Breaking: "no markers" now errors instead of silently appending |
-| `04`, `05`, `07` | `2.1.0`, `2.2.0`, `2.3.0` (or batch into `2.1.0`) | New features |
-| `06`, `08`–`18` | PATCH (or batch) | Bug fixes, refactors, tooling |
-
-**Release flow** (post-spec `06`):
-
-```sh
-# 1. Edit version field in .claude-plugin/plugin.json
-# 2. Update CHANGELOG.md: move items from [Unreleased] to new version heading with today's date
-# 3. Run the sync + commit + tag script:
-pnpm release
-```
-```
-
-### `CLAUDE.md` — Plugin versioning section
-
-```markdown
-## Plugin versioning
-
-Both `plugin.json` and `marketplace.json` carry a `version` field. Keep them in sync manually when releasing. There is no automated publish step.
-```
-
-## Target behaviour
-
-### `PROMPT.md`
-
-**Ground rule update:** Replace the line:
-```
-- **Never hand-edit** `.claude-plugin/marketplace.json` version — use `pnpm release` (available after spec 06)
-```
-with:
-```
-- **Never hand-edit** `.claude-plugin/marketplace.json` version — use `pnpm bump` (available after spec 19)
-```
-
-**New step inserted between Step 4 (Verify) and Step 5 (Mark done and commit):**
-
-```markdown
-## Step 4.5 — Bump version + CHANGELOG
-
-1. Read the spec's `**Version impact:** patch|minor|major` line at the top of the spec file.
-   If the line is absent, infer: breaking CLI/contract change → `major`; new flag/feature → `minor`; bug fix, refactor, docs → `patch`.
-
-2. Append **one bullet** to the matching subsection under `## [Unreleased]` in `CHANGELOG.md`, replacing the `- (none)` placeholder on first use:
-   - `### Breaking` — CLI flag renamed/removed, exit-code change, on-disk file schema change
-   - `### Added` — new flag, subcommand, or user-visible feature
-   - `### Fixed` — bug fix, refactor, docs, internal tooling
-   Wording: one line, user-facing, present-tense description (e.g. `- \`pnpm bump\` command for atomic version bumping`).
-
-3. Run:
-   ```sh
-   pnpm bump <patch|minor|major>
-   ```
-   This atomically: increments `plugin.json` version, mirrors into `marketplace.json`, and promotes `[Unreleased]` → a new dated section.
-
-4. Run `pnpm verify:changelog` to confirm the check passes.
-```
-
-**Step 5 commit message format update:** In Step 5, change the commit message example to include the version:
-
-Before:
-```sh
-git commit -m "feat(spec-NN): <short description of what was implemented>"
-```
-
-After:
-```sh
-git commit -m "feat(spec-NN): <short description of what was implemented> (vX.Y.Z)"
-```
-
-Replace `X.Y.Z` with the actual new version that `pnpm bump` produced.
-
-### `docs/plans/README.md`
-
-**Intro paragraph** — Append to the first paragraph in "How to use (for Ralph)":
-```
-Each spec starts with a `**Version impact:** patch|minor|major` line just under the title. Use it in Step 4.5.
-```
-
-**"Bump plan" table** — Remove the entire table (it was never followed and is now stale). Replace with a single sentence:
-```
-Each spec bumps per its `**Version impact:**` declaration. See `PROMPT.md` Step 4.5.
-```
-
-**"Release flow" code block** — Replace with:
-
-```markdown
-**Release flow:**
-
-```sh
-# Per change (every spec commit):
-pnpm bump <patch|minor|major>   # bumps version, promotes CHANGELOG
-pnpm verify:changelog            # confirm CI check passes
-
-# Periodic: tag and push a release boundary
-pnpm tag                         # creates local git tag vX.Y.Z
-git push --follow-tags           # publishes tag to GitHub
-```
-```
-
-**Ground rules** — Update the "Plugin versioning" bullet:
-
-Before:
-```
-- **Plugin versioning**: `.claude-plugin/plugin.json` is the single source of truth for the version number. **Never hand-edit `.claude-plugin/marketplace.json`** — the `pnpm release` script (introduced in spec `06`) syncs it.
-```
-
-After:
-```
-- **Plugin versioning**: `.claude-plugin/plugin.json` is the single source of truth for the version number. **Never hand-edit `.claude-plugin/marketplace.json`** — use `pnpm bump` (spec `19`) which bumps, syncs, and promotes the CHANGELOG in one step.
-```
-
-### `CLAUDE.md`
-
-Replace the entire "Plugin versioning" section. Current text:
-
-```markdown
-## Plugin versioning
-
-Both `plugin.json` and `marketplace.json` carry a `version` field. Keep them in sync manually when releasing. There is no automated publish step.
-```
-
-Replace with:
-
-```markdown
-## Plugin versioning
-
-`.claude-plugin/plugin.json` is the single source of truth. Every `feat(spec-NN)` / `fix(spec-NN)` commit includes a version bump and a dated `CHANGELOG.md` entry, via `pnpm bump <patch|minor|major>`. Never hand-edit `.claude-plugin/marketplace.json` — `pnpm bump` mirrors into it automatically.
-
-**SemVer policy:**
-- **major**: breaking change to CLI flags, exit codes, or on-disk contracts (`confluence-pages.json` schema, `~/.unic-confluence.json` format).
-- **minor**: new flag, subcommand, or user-visible feature.
-- **patch**: bug fix, refactor, docs, internal tooling.
-
-`pnpm verify:changelog` (and CI on PRs) rejects changes that modify source or user-facing docs without a version bump and CHANGELOG entry.
-```
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `PROMPT.md` | Update ground rule; add Step 4.5; update Step 5 commit example |
-| `docs/plans/README.md` | Intro note; remove stale bump table; update release-flow block; update ground rule bullet |
-| `CLAUDE.md` | Replace "Plugin versioning" section |
-
-## Implementation steps
-
-### Step 1 — Update `PROMPT.md`
-
-1a. Find the line:
-```
-- **Never hand-edit** `.claude-plugin/marketplace.json` version — use `pnpm release` (available after spec 06)
-```
-Replace `pnpm release` with `pnpm bump` and `spec 06` with `spec 19`.
-
-1b. After the `## Step 4 — Verify` section (ends with "Fix any failures before proceeding.") and before `## Step 5 — Mark done and commit`, insert the full Step 4.5 block described in "Target behaviour" above.
-
-1c. In `## Step 5 — Mark done and commit`, locate the `git commit` command and update it:
-- Before: `git commit -m "feat(spec-NN): <short description of what was implemented>"`
-- After: `git commit -m "feat(spec-NN): <short description of what was implemented> (vX.Y.Z)"`
-
-Add a note on the line below:
-```
-Replace `X.Y.Z` with the version output by `pnpm bump`.
-```
-
-### Step 2 — Update `docs/plans/README.md`
-
-2a. In the intro list under "How to use (for Ralph)", append as item 5:
-```
-5. Read the spec's `**Version impact:** patch|minor|major` line and execute `PROMPT.md` Step 4.5 before committing.
-```
-
-2b. In the "Versioning policy" section, find the "Bump plan" table (starts with `| Spec(s) | Target version | Reason |`). Remove the entire table including its header and all rows. Replace with the sentence: `Each spec bumps per its \`**Version impact:**\` declaration. See \`PROMPT.md\` Step 4.5.`
-
-2c. In the "Versioning policy" section, find the `**Release flow** (post-spec \`06\`):` block and replace it with the new "Release flow" block described in "Target behaviour" above.
-
-2d. In the "Ground rules" list, update the "Plugin versioning" bullet as described above.
-
-### Step 3 — Update `CLAUDE.md`
-
-Find the `## Plugin versioning` section. Replace its body (everything after the `## Plugin versioning` heading down to the next `##` heading, exclusive) with the new text described in "Target behaviour" above.
-
-## Acceptance criteria
-
-- [ ] `PROMPT.md` has a Step 4.5 section between Step 4 and Step 5 with instructions to add a CHANGELOG bullet, run `pnpm bump`, and run `pnpm verify:changelog`.
-- [ ] `PROMPT.md` ground rule references `pnpm bump` not `pnpm release`.
-- [ ] `PROMPT.md` Step 5 commit message example includes `(vX.Y.Z)`.
-- [ ] `docs/plans/README.md` has no "Bump plan" table.
-- [ ] `docs/plans/README.md` "Release flow" block references `pnpm bump` and `pnpm tag` (not `pnpm release`).
-- [ ] `docs/plans/README.md` ground rule for plugin versioning references `pnpm bump`.
-- [ ] `CLAUDE.md` "Plugin versioning" section describes the per-commit flow, SemVer policy, and CI enforcement.
-- [ ] `CLAUDE.md` does not reference `pnpm release` anywhere in "Plugin versioning".
-- [ ] No code files are modified — docs only.
-
-## Verification
-
-```sh
-# 1. Confirm Step 4.5 exists in PROMPT.md
-grep "Step 4.5" PROMPT.md
-# Expected: at least one match
-
-# 2. Confirm pnpm bump referenced in PROMPT.md
-grep "pnpm bump" PROMPT.md
-# Expected: matches in Step 3 and Step 4.5
-
-# 3. Confirm pnpm release not referenced in PROMPT.md
-grep "pnpm release" PROMPT.md
-# Expected: no matches
-
-# 4. Confirm stale bump plan table removed from docs/plans/README.md
-grep "Target version" docs/plans/README.md
-# Expected: no matches
-
-# 5. Confirm pnpm bump in README.md
-grep "pnpm bump" docs/plans/README.md
-# Expected: at least one match
-
-# 6. Confirm CLAUDE.md plugin versioning section updated
-grep "pnpm bump" CLAUDE.md
-# Expected: at least one match
-
-grep "pnpm release" CLAUDE.md
-# Expected: no matches in "Plugin versioning" section
-```
-
-## Out of scope
-
-- Backfilling `**Version impact:**` lines into specs 00–18 (completed, history).
-- Modifying any script files (`scripts/push-to-confluence.mjs`, `scripts/sync-version.mjs`, etc.).
-- Adding `**Version impact:**` lines to this spec file (spec 21) from within itself — Ralph should add the `**Version impact:**` line to future specs as part of the normal spec-writing process, not retroactively.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/22-auto-populate-aliases.md b/apps/claude-code/unic-confluence/docs/plans/22-auto-populate-aliases.md
deleted file mode 100644
index 7533a8dc..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/22-auto-populate-aliases.md
+++ /dev/null
@@ -1,585 +0,0 @@
-# 22. Auto-populate `confluence-pages.json` aliases
-**Status: done — 2026-04-27**
-
-**Priority:** P1
-**Effort:** M
-**Version impact:** minor
-**Depends on:** spec 09 (Vitest harness), spec 15 (`CliError` class), spec 19 (`pnpm bump`), spec 20 (`pnpm verify:changelog`)
-**Touches:** `scripts/push-to-confluence.mjs`, `scripts/lib/slug.mjs` (create), `scripts/lib/pages-file.mjs` (create), `scripts/lib/slug.test.mjs` (create), `scripts/lib/pages-file.test.mjs` (create), `commands/unic-confluence.md`, `README.md`, `CHANGELOG.md`, `docs/plans/README.md`, `package.json` (extend `test` script)
-
-## Context
-
-A Confluence page is addressable two ways: by a key looked up in `confluence-pages.json` (e.g. `spec-42410`) or by a raw numeric page ID (e.g. `804848595`). The lookup file lives at `process.cwd()/confluence-pages.json` and is read by `resolvePageId()` in `scripts/lib/resolve.mjs:24-59`. Today nothing writes to it — users who publish by raw ID either re-type the ID forever or hand-edit the file. There is also no built-in way to list configured aliases short of `cat confluence-pages.json`.
-
-This spec closes both gaps:
-
-1. After a successful publish-by-numeric-ID, the CLI slugifies the Confluence page's `title` and writes it as a new alias into `confluence-pages.json` at `cwd`, telling the user `✓ Saved alias "<slug>" → <id>`.
-2. A new `--list` flag prints the configured aliases as a sorted two-column table.
-
-Both entry points (the `node scripts/push-to-confluence.mjs` npm path and the `/unic-confluence` slash command) benefit because `commands/unic-confluence.md` is a thin pass-through that shells out to the same script. The on-disk schema stays `key → integer`, keeping this a **minor** bump and avoiding the "richer record" path explicitly flagged as out-of-scope in `CLAUDE.md` "Do not add" (lines 74-76).
-
-## Current behaviour
-
-- `scripts/push-to-confluence.mjs:386-388`: `pageArg = positionalArgs[0]; … const pageId = resolvePageId(pageArg);` — once `resolvePageId` returns, the script no longer knows whether the original arg was numeric.
-- `scripts/lib/resolve.mjs:12-14`: `isNumericId(arg)` exists and is reusable.
-- `confluence-pages.json` is purely read-only: only read sites are `resolve.mjs:41` and `runVerify` at `push-to-confluence.mjs:188`. Zero write sites.
-- `runVerify` already filters the `_comment` key (`push-to-confluence.mjs:193,199`) — same convention applies here.
-- `commands/unic-confluence.md:49-52` shells out to `node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" <page-key-or-id> <markdown-file>`. CLI changes are inherited.
-- No slug helper anywhere (`grep -r slug scripts/ commands/` → 0 hits).
-- The script's only `writeFileSync` is the credentials write at `push-to-confluence.mjs:322` (chmod 600). `confluence-pages.json` must NOT be chmod'd — it is checked into consumer repos.
-- Test suite: `scripts/lib/resolve.test.mjs` backs up `confluence-pages.json` around runs (lines 48-65) and spawns subprocesses with `cwd: "/tmp"` (lines 88-92, 107-111). The new auto-save logic must only fire on the publish code path so these tests remain green untouched.
-- `package.json` `test` script lists test files explicitly:
-  ```json
-  "test": "node --test scripts/lib/frontmatter.test.mjs scripts/lib/inject.test.mjs scripts/lib/resolve.test.mjs"
-  ```
-
-## Target behaviour
-
-### A. Auto-save alias after publish-by-numeric-ID
-
-Triggers iff **all** are true:
-- `isNumericId(pageArg)` was true (capture this before `resolvePageId` normalizes it).
-- `--dry-run` was not passed.
-- `--no-save` (new flag) was not passed.
-- The PUT succeeded (i.e. control reached the `✓ Published` log at line 476).
-
-Then, immediately before the `✓ Published` log:
-
-1. Compute `pagesPath = path.join(process.cwd(), "confluence-pages.json")`.
-2. Load the file via `readPagesFile(cwd)` — returns `{ pages: object | null, path }`. If absent, treat `pages` as `{}`.
-3. Reverse-lookup via `findAliasForId(pages, pageId)`. If non-null → log `ℹ Page ${pageId} already aliased as "${existing}"` and return without writing.
-4. Otherwise compute `baseSlug = slugify(title)`. Compute `alias = pickAvailableAlias(pages, baseSlug, pageId)` (resolves collisions and empty-slug fallback).
-5. `writePagesFile(pagesPath, appendAlias(pages, alias, pageId))`.
-6. Log `✓ Saved alias "${alias}" → ${pageId} in confluence-pages.json`.
-
-Any thrown error from steps 1-5 is non-fatal: catch and log `⚠ Could not save alias: ${err.message}`. The publish itself already succeeded — exit code stays 0.
-
-### B. New flag: `--list`
-
-Dispatched at the top of `main()` alongside `--setup` / `--check-auth` / `--verify`.
-
-- Reads `confluence-pages.json` from `process.cwd()`.
-- Missing file → `console.log("No confluence-pages.json in " + process.cwd())`; exit 0.
-- Empty or only `_comment` → `console.log("No aliases configured yet.")`; exit 0.
-- Otherwise prints a two-column table sorted alphabetically by alias, `_comment` filtered:
-  ```
-  alias                                     page id
-  ────────────────────────────────────────  ──────────────
-  profile-features                          804848595
-  spec-42410                                912345678
-  ```
-- No auth needed; no network call.
-
-### C. New flag: `--no-save`
-
-Opt-out for §A. Detected via `args.includes("--no-save")` next to `--dry-run` (`push-to-confluence.mjs:378-379`).
-
-### D. Slugification rules (`slugify(title)`)
-
-1. `String(title ?? "").normalize("NFD").replace(/[̀-ͯ]/g, "")` — strip combining diacritics.
-2. `.toLowerCase()`.
-3. Replace any `[^a-z0-9]+` run with `"-"`.
-4. Trim leading and trailing `-`.
-5. Truncate to 60 chars; re-trim trailing `-` after truncation.
-6. Return `""` if nothing remains. Caller substitutes `page-${pageId}`.
-
-### E. Collision rules (`pickAvailableAlias(pages, baseSlug, pageId)`)
-
-- If `baseSlug === ""` → start from `page-${pageId}`.
-- If the candidate is not a key in `pages` → use it.
-- Else append `-2`, `-3`, … until free. (`page-<id>-2` is acceptable in the empty-slug fallback path.)
-
-### F. Slash command (`commands/unic-confluence.md`)
-
-- Step 3 flag list (lines 55-57): add `--no-save — skip auto-saving the alias when publishing by numeric page ID`.
-- New short Step 5 documenting `node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --list`.
-- Step 4 "Report result" (lines 59-62): if the script's stdout contains `✓ Saved alias` or `ℹ Page … already aliased`, surface that line verbatim to the user.
-
-No logic is duplicated — the command remains a pass-through.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `scripts/lib/slug.mjs` | Create — single export `slugify(title)`. |
-| `scripts/lib/pages-file.mjs` | Create — pure helpers: `readPagesFile`, `findAliasForId`, `pickAvailableAlias`, `appendAlias`, `writePagesFile`, `listAliases`. |
-| `scripts/lib/slug.test.mjs` | Create — Node test runner cases for slugifier. |
-| `scripts/lib/pages-file.test.mjs` | Create — Node test runner cases for pure helpers. |
-| `scripts/push-to-confluence.mjs` | Wire `--list`, `--no-save`, and the post-PUT auto-save block. |
-| `commands/unic-confluence.md` | Document `--list`, `--no-save`, surface save-confirmation lines. |
-| `README.md` | New "Auto-aliasing" subsection; document `--list` and `--no-save`; note known concurrency limitation. |
-| `CHANGELOG.md` | Three bullets under `## [Unreleased]` → `### Added` (auto-save, `--list`, `--no-save`). |
-| `docs/plans/README.md` | Append row 22 to the execution table. |
-| `package.json` | Append the two new test files to the `test` script. |
-
-## Implementation steps
-
-### Step 1 — Create `scripts/lib/slug.mjs`
-
-```js
-// @ts-check
-// SPDX-License-Identifier: LGPL-3.0-or-later
-
-const MAX_LEN = 60;
-
-/**
- * Slugify a Confluence page title into a safe alias key for confluence-pages.json.
- * Returns "" when the title produces no slug characters; caller substitutes a fallback.
- *
- * @param {unknown} title
- * @returns {string}
- */
-export function slugify(title) {
-	const raw = String(title ?? "")
-		.normalize("NFD")
-		.replace(/[̀-ͯ]/g, "")
-		.toLowerCase()
-		.replace(/[^a-z0-9]+/g, "-")
-		.replace(/^-+|-+$/g, "");
-	if (!raw) return "";
-	return raw.slice(0, MAX_LEN).replace(/-+$/, "");
-}
-```
-
-### Step 2 — Create `scripts/lib/pages-file.mjs`
-
-```js
-// @ts-check
-// SPDX-License-Identifier: LGPL-3.0-or-later
-import { existsSync, readFileSync, writeFileSync } from "node:fs";
-import path from "node:path";
-
-const FILENAME = "confluence-pages.json";
-
-/**
- * @param {string} cwd
- * @returns {{ pages: Record<string, unknown>, path: string, existed: boolean }}
- */
-export function readPagesFile(cwd) {
-	const p = path.join(cwd, FILENAME);
-	if (!existsSync(p)) return { pages: {}, path: p, existed: false };
-	const pages = /** @type {Record<string, unknown>} */ (
-		JSON.parse(readFileSync(p, "utf8"))
-	);
-	return { pages, path: p, existed: true };
-}
-
-/**
- * @param {Record<string, unknown>} pages
- * @param {number} pageId
- * @returns {string | null}
- */
-export function findAliasForId(pages, pageId) {
-	for (const [k, v] of Object.entries(pages)) {
-		if (k === "_comment") continue;
-		if (typeof v === "number" && v === pageId) return k;
-	}
-	return null;
-}
-
-/**
- * @param {Record<string, unknown>} pages
- * @param {string} baseSlug
- * @param {number} pageId
- * @returns {string}
- */
-export function pickAvailableAlias(pages, baseSlug, pageId) {
-	const start = baseSlug || `page-${pageId}`;
-	if (!(start in pages)) return start;
-	for (let i = 2; ; i++) {
-		const candidate = `${start}-${i}`;
-		if (!(candidate in pages)) return candidate;
-	}
-}
-
-/**
- * Returns a new object with the alias appended, preserving existing key order.
- *
- * @param {Record<string, unknown>} pages
- * @param {string} alias
- * @param {number} pageId
- * @returns {Record<string, unknown>}
- */
-export function appendAlias(pages, alias, pageId) {
-	return { ...pages, [alias]: pageId };
-}
-
-/**
- * @param {string} filePath
- * @param {Record<string, unknown>} pages
- */
-export function writePagesFile(filePath, pages) {
-	writeFileSync(filePath, `${JSON.stringify(pages, null, 2)}\n`, "utf8");
-}
-
-/**
- * @param {Record<string, unknown>} pages
- * @returns {Array<[string, number]>}
- */
-export function listAliases(pages) {
-	const out = /** @type {Array<[string, number]>} */ ([]);
-	for (const [k, v] of Object.entries(pages)) {
-		if (k === "_comment") continue;
-		if (typeof v === "number") out.push([k, v]);
-	}
-	out.sort(([a], [b]) => a.localeCompare(b));
-	return out;
-}
-```
-
-### Step 3 — Create `scripts/lib/slug.test.mjs`
-
-```js
-import { test } from "node:test";
-import assert from "node:assert/strict";
-import { slugify } from "./slug.mjs";
-
-test("plain ASCII title", () => assert.equal(slugify("My Page Title"), "my-page-title"));
-test("strips diacritics", () => assert.equal(slugify("Perfil de l'usuari"), "perfil-de-l-usuari"));
-test("collapses repeated separators", () => assert.equal(slugify("a   b///c"), "a-b-c"));
-test("trims leading/trailing dashes", () => assert.equal(slugify("--Hello--"), "hello"));
-test("CJK only returns empty", () => assert.equal(slugify("漢字"), ""));
-test("emoji only returns empty", () => assert.equal(slugify("🚀✨"), ""));
-test("null returns empty", () => assert.equal(slugify(null), ""));
-test("undefined returns empty", () => assert.equal(slugify(undefined), ""));
-test("truncates at 60 chars", () => {
-	const out = slugify("a".repeat(80));
-	assert.equal(out.length, 60);
-	assert.ok(!out.endsWith("-"));
-});
-test("no trailing dash after truncation at word boundary", () => {
-	const out = slugify("x".repeat(58) + "    " + "y".repeat(10));
-	assert.ok(!out.endsWith("-"));
-	assert.ok(out.length <= 60);
-});
-```
-
-### Step 4 — Create `scripts/lib/pages-file.test.mjs`
-
-```js
-import { test } from "node:test";
-import assert from "node:assert/strict";
-import { mkdtempSync, readFileSync, rmSync } from "node:fs";
-import path from "node:path";
-import os from "node:os";
-import {
-	appendAlias,
-	findAliasForId,
-	listAliases,
-	pickAvailableAlias,
-	readPagesFile,
-	writePagesFile,
-} from "./pages-file.mjs";
-
-test("findAliasForId skips _comment and non-numbers", () => {
-	const pages = { _comment: "x", foo: 111, bar: "not-a-number", baz: 222 };
-	assert.equal(findAliasForId(pages, 222), "baz");
-	assert.equal(findAliasForId(pages, 999), null);
-});
-
-test("pickAvailableAlias returns base when free", () => {
-	assert.equal(pickAvailableAlias({}, "foo", 42), "foo");
-});
-
-test("pickAvailableAlias appends -2 then -3 on collision", () => {
-	const pages = { foo: 1, "foo-2": 2 };
-	assert.equal(pickAvailableAlias(pages, "foo", 9), "foo-3");
-});
-
-test("pickAvailableAlias falls back to page-<id> on empty slug", () => {
-	assert.equal(pickAvailableAlias({}, "", 555), "page-555");
-});
-
-test("appendAlias preserves _comment-first ordering", () => {
-	const pages = { _comment: "x", a: 1, b: 2 };
-	const out = appendAlias(pages, "c", 3);
-	assert.deepEqual(Object.keys(out), ["_comment", "a", "b", "c"]);
-	assert.equal(out["c"], 3);
-});
-
-test("writePagesFile produces 2-space indent with trailing newline", () => {
-	const dir = mkdtempSync(path.join(os.tmpdir(), "pages-"));
-	const p = path.join(dir, "confluence-pages.json");
-	writePagesFile(p, { foo: 1 });
-	assert.equal(readFileSync(p, "utf8"), '{\n  "foo": 1\n}\n');
-	rmSync(dir, { recursive: true, force: true });
-});
-
-test("readPagesFile returns existed=false when absent", () => {
-	const dir = mkdtempSync(path.join(os.tmpdir(), "pages-"));
-	const out = readPagesFile(dir);
-	assert.equal(out.existed, false);
-	assert.deepEqual(out.pages, {});
-	rmSync(dir, { recursive: true, force: true });
-});
-
-test("listAliases sorts alphabetically and filters _comment and non-numbers", () => {
-	const pages = { _comment: "x", zoo: 3, alpha: 1, beta: "no" };
-	assert.deepEqual(listAliases(pages), [["alpha", 1], ["zoo", 3]]);
-});
-```
-
-### Step 5 — Wire the CLI in `scripts/push-to-confluence.mjs`
-
-a. Add imports after the existing `./lib/*` import block:
-
-```js
-import { isNumericId } from "./lib/resolve.mjs";
-import { slugify } from "./lib/slug.mjs";
-import {
-	appendAlias,
-	findAliasForId,
-	listAliases,
-	pickAvailableAlias,
-	readPagesFile,
-	writePagesFile,
-} from "./lib/pages-file.mjs";
-```
-
-b. Add a `runList()` helper near the other top-level helpers (before `main()`):
-
-```js
-function runList() {
-	let res;
-	try {
-		res = readPagesFile(process.cwd());
-	} catch (err) {
-		throw new CliError(
-			`invalid JSON in confluence-pages.json — ${/** @type {Error} */ (err).message}`,
-		);
-	}
-	if (!res.existed) {
-		console.log(`No confluence-pages.json in ${process.cwd()}`);
-		return;
-	}
-	const rows = listAliases(res.pages);
-	if (rows.length === 0) {
-		console.log("No aliases configured yet.");
-		return;
-	}
-	const aliasW = Math.max(5, ...rows.map(([k]) => k.length));
-	console.log(`${"alias".padEnd(aliasW)}  page id`);
-	console.log(`${"─".repeat(aliasW)}  ──────────────`);
-	for (const [k, v] of rows) console.log(`${k.padEnd(aliasW)}  ${v}`);
-}
-```
-
-c. In `main()`, add `--list` dispatch after the `--verify` branch (~line 376):
-
-```js
-if (args[0] === "--list") {
-	runList();
-	return;
-}
-```
-
-d. Extend flag parsing at line ~378:
-
-```js
-const dryRun = args.includes("--dry-run");
-const replaceAll = args.includes("--replace-all");
-const noSave = args.includes("--no-save");
-```
-
-e. Capture `wasNumericArg` right after `pageArg` is read (~line 386), before `resolvePageId`:
-
-```js
-const pageArg = positionalArgs[0] ?? "";
-const wasNumericArg = isNumericId(pageArg);
-```
-
-f. Update the usage string in the `positionalArgs.length < 2` error at line 383:
-
-```
-"Usage: node scripts/push-to-confluence.mjs [--dry-run] [--replace-all] [--no-save] {pageId} {file.md}"
-```
-
-g. Insert the auto-save block immediately before the `✓ Published` log at line 476:
-
-```js
-if (wasNumericArg && !dryRun && !noSave) {
-	try {
-		const { pages, path: pagesPath } = readPagesFile(process.cwd());
-		const existing = findAliasForId(pages, pageId);
-		if (existing) {
-			console.log(`ℹ Page ${pageId} already aliased as "${existing}"`);
-		} else {
-			const alias = pickAvailableAlias(pages, slugify(title), pageId);
-			writePagesFile(pagesPath, appendAlias(pages, alias, pageId));
-			console.log(`✓ Saved alias "${alias}" → ${pageId} in confluence-pages.json`);
-		}
-	} catch (err) {
-		console.error(`⚠ Could not save alias: ${/** @type {Error} */ (err).message}`);
-	}
-}
-```
-
-### Step 6 — Update `commands/unic-confluence.md`
-
-In Step 3 flag list (after `--replace-all`), append:
-
-```markdown
-- `--no-save` — skip auto-saving the alias when publishing by numeric page ID
-```
-
-In Step 4 "Report result", append a sentence:
-
-```markdown
-If stdout includes `✓ Saved alias` or `ℹ Page <id> already aliased as`, surface that line verbatim to the user.
-```
-
-After Step 4, add a new Step 5:
-
-```markdown
-### 5. List configured aliases
-
-To list all configured aliases in the current repo:
-
-```sh
-node "${CLAUDE_PLUGIN_ROOT}/scripts/push-to-confluence.mjs" --list
-```
-
-Prints a two-column table of `alias → page id` from `confluence-pages.json`. Reports a friendly message if the file is missing or has no aliases.
-```
-
-### Step 7 — README "Auto-aliasing" subsection
-
-Add after the CLI usage block in the "Usage" section:
-
-```markdown
-### Auto-aliasing
-
-Publishing by raw numeric page ID auto-saves a slugified alias derived from the Confluence page's title into `confluence-pages.json` at the current working directory. Subsequent publishes can reference the page by its alias.
-
-- The slug is lowercased, ASCII-only, dash-separated, max 60 chars.
-- If the slug collides with another alias, `-2`, `-3`, … is appended.
-- If the page ID is already aliased, no write happens (the existing alias is reported).
-- Pass `--no-save` to skip auto-saving for a single run.
-
-List all aliases:
-
-```sh
-node scripts/push-to-confluence.mjs --list
-```
-
-**Known limitation:** the file write is not atomic. Two concurrent publishes against the same `confluence-pages.json` may race; one update will be lost.
-```
-
-### Step 8 — CHANGELOG
-
-Under `## [Unreleased]` → `### Added`, replace `- (none)` with:
-
-```markdown
-- Auto-saves a slugified alias into `confluence-pages.json` after publishing by raw numeric page ID; reports the new alias to the user.
-- `--list` flag prints configured aliases as a sorted two-column table.
-- `--no-save` flag opts out of alias auto-saving.
-```
-
-### Step 9 — `package.json` test script
-
-Replace the `"test"` entry value:
-
-```json
-"test": "node --test scripts/lib/frontmatter.test.mjs scripts/lib/inject.test.mjs scripts/lib/resolve.test.mjs scripts/lib/slug.test.mjs scripts/lib/pages-file.test.mjs"
-```
-
-### Step 10 — Update `docs/plans/README.md`
-
-Append after the spec 21 row in the execution table:
-
-```markdown
-| 22 | [Auto-populate confluence-pages.json aliases](./22-auto-populate-aliases.md) | P1 | — |
-```
-
-### Step 11 — Bump and verify
-
-```sh
-pnpm typecheck
-pnpm lint
-pnpm test
-pnpm bump minor
-pnpm verify:changelog
-git add -A
-git commit -m "feat(spec-22): auto-populate confluence-pages.json aliases (vX.Y.Z)"
-```
-
-## Test cases
-
-| Scenario | Action | Expected outcome |
-|---|---|---|
-| First publish by numeric ID, no file | `node push-to-confluence.mjs 12345 doc.md` (cwd has no `confluence-pages.json`) | After PUT: `confluence-pages.json` created with `{ "<slug>": 12345 }`; stdout includes `✓ Saved alias`. |
-| First publish by numeric ID, existing file with `_comment` | Same, but file pre-populated with `{ "_comment": "x", "other": 999 }` | New key appended after `other`; `_comment` still first; stdout includes `✓ Saved alias`. |
-| Slug collision | File already has `{ "my-page": 111 }`; publish ID 222 whose title slugifies to `my-page` | New entry `"my-page-2": 222`; stdout reports the `-2` alias. |
-| Page ID already aliased | File has `{ "alpha": 12345 }`; publish 12345 again | No write; stdout: `ℹ Page 12345 already aliased as "alpha"`. |
-| Empty slug fallback | Title is `🚀✨`; publish ID 777 | Alias becomes `page-777`. |
-| `--dry-run` with numeric ID | `--dry-run 12345 doc.md` | No write; no `✓ Saved alias` log; existing dry-run block runs. |
-| `--no-save` with numeric ID | `--no-save 12345 doc.md` | PUT succeeds; file untouched; no save log. |
-| Publish by KEY (not numeric) | `node push-to-confluence.mjs spec-1 doc.md` | No auto-save attempted; existing behaviour unchanged. |
-| `--list` with no file | `--list` in clean cwd | Prints `No confluence-pages.json in <cwd>`; exit 0. |
-| `--list` with only `_comment` | `--list` against `{ "_comment": "x" }` | Prints `No aliases configured yet.`; exit 0. |
-| `--list` populated | `--list` against three entries | Prints alphabetised table; `_comment` filtered. |
-| Save failure non-fatal | Make `confluence-pages.json` read-only, publish numeric | PUT succeeds; stderr has `⚠ Could not save alias`; exit code is still 0. |
-| Malformed JSON | Existing file is invalid JSON, publish numeric | Stderr `⚠ Could not save alias: …`; exit 0; file untouched. |
-
-## Verification
-
-```sh
-# 1. Pure-function tests
-pnpm test
-# Expected: all new slug + pages-file cases pass; existing suite green.
-
-# 2. Type + lint
-pnpm typecheck && pnpm lint
-
-# 3. Manual --list (no auth needed)
-mkdir -p /tmp/confl-list-test && cd /tmp/confl-list-test
-node /path/to/scripts/push-to-confluence.mjs --list
-# Expected: "No confluence-pages.json in /tmp/confl-list-test"
-
-echo '{"_comment":"x","spec-1":111,"alpha":222}' > confluence-pages.json
-node /path/to/scripts/push-to-confluence.mjs --list
-# Expected: two-row table sorted (alpha, spec-1).
-
-# 4. End-to-end auto-save (requires real Confluence creds + test page)
-mkdir -p /tmp/confl-save-test && cd /tmp/confl-save-test
-node /path/to/scripts/push-to-confluence.mjs <real-test-page-id> /path/to/some.md
-cat confluence-pages.json
-# Expected: file contains a slugified alias → <real-test-page-id>.
-
-# Run again:
-node /path/to/scripts/push-to-confluence.mjs <real-test-page-id> /path/to/some.md
-# Expected: stdout has "ℹ Page <id> already aliased as ..."
-
-# 5. Changelog gate
-pnpm verify:changelog
-# Expected: passes.
-```
-
-## Acceptance criteria
-
-- [ ] `docs/plans/README.md` execution table includes a row for spec 22.
-- [ ] `scripts/lib/slug.mjs` + `scripts/lib/pages-file.mjs` created, JSDoc-typed, `// @ts-check`'d.
-- [ ] `pnpm test` runs the two new test files; all cases pass; existing tests still pass.
-- [ ] `--list` prints sorted aliases with `_comment` filtered; handles missing/empty file gracefully.
-- [ ] Publishing by numeric ID without `--dry-run`/`--no-save` writes a slugified alias and logs `✓ Saved alias …`.
-- [ ] Publishing by numeric ID when the page is already aliased logs `ℹ Page <id> already aliased as "<key>"` and does NOT rewrite the file.
-- [ ] Publishing by KEY (non-numeric) does NOT write the file.
-- [ ] `--dry-run` and `--no-save` paths do NOT write the file.
-- [ ] Save failures are non-fatal: warning on stderr, exit code 0.
-- [ ] `_comment` key (if present) stays first in `confluence-pages.json` after writes.
-- [ ] `commands/unic-confluence.md` documents `--list` and `--no-save`; the command remains a pass-through.
-- [ ] `README.md` has the "Auto-aliasing" subsection with the known-concurrency limitation noted.
-- [ ] `CHANGELOG.md` `[Unreleased]` has the three `### Added` bullets.
-- [ ] `pnpm bump minor` lands the version bump; `pnpm verify:changelog` passes.
-
-## Out of scope
-
-- Writing `confluence-pages.json` on publish-by-key (already aliased by definition).
-- Any schema change beyond `key → integer` (titles, timestamps, nested objects). Explicitly forbidden by `CLAUDE.md` lines 74-76 without maintainer sign-off.
-- Renaming or deleting aliases via CLI — hand-edit the file.
-- Atomic file writes / cross-process locking. The known concurrency limitation is documented; not fixed.
-- A `package.json` `confluence:list` shortcut. Consumer repos add their own per `CLAUDE.md` line 23.
-- Any change to the slash command behaviour beyond documentation — the command stays a pure pass-through.
-- Slug locale customisation (no `localeCompare` collation, no per-language stemming). The slug is ASCII-stripped Unicode normalisation, full stop.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/23-fix-ci-pnpm-version-conflict.md b/apps/claude-code/unic-confluence/docs/plans/23-fix-ci-pnpm-version-conflict.md
deleted file mode 100644
index a9d6d32d..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/23-fix-ci-pnpm-version-conflict.md
+++ /dev/null
@@ -1,238 +0,0 @@
-# 23. Fix CI pnpm version conflict
-**Status: done — 2026-04-27**
-
-**Priority:** P0
-**Effort:** S
-**Version impact:** patch
-**Depends on:** spec 22 (last sequential entry; no real dependency)
-**Touches:** `.github/workflows/ci.yml`, `CHANGELOG.md`, `docs/plans/README.md`
-
-## Context
-
-CI is currently broken on `feature/autopopulate-confluence-pages-aliases` (and will break on every PR / push to `main`) with:
-
-```
-Error: Error: Multiple versions of pnpm specified:
-  - version 10 in the GITHUB_ACTION_INPUT_VERSION environment variable
-  - version pnpm@10.24.0 in the package.json#packageManager
-Remove one of these to continue.
-```
-
-Failing run: <https://github.com/unic/unic-claude-code-confluence/actions/runs/24897403641/job/72905826220>
-
-The conflict came in implicitly: spec `00` set up `pnpm/action-setup@v4` with `version: 10`, and a later commit (around the spec-19/20 versioning work) added `"packageManager": "pnpm@10.24.0"` to `package.json`. `pnpm/action-setup@v4` ≥ June 2024 enforces "single source of truth" and errors out when both are declared, instead of silently picking one.
-
-The right fix is to **drop the workflow `version` input** and let `package.json#packageManager` be the canonical pin. Reasons:
-
-1. `package.json#packageManager` is the npm-defined, tool-agnostic standard (Corepack, Volta, Yarn, pnpm all read it).
-2. It already exists in the repo, is exact-pinned (`10.24.0`), and is what local developers use.
-3. `pnpm bump` and Renovate/dependabot bumps update `package.json` — keeping CI in sync with the workflow file would require a second update site.
-
-`cache: pnpm` on `actions/setup-node@v4` is unaffected by this change — it only requires that pnpm is on `PATH` by the time `setup-node` runs, which `pnpm/action-setup` (without `version`) still guarantees.
-
-## Current behaviour
-
-`.github/workflows/ci.yml:16-23`:
-
-```yaml
-- uses: actions/checkout@v4
-- uses: pnpm/action-setup@v4
-  with:
-    version: 10
-- uses: actions/setup-node@v4
-  with:
-    node-version: ${{ matrix.node-version }}
-    cache: pnpm
-```
-
-`package.json:6` (single source of truth that the workflow conflicts with):
-
-```json
-"packageManager": "pnpm@10.24.0",
-```
-
-Result: every job in the matrix (`node-version: ["22", "24"]`) fails at the `pnpm/action-setup@v4` step before any install / lint / test runs.
-
-## Target behaviour
-
-### A. Workflow defers to `package.json#packageManager`
-
-`.github/workflows/ci.yml` no longer passes a `version` input to `pnpm/action-setup@v4`. The action reads the version from `package.json#packageManager` (its documented fallback when no `version` is provided). The matrix job runs `pnpm install --frozen-lockfile`, `pnpm ci:check`, `pnpm test`, and (on PRs) `pnpm verify:changelog` to green.
-
-### B. CHANGELOG records the fix
-
-A bullet under `## [Unreleased]` → `### Fixed` documents the CI repair in user-facing language. After `pnpm bump patch` runs, the bullet is promoted under a new dated heading.
-
-### C. Plan README execution table updated
-
-`docs/plans/README.md` execution-order table gains a row 23 with the new spec's link and `P0` priority.
-
-### D. Plan README ground rule prevents regression
-
-`docs/plans/README.md` "Ground rules" section gains a bullet stating that `package.json#packageManager` is the single source of truth for the pnpm version, that CI workflows must NOT pass a `version` input to `pnpm/action-setup`, and that bumping pnpm is done by editing `packageManager` (after which `pnpm install` regenerates the lockfile). This converts the one-off fix into an enforceable convention so a future contributor doesn't reintroduce the conflict.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `.github/workflows/ci.yml` | Remove the `with: version: 10` block from the `pnpm/action-setup@v4` step |
-| `CHANGELOG.md` | Add `### Fixed` bullet under `## [Unreleased]`; promoted by `pnpm bump patch` |
-| `docs/plans/README.md` | Append row 23 to the execution-order table; add a ground-rule bullet about pnpm version pinning |
-
-## Implementation steps
-
-### Step 1 — Trim the workflow
-
-Edit `.github/workflows/ci.yml`. Replace lines 17-19:
-
-```yaml
-      - uses: pnpm/action-setup@v4
-        with:
-          version: 10
-```
-
-with:
-
-```yaml
-      - uses: pnpm/action-setup@v4
-```
-
-Keep tabs/spaces consistent with the surrounding YAML (the existing file uses 2-space indentation inside `steps:` — match it; do **not** convert to tabs even though the repo's `.editorconfig` says tabs, because YAML in this file already uses spaces and a mid-file switch would break parsing).
-
-The final shape of the `steps:` block must be exactly:
-
-```yaml
-    steps:
-      - uses: actions/checkout@v4
-      - uses: pnpm/action-setup@v4
-      - uses: actions/setup-node@v4
-        with:
-          node-version: ${{ matrix.node-version }}
-          cache: pnpm
-      - run: pnpm install --frozen-lockfile
-      - run: pnpm ci:check
-      - run: pnpm test
-      - run: pnpm verify:changelog
-        if: github.event_name == 'pull_request'
-```
-
-### Step 2 — Add CHANGELOG entry
-
-Open `CHANGELOG.md`. Under `## [Unreleased]` → `### Fixed`, replace `- (none)` (or append, if other entries already exist) with:
-
-```markdown
-- CI: drop redundant `version: 10` from `pnpm/action-setup@v4` so `package.json#packageManager` is the single source of truth (CI was failing with "Multiple versions of pnpm specified")
-```
-
-### Step 3 — Update `docs/plans/README.md`
-
-3a. After the row for spec 22 in the "Execution order" table, append:
-
-```markdown
-| 23 | [Fix CI pnpm version conflict](./23-fix-ci-pnpm-version-conflict.md) | P0 | — |
-```
-
-3b. In the "Ground rules" section, insert a new bullet immediately after the existing `**Package manager**:` bullet:
-
-```markdown
-- **pnpm version pinning**: `package.json#packageManager` (e.g. `pnpm@10.24.0`) is the single source of truth. CI workflows MUST NOT pass a `version` input to `pnpm/action-setup` — the action reads `packageManager` automatically. To bump pnpm: edit `packageManager` in `package.json`, run `pnpm install` to regenerate the lockfile, and commit both. Never declare the pnpm version in two places.
-```
-
-### Step 4 — Bump and verify
-
-Run, in order:
-
-```sh
-pnpm install --frozen-lockfile     # sanity check that local install still works
-pnpm ci:check                       # lint + format + typecheck (matches CI)
-pnpm test                           # full suite
-pnpm bump patch                     # increments plugin.json, mirrors marketplace.json, promotes CHANGELOG
-pnpm verify:changelog               # confirms version bump + CHANGELOG entry are in sync
-```
-
-`pnpm bump patch` will turn the current `2.0.1` into `2.0.2` (or whatever the current head is — read `.claude-plugin/plugin.json` first; do not assume).
-
-### Step 5 — Mark done and commit
-
-Prepend the H1 with `**Status: done — <YYYY-MM-DD>**`, then:
-
-```sh
-git add .github/workflows/ci.yml CHANGELOG.md docs/plans/README.md \
-        docs/plans/23-fix-ci-pnpm-version-conflict.md \
-        .claude-plugin/plugin.json .claude-plugin/marketplace.json
-git commit -m "fix(spec-23): drop redundant pnpm version pin from CI (vX.Y.Z)"
-```
-
-Replace `X.Y.Z` with the version output by `pnpm bump`.
-
-## Test cases
-
-| Scenario | Action | Expected outcome |
-|---|---|---|
-| Push to feature branch | `git push` | CI workflow starts; `Setup pnpm` step succeeds (no "Multiple versions" error) |
-| `cache: pnpm` still resolves | CI runs `actions/setup-node@v4` step | `setup-node` finds pnpm on PATH; cache key derived correctly |
-| `pnpm install --frozen-lockfile` | CI runs install | Resolves with pnpm 10.24.0 (from `packageManager`); lockfile honoured |
-| Node 22 matrix leg | CI runs full pipeline on Node 22 | `pnpm ci:check`, `pnpm test`, and (on PR) `pnpm verify:changelog` all pass |
-| Node 24 matrix leg | CI runs full pipeline on Node 24 | Same as Node 22 leg |
-| Local dev unaffected | Developer runs `pnpm install` locally | Uses pnpm 10.24.0 (from `packageManager`); no behaviour change |
-| Future pnpm bump | Bump `packageManager` to e.g. `pnpm@10.25.0` | CI picks up the new version on next push without touching the workflow file |
-
-## Verification
-
-```sh
-# 1. Confirm the version arg is gone from the workflow
-grep -n "version: 10" .github/workflows/ci.yml
-# Expected: no matches
-
-grep -nA1 "pnpm/action-setup" .github/workflows/ci.yml
-# Expected: the line "uses: pnpm/action-setup@v4" with no "with:" block beneath it
-
-# 2. Confirm packageManager is still authoritative
-grep -n "packageManager" package.json
-# Expected: "packageManager": "pnpm@10.24.0",
-
-# 3. Confirm CHANGELOG entry exists under Unreleased > Fixed
-sed -n '/## \[Unreleased\]/,/## \[/p' CHANGELOG.md | grep -i "Multiple versions"
-# Expected: at least one match
-
-# 4. Confirm execution-order row added
-grep -n "23 |" docs/plans/README.md
-# Expected: a row pointing at ./23-fix-ci-pnpm-version-conflict.md
-
-# 4b. Confirm ground-rule bullet added
-grep -n "pnpm version pinning" docs/plans/README.md
-# Expected: one match in the "Ground rules" section
-
-# 5. Local mirror of CI
-pnpm install --frozen-lockfile
-pnpm ci:check
-pnpm test
-pnpm verify:changelog
-# Expected: all four exit 0
-
-# 6. After commit + push, confirm GH Actions green
-gh run watch --exit-status
-# Expected: workflow concludes with conclusion=success on both Node 22 and Node 24 legs
-```
-
-## Acceptance criteria
-
-- [ ] `.github/workflows/ci.yml` has no `version:` input under `pnpm/action-setup@v4`.
-- [ ] `package.json#packageManager` remains `pnpm@10.24.0` (untouched).
-- [ ] `CHANGELOG.md` has a `### Fixed` bullet describing the CI repair (promoted by `pnpm bump patch` to a dated section).
-- [ ] `.claude-plugin/plugin.json` and `.claude-plugin/marketplace.json` versions match and are one patch above the previous release.
-- [ ] `docs/plans/README.md` execution-order table has a row 23 for this spec.
-- [ ] `docs/plans/README.md` "Ground rules" section has a bullet declaring `package.json#packageManager` as the single source of truth for the pnpm version and forbidding a `version` input to `pnpm/action-setup`.
-- [ ] `pnpm install --frozen-lockfile`, `pnpm ci:check`, `pnpm test`, and `pnpm verify:changelog` all pass locally.
-- [ ] After push, the GitHub Actions run on this branch is green for both `node-version: 22` and `node-version: 24`.
-- [ ] No source code under `scripts/` is modified.
-
-## Out of scope
-
-- **Pinning the action by SHA.** Switching `pnpm/action-setup@v4` to a SHA pin would harden supply-chain posture but is unrelated to the breakage and would expand the diff. Track separately if desired.
-- **Bumping pnpm itself.** Stay on `10.24.0`. A pnpm minor/major bump is a separate decision with separate risk (lockfile churn, `pnpm-workspace.yaml` schema).
-- **Migrating to Corepack-only setup.** Replacing `pnpm/action-setup` with `corepack enable` + `actions/setup-node`'s native pnpm support is viable but is a larger refactor and changes how `cache: pnpm` resolves. Out of scope.
-- **Adding a Renovate / Dependabot config to keep `packageManager` fresh.** Useful, but a separate spec.
-- **Changing the matrix.** `node-version: ["22", "24"]` stays as-is. Engines say `>=24`; 22 is kept as a forward-compat smoke test.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/24-fix-catalog-leak-in-git-install.md b/apps/claude-code/unic-confluence/docs/plans/24-fix-catalog-leak-in-git-install.md
deleted file mode 100644
index 0f8c09d9..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/24-fix-catalog-leak-in-git-install.md
+++ /dev/null
@@ -1,311 +0,0 @@
-# 24. Fix `catalog:` specifier leak in git-install consumers
-**Status: done — 2026-04-27**
-
-**Priority:** P0
-**Effort:** S
-**Version impact:** patch
-**Depends on:** none (orthogonal to spec 23)
-**Touches:** `package.json`, `pnpm-workspace.yaml`, `pnpm-lock.yaml`, `CHANGELOG.md`, `docs/plans/README.md`
-
-## Context
-
-A downstream project (`agentic-delivery-model`) tries to install this plugin as an npm dep via the documented git-URL install path:
-
-```sh
-pnpm install -w -D git+ssh://git@github.com:unic/unic-claude-code-confluence
-```
-
-…and crashes immediately:
-
-```
-ERR_PNPM_SPEC_NOT_SUPPORTED_BY_ANY_RESOLVER  marked@catalog: isn't supported by any available resolver.
-
-This error happened while installing the dependencies of unic-confluence@0.0.0
-
-An external package outside of the pnpm workspace declared a dependency using
-the catalog protocol. This is likely a bug in that external package. Only
-packages within the pnpm workspace may use catalogs. Usages of the catalog
-protocol are replaced with real specifiers on 'pnpm publish'.
-```
-
-Root cause: `unic-confluence` ships `"marked": "catalog:"` in its runtime `dependencies` (`package.json:26`). The `catalog:` protocol is a pnpm workspace-internal feature. When the package is consumed via `pnpm publish` it is rewritten to a concrete version automatically; when consumed via a git URL (the install path documented in `README.md:121,127,224,227`), pnpm clones the repo verbatim and reads `package.json` as-is — so the literal string `catalog:` reaches the downstream resolver, which has no workspace catalog to consult and aborts the install. `npm install <git-url>` fails the same way (npm has no catalog support at all).
-
-This repo never publishes to npm. Evidence:
-
-- `package.json#private: true` (`package.json:3`)
-- no `"files"`, no `"publishConfig"`, no `"main"`/`"bin"`/`"exports"` field
-- no `prepare` / `prepack` / `publish` script in `package.json#scripts`
-- no `npm publish` / `pnpm publish` step in `.github/workflows/ci.yml`
-- README "Install as npm package" section instructs consumers to use `git+ssh://…` (`README.md:121,127`)
-
-The only supported install path is the git URL. Therefore runtime dependency specifiers must be valid in isolation — they cannot rely on workspace context that consumers do not share.
-
-**Why not just adopt `pnpm publish`?** That is the canonical fix recommended by the pnpm error itself, but the change surface is much larger: drop `private: true`, define a `"files"` allowlist plus `"publishConfig"`, add a release job with an `NPM_TOKEN` secret, change README install instructions, coordinate the consumer migration. Track separately as a future spec; this one unblocks the consumer today with a one-line dep change.
-
-**Why not a `prepare` script that rewrites `catalog:` at install time?** pnpm and npm do run `prepare` for git deps, but the rewrite would need a YAML parser to read `pnpm-workspace.yaml` (and that parser itself can't be `catalog:`), would mutate `package.json` mid-install, and would behave inconsistently across npm/pnpm/yarn. The complexity exceeds the benefit when the actual workspace has exactly one runtime dep.
-
-**Why is keeping devDeps on `catalog:` safe?** Downstream `pnpm install <git-url>` only resolves the package's `dependencies` (and `peerDependencies` / `optionalDependencies`). `devDependencies` are skipped for non-root packages. Local development still happens inside the workspace, so `catalog:` continues to resolve normally for devDeps. The leak is strictly a runtime-dep problem.
-
-## Current behaviour
-
-`package.json:25-33`:
-
-```json
-"dependencies": {
-    "marked": "catalog:"
-},
-"devDependencies": {
-    "@biomejs/biome": "catalog:",
-    "@ralph-orchestrator/ralph-cli": "catalog:",
-    "@types/node": "catalog:",
-    "typescript": "catalog:"
-}
-```
-
-`pnpm-workspace.yaml:7-12`:
-
-```yaml
-catalog:
-  "@biomejs/biome": 2.4.0
-  "@ralph-orchestrator/ralph-cli": 2.9.2
-  "@types/node": 24.12.2
-  marked: 17.0.5
-  typescript: 5.8.3
-```
-
-Result: any downstream `pnpm install git+ssh://...unic-claude-code-confluence` (or the `npm install` equivalent) fails at the resolver step with `ERR_PNPM_SPEC_NOT_SUPPORTED_BY_ANY_RESOLVER  marked@catalog:`. The plugin is unconsumable as an npm dependency.
-
-## Target behaviour
-
-### A. Runtime deps pinned directly, not via catalog
-
-`package.json#dependencies.marked` becomes the exact version string `17.0.5` (the value previously held in the catalog). The catalog entry is removed because no other package references it (this repo is a single-package workspace).
-
-After the change, downstream `pnpm install git+ssh://...` resolves `marked@17.0.5` straight from npm — no workspace context required.
-
-### B. devDeps stay on `catalog:`
-
-`@biomejs/biome`, `@ralph-orchestrator/ralph-cli`, `@types/node`, and `typescript` keep their `catalog:` specifiers. Consumers don't install transitive devDeps, so the leak doesn't reach them. Local development continues to use the catalog as the single source of truth for devDep versions.
-
-### C. Ground rule prevents regression
-
-`docs/plans/README.md` "Ground rules" section gains a bullet stating that `dependencies`, `peerDependencies`, and `optionalDependencies` MUST use concrete pinned versions (never `catalog:` or `workspace:`), because the package is consumed via git URL. Only `devDependencies` may use `catalog:`.
-
-This converts the one-off fix into an enforceable convention so a future contributor doesn't reintroduce the same leak the next time a runtime dep is added.
-
-### D. CHANGELOG records the fix
-
-A bullet under `## [Unreleased]` → `### Fixed` documents the unblock for downstream consumers in user-facing language. `pnpm bump patch` promotes it under a new dated heading.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `package.json` | Replace `"marked": "catalog:"` with `"marked": "17.0.5"` in `dependencies` |
-| `pnpm-workspace.yaml` | Remove the `marked: 17.0.5` line from the `catalog:` block |
-| `pnpm-lock.yaml` | Regenerated automatically by `pnpm install` after the edits — commit the diff |
-| `CHANGELOG.md` | Add `### Fixed` bullet under `## [Unreleased]`; promoted by `pnpm bump patch` |
-| `docs/plans/README.md` | Append row 24 to the execution-order table; add a ground-rule bullet about runtime-dep pinning |
-
-No source under `scripts/` is touched — `scripts/push-to-confluence.mjs` continues to `import { marked } from 'marked'` with zero behaviour change.
-
-## Implementation steps
-
-### Step 1 — De-catalog `marked` in `package.json`
-
-Read the current catalog value first (do not assume `17.0.5` if the catalog has drifted by the time this spec runs):
-
-```sh
-grep -E "^\s+marked:" pnpm-workspace.yaml
-```
-
-Then edit `package.json:25-27`. Replace:
-
-```json
-"dependencies": {
-    "marked": "catalog:"
-},
-```
-
-with (substituting the catalog value you just read):
-
-```json
-"dependencies": {
-    "marked": "17.0.5"
-},
-```
-
-`package.json` indentation is **two spaces** (project convention for `.json` regardless of `.editorconfig`'s tab default — match the existing file).
-
-### Step 2 — Remove `marked` from the catalog
-
-Edit `pnpm-workspace.yaml`. Remove the line:
-
-```yaml
-  marked: 17.0.5
-```
-
-from the `catalog:` block (lines 7-12). The remaining four catalog entries (`@biomejs/biome`, `@ralph-orchestrator/ralph-cli`, `@types/node`, `typescript`) stay untouched and pinned exactly. Indentation in this file is two spaces (YAML convention) — preserve it.
-
-### Step 3 — Refresh the lockfile
-
-```sh
-pnpm install
-```
-
-This rewrites `pnpm-lock.yaml` to record `marked@17.0.5` as a direct registry dep instead of a catalog-resolved one. Commit the updated lockfile alongside `package.json` and `pnpm-workspace.yaml`.
-
-### Step 4 — Add CHANGELOG entry
-
-Open `CHANGELOG.md`. Under `## [Unreleased]` → `### Fixed`, replace `- (none)` (or append, if other entries already exist) with:
-
-```markdown
-- Pin `marked` directly in `package.json#dependencies` instead of `catalog:` so the plugin can be installed via git URL by external pnpm/npm projects (was failing with `ERR_PNPM_SPEC_NOT_SUPPORTED_BY_ANY_RESOLVER  marked@catalog:` because `catalog:` is a pnpm workspace-internal feature that only `pnpm publish` rewrites, and this plugin is consumed via git URL, not the registry)
-```
-
-### Step 5 — Update `docs/plans/README.md`
-
-5a. After the row for spec 23 in the "Execution order" table, append:
-
-```markdown
-| 24 | [Fix `catalog:` leak in git-install consumers](./24-fix-catalog-leak-in-git-install.md) | P0 | — |
-```
-
-5b. In the "Ground rules" section, immediately after the existing `- **Dep versioning**: …` bullet, insert:
-
-```markdown
-- **Runtime dep specifiers**: `dependencies`, `peerDependencies`, and `optionalDependencies` in `package.json` MUST use exact pinned versions (e.g. `"marked": "17.0.5"`) — never `catalog:` or `workspace:`. The plugin is consumed via git URL (`git+ssh://...`), so consumers don't share our pnpm workspace and cannot resolve catalog specifiers; pnpm only rewrites `catalog:` at `pnpm publish` time, which this repo does not run. `devDependencies` may continue to use `catalog:` because consumers don't install transitive devDeps. To bump a runtime dep: edit `package.json` directly; do not add it to the catalog.
-```
-
-This bullet is the regression guard. Without it, the next runtime dep added to the project will reintroduce the same break.
-
-### Step 6 — Local verification mirroring downstream
-
-```sh
-# 6a. Finishing-your-work sequence (per CONTRIBUTING.md)
-pnpm install --frozen-lockfile
-pnpm format          # write any formatting fixes
-pnpm check           # biome lint + format check (must be clean before bump)
-pnpm test            # full test suite
-pnpm typecheck       # JSDoc / @ts-check
-
-# 6b. Simulate a downstream git-URL consumer to confirm the leak is fixed.
-#     Use a clean tmpdir so we don't pollute the workspace.
-repo=$(git -C /Users/oriol.torrent/Sites/UNIC/unic-claude-code-confluence rev-parse --show-toplevel)
-tmp=$(mktemp -d) && cd "$tmp"
-pnpm init >/dev/null
-pnpm add -D "git+file://$repo"
-# Expected:
-#   - install succeeds (no ERR_PNPM_SPEC_NOT_SUPPORTED_BY_ANY_RESOLVER)
-#   - node_modules/unic-confluence/package.json#dependencies.marked === "17.0.5"
-#   - node_modules/marked exists
-node -e "console.log(require('./node_modules/unic-confluence/package.json').dependencies)"
-test -d node_modules/marked && echo OK
-cd - && rm -rf "$tmp"
-```
-
-### Step 7 — Bump, mark done, and commit
-
-```sh
-pnpm bump patch                 # increments plugin.json, mirrors marketplace.json, promotes CHANGELOG
-pnpm verify:changelog           # confirms version bump + CHANGELOG entry coherent
-```
-
-Then prepend the H1 of this spec with `**Status: done — <YYYY-MM-DD>**` and stage all changed files:
-
-```sh
-git add package.json pnpm-workspace.yaml pnpm-lock.yaml \
-        CHANGELOG.md docs/plans/README.md \
-        docs/plans/24-fix-catalog-leak-in-git-install.md \
-        .claude-plugin/plugin.json .claude-plugin/marketplace.json
-git commit -m "fix(spec-24): pin marked directly so git-URL consumers can install (vX.Y.Z)"
-```
-
-Replace `X.Y.Z` with the version output by `pnpm bump`.
-
-## Test cases
-
-| Scenario | Action | Expected outcome |
-|---|---|---|
-| Local install in workspace | `pnpm install --frozen-lockfile` | Lockfile honoured; `marked@17.0.5` resolves directly from registry |
-| Local lint/test/typecheck | `pnpm ci:check && pnpm test && pnpm typecheck` | All pass; `scripts/push-to-confluence.mjs` still imports `marked` correctly |
-| Downstream pnpm git install | `pnpm add git+ssh://git@github.com:unic/unic-claude-code-confluence` from a non-workspace project | Install succeeds; `marked` resolves to `17.0.5` in the consumer's `node_modules` |
-| Downstream npm git install | `npm install git+ssh://git@github.com:unic/unic-claude-code-confluence` | Install succeeds (npm previously also broke on `catalog:`; pinning fixes both) |
-| Real `agentic-delivery-model` install | Re-run `pnpm install -w -D git+ssh://…` in `~/Sites/UNIC/agentic-delivery-model` | No `ERR_PNPM_SPEC_NOT_SUPPORTED_BY_ANY_RESOLVER` |
-| devDep `catalog:` still works locally | `pnpm exec biome --version`, `pnpm exec tsc --version` | Versions match `pnpm-workspace.yaml#catalog` entries |
-| Ground-rule visible to future contributors | `grep "Runtime dep specifiers" docs/plans/README.md` | One match in "Ground rules" section |
-| Markdown rendering still works | Run `pnpm confluence --dry-run <key> README.md` against any page | `marked` converts MD → HTML as before |
-
-## Verification
-
-```sh
-# 1. Confirm marked is pinned directly
-grep -n '"marked"' package.json
-# Expected: "marked": "17.0.5",   (no `catalog:`)
-
-# 2. Confirm marked is NOT in the catalog anymore
-grep -nE "^\s+marked:" pnpm-workspace.yaml
-# Expected: no matches
-
-# 3. Confirm devDeps still use catalog (count of remaining `catalog:` literals in package.json)
-grep -c '"catalog:"' package.json
-# Expected: 4 (the four devDep entries; runtime deps no longer use catalog:)
-
-# 4. Confirm lockfile reflects the change
-grep -nE "^\s+marked:" pnpm-lock.yaml | head
-# Expected: an entry under `importers > .` listing marked at 17.0.5 from the registry, not via catalog
-
-# 5. Confirm CHANGELOG entry exists under Unreleased > Fixed
-sed -n '/## \[Unreleased\]/,/## \[/p' CHANGELOG.md | grep -i "catalog"
-# Expected: at least one match describing the fix
-
-# 6. Confirm execution-order row added
-grep -n "24 |" docs/plans/README.md
-# Expected: row pointing at ./24-fix-catalog-leak-in-git-install.md
-
-# 7. Confirm ground-rule bullet added
-grep -n "Runtime dep specifiers" docs/plans/README.md
-# Expected: one match in the "Ground rules" section
-
-# 8. End-to-end: simulate a git-URL consumer in a tmpdir
-repo=$(git -C /Users/oriol.torrent/Sites/UNIC/unic-claude-code-confluence rev-parse --show-toplevel)
-tmp=$(mktemp -d) && cd "$tmp" && pnpm init >/dev/null && \
-  pnpm add -D "git+file://$repo" && \
-  node -e "console.log(require('./node_modules/unic-confluence/package.json').dependencies.marked)" && \
-  test -d node_modules/marked && cd - && rm -rf "$tmp"
-# Expected: prints "17.0.5"; node_modules/marked exists; whole script exits 0
-
-# 9. Verify the actual downstream consumer
-cd /Users/oriol.torrent/Sites/UNIC/agentic-delivery-model && pnpm install
-# Expected: no ERR_PNPM_SPEC_NOT_SUPPORTED_BY_ANY_RESOLVER
-
-# 10. CI smoke after push
-gh run watch --exit-status
-# Expected: green on both Node 22 and Node 24 legs
-```
-
-## Acceptance criteria
-
-- [ ] `package.json#dependencies.marked` is a concrete version string (matching the value previously in the catalog), not `"catalog:"`.
-- [ ] `pnpm-workspace.yaml#catalog` block no longer contains a `marked` entry.
-- [ ] `pnpm-lock.yaml` reflects `marked` resolved as a direct registry dep, regenerated by `pnpm install`.
-- [ ] All four devDeps (`@biomejs/biome`, `@ralph-orchestrator/ralph-cli`, `@types/node`, `typescript`) still use `"catalog:"` and resolve via the catalog.
-- [ ] `CHANGELOG.md` has a `### Fixed` bullet describing the unblock (promoted by `pnpm bump patch` to a dated section).
-- [ ] `.claude-plugin/plugin.json` and `.claude-plugin/marketplace.json` versions match and are one patch above the previous release.
-- [ ] `docs/plans/README.md` execution-order table has a row 24 for this spec.
-- [ ] `docs/plans/README.md` "Ground rules" section has a bullet declaring that runtime-dep specifiers MUST be concrete versions (never `catalog:`/`workspace:`) because consumers install via git URL.
-- [ ] `pnpm install --frozen-lockfile`, `pnpm ci:check`, `pnpm test`, `pnpm typecheck`, and `pnpm verify:changelog` all pass locally.
-- [ ] A clean tmpdir consumer running `pnpm add git+file://<repo>` succeeds and prints `17.0.5` for `marked`.
-- [ ] `cd ~/Sites/UNIC/agentic-delivery-model && pnpm install` succeeds (the original failure mode is gone).
-- [ ] No source code under `scripts/` is modified; `scripts/push-to-confluence.mjs` continues to `import { marked } from 'marked'` with no behaviour change.
-
-## Out of scope
-
-- **Migrate to real `pnpm publish` / npm registry.** Recommended long-term path: drop `private: true`, add `"files"` and `"publishConfig"`, add a `release.yml` workflow that runs `pnpm publish --access public` on tag push (which auto-resolves `catalog:`), update README install instructions to `pnpm add unic-confluence`. This is the canonical fix per pnpm's own error message but it is a multi-touch change involving secrets (`NPM_TOKEN`), package metadata, file allowlist auditing, and consumer migration. Track in a separate spec (e.g. `25-publish-to-npm-registry.md`).
-- **`prepare`-script-based catalog rewrite.** Adding a script that runs at consumer-install time to materialise resolved versions in `package.json` was considered and rejected: it requires a YAML parser, mutates `package.json` mid-install, and behaves inconsistently across npm/pnpm/yarn. The benefit (keep `marked` in the catalog for symmetry) is not worth the complexity for one runtime dep.
-- **Convert `marked` to a peerDependency.** Making consumers install `marked` themselves would dodge the catalog issue but breaks the plugin's "drop in and run" promise and changes the consumer contract. Out of scope.
-- **Add `pnpm pack` smoke check to CI.** A workflow step that runs `pnpm pack` and inspects the tarball's `package.json` for `catalog:` strings would catch future regressions, but the ground-rule bullet plus existing reviewer attention should keep regressions cheap. Track separately if a regression actually occurs.
-- **Bump `marked` itself.** Stay on `17.0.5` (whatever the catalog currently holds at execution time). A version bump is a separate decision with separate risk.
-- **Audit other transitive `catalog:` leaks.** This spec verifies the only runtime dep (`marked`). If a future spec adds `peerDependencies` or `optionalDependencies`, the new ground rule covers them — no preemptive audit needed.
-
-_Ralph: append findings here._
diff --git a/apps/claude-code/unic-confluence/docs/plans/README.md b/apps/claude-code/unic-confluence/docs/plans/README.md
deleted file mode 100644
index 750cd68e..00000000
--- a/apps/claude-code/unic-confluence/docs/plans/README.md
+++ /dev/null
@@ -1,90 +0,0 @@
-# Plugin Roadmap
-
-This directory contains one Ralph-ready implementation spec per feature. Each file is a self-contained brief: context, affected files, step-by-step implementation, test cases, and acceptance criteria.
-
-## How to use (for Ralph)
-
-1. Pick the next unstarted spec in numerical order.
-2. Read the spec top to bottom **before** touching any code.
-3. If the "Current behaviour" section doesn't match what you find in the file, STOP and add a `## Deviations` section to the spec file documenting the discrepancy. Do NOT silently adapt.
-4. After landing the change, add `**Status: done — <YYYY-MM-DD>**` at the top of the spec.
-5. Read the spec's `**Version impact:** patch|minor|major` line and execute `PROMPT.md` Step 4.5 before committing.
-
-## Ground rules
-
-- **Package manager**: `pnpm` (workspace mode, after spec `00` lands). Before `00`: use `npm`.
-- **pnpm version pinning**: `package.json#packageManager` (e.g. `pnpm@10.24.0`) is the single source of truth. CI workflows MUST NOT pass a `version` input to `pnpm/action-setup` — the action reads `packageManager` automatically. To bump pnpm: edit `packageManager` in `package.json`, run `pnpm install` to regenerate the lockfile, and commit both. Never declare the pnpm version in two places.
-- **Indentation**: tabs. Line endings: LF. (See `.editorconfig`.)
-- **Commits**: conventional commits — `feat(scope): description`, `fix(scope): description`, `chore(scope): description`. No ticket numbers required.
-- **Dep versioning**: `devDependencies` go through the `catalog:` section in `pnpm-workspace.yaml`, pinned exactly. Never add `^` or `~` ranges. To add/bump a dev-only dep that uses the catalog: edit the catalog entry only.
-- **Runtime dep specifiers**: `dependencies`, `peerDependencies`, and `optionalDependencies` in `package.json` MUST use exact pinned versions (e.g. `"marked": "17.0.5"`) — never `catalog:` or `workspace:`. The plugin is consumed via git URL (`git+ssh://...`), so consumers don't share our pnpm workspace and cannot resolve catalog specifiers; pnpm only rewrites `catalog:` at `pnpm publish` time, which this repo does not run. `devDependencies` may continue to use `catalog:` because consumers don't install transitive devDeps. To bump a runtime dep: edit `package.json` directly; do not add it to the catalog.
-- **Supply-chain guards**: `minimumReleaseAge: 1440` blocks packages younger than 24 hours. If a bump is urgent, add the dep to `minimumReleaseAgeExclude` in `pnpm-workspace.yaml` and document the justification in the PR description.
-- **Node version**: Node 24.x (Active LTS). `useNodeVersion` in `pnpm-workspace.yaml` pins the exact patch. Bump it only when explicitly needed, not opportunistically.
-- **Plugin versioning**: `.claude-plugin/plugin.json` is the single source of truth for the version number. **Never hand-edit `.claude-plugin/marketplace.json`** — use `pnpm bump` (spec `19`) which bumps, syncs, and promotes the CHANGELOG in one step.
-- **Before opening a PR**: run `pnpm lint` (available after spec `10`), `pnpm test` (after spec `09`), and do one end-to-end publish to a throwaway Confluence test page.
-- **Scope guard**: do NOT add MCP servers, skills, agents, image-upload support, create-page support, or multi-space support. See `CLAUDE.md` "do not add" section (added in spec `17`).
-
-## Versioning policy
-
-**SemVer, strictly.**
-
-| Change type | Bump |
-|---|---|
-| Breaking change to slash-command interface, CLI args, or on-disk contracts (`confluence-pages.json` schema, credential file format at `~/.unic-confluence.json`) | MAJOR |
-| New flag, new subcommand, new feature that doesn't break existing usage | MINOR |
-| Bug fix, docs update, internal refactor, dependency bump | PATCH |
-
-Each spec bumps per its `**Version impact:**` declaration. See `PROMPT.md` Step 4.5.
-
-**Release flow:**
-
-```sh
-# Per change (every spec commit):
-pnpm bump <patch|minor|major>   # bumps version, promotes CHANGELOG
-pnpm verify:changelog            # confirm CI check passes
-
-# Periodic: tag and push a release boundary
-pnpm tag                         # creates local git tag vX.Y.Z
-git push --follow-tags           # publishes tag to GitHub
-```
-
-## Execution order
-
-| # | Spec | Priority | Status |
-|---|---|---|---|
-| 00 | [Convert repo tooling to pnpm](./00-convert-repo-tooling-to-pnpm.md) | P0 | — |
-| 01 | [Drop --verify from slash command](./01-drop-verify-from-slash-command.md) | P0 | — |
-| 02 | [Document marker syntax in README](./02-document-marker-syntax-readme.md) | P0 | — |
-| 03 | [Refuse publish without markers](./03-refuse-publish-without-markers.md) | P1 | — |
-| 04 | [Add --dry-run flag](./04-dry-run-flag.md) | P1 | — |
-| 05 | [Confluence code-macro fidelity](./05-confluence-code-macro-fidelity.md) | P1 | — |
-| 06 | [Version-sync script and CHANGELOG](./06-version-sync-script-and-changelog.md) | P1 | — |
-| 07 | [HTTP timeout and --check-auth](./07-http-timeout-and-check-auth.md) | P1 | — |
-| 08 | [Tighten stripFrontmatter](./08-tighten-strip-frontmatter.md) | P1 | — |
-| 09 | [Tests for pure functions](./09-tests-for-pure-functions.md) | P1 | — |
-| 10 | [Biome lint/format and CI](./10-biome-lint-format-and-ci.md) | P1 | — |
-| 11 | [README troubleshooting section](./11-readme-troubleshooting-section.md) | P1 | — |
-| 12 | [Fix handleHttpError argv leak](./12-fix-handlehttperror-argv-leak.md) | P2 | — |
-| 13 | [Normalise \<p\> wrapping around markers](./13-normalise-p-wrapping-markers.md) | P2 | — |
-| 14 | [JSDoc types and @ts-check](./14-jsdoc-types-ts-check.md) | P2 | — |
-| 15 | [CliError class](./15-clierror-class.md) | P2 | — |
-| 16 | [Plugin-level README](./16-claude-plugin-readme.md) | P2 | — |
-| 17 | [CLAUDE.md do-not-add section](./17-claude-md-do-not-add-section.md) | P2 | — |
-| 18 | [Parallelise runVerify](./18-parallelise-runverify.md) | P2 | — |
-| 19 | [pnpm bump tooling](./19-pnpm-bump-tooling.md) | P1 | — |
-| 20 | [pnpm verify:changelog enforcement](./20-pnpm-verify-changelog.md) | P1 | — |
-| 21 | [Versioning docs and Ralph integration](./21-versioning-docs-and-prompt.md) | P1 | — |
-| 22 | [Auto-populate confluence-pages.json aliases](./22-auto-populate-aliases.md) | P1 | — |
-| 23 | [Fix CI pnpm version conflict](./23-fix-ci-pnpm-version-conflict.md) | P0 | — |
-| 24 | [Fix `catalog:` leak in git-install consumers](./24-fix-catalog-leak-in-git-install.md) | P0 | — |
-
-## Cross-cutting dependencies
-
-- **`07` → `01`**: Spec `01` can either (a) drop the credential check entirely or (b) replace it with the new `--check-auth` subcommand from spec `07`. If you do `01` before `07`, choose option (a). If you land `07` first, choose option (b).
-- **`09` → `08` + code**:  Spec `09` extracts pure functions into `scripts/lib/`. Specs that modify those functions (`08`, `13`) should land before or together with `09` to avoid merge conflicts.
-- **`15` → `09`**: The `CliError` class refactor touches ~20 `process.exit(1)` sites. Land after `09` so tests catch regressions.
-- **`14` → `09` + `10`**: JSDoc pass should come after `09` (lib extraction) and `10` (Biome formatting) to avoid churning JSDoc during the format pass.
-- **`18` depends on `01` outcome**: If spec `01` deletes `--verify` entirely, skip spec `18`.
-- **`19` → `15`**: Spec `19` (`pnpm bump`) imports `CliError` from `scripts/lib/errors.mjs`. Land spec `15` first.
-- **`20` → `19`**: Spec `20` (`pnpm verify:changelog`) references `pnpm bump` in its error messages. Land spec `19` first.
-- **`21` → `19` + `20`**: Spec `21` (docs) documents `pnpm bump` and `pnpm verify:changelog`. Land both first.
diff --git a/apps/claude-code/unic-confluence/package.json b/apps/claude-code/unic-confluence/package.json
index 21a4ad3f..032d166d 100644
--- a/apps/claude-code/unic-confluence/package.json
+++ b/apps/claude-code/unic-confluence/package.json
@@ -3,7 +3,6 @@
     "marked": "17.0.5"
   },
   "devDependencies": {
-    "@ralph-orchestrator/ralph-cli": "catalog:",
     "@types/node": "catalog:",
     "@unic/release-tools": "workspace:*",
     "@unic/tsconfig": "workspace:*",
@@ -20,7 +19,6 @@
   "scripts": {
     "bump": "unic-bump",
     "confluence": "node scripts/push-to-confluence.mjs",
-    "ralph": "ralph run -c ralph.yml -H builtin:code-assist",
     "sync-version": "unic-sync-version",
     "tag": "unic-tag",
     "test": "node --test scripts/lib/frontmatter.test.mjs scripts/lib/inject.test.mjs scripts/lib/resolve.test.mjs scripts/lib/slug.test.mjs scripts/lib/pages-file.test.mjs",
diff --git a/apps/claude-code/unic-confluence/ralph.yml b/apps/claude-code/unic-confluence/ralph.yml
deleted file mode 100644
index 8b24f7c0..00000000
--- a/apps/claude-code/unic-confluence/ralph.yml
+++ /dev/null
@@ -1,35 +0,0 @@
-# Ralph Orchestrator Configuration
-# Generated by: ralph init --backend claude
-# Project: unic-confluence
-# Docs: https://github.com/mikeyobrien/ralph-orchestrator
-
-cli:
-  backend: 'claude'
-
-event_loop:
-  prompt_file: 'PROMPT.md'
-  completion_promise: 'LOOP_COMPLETE'
-  max_iterations: 100 # 15 specs * generous retry budget
-  max_runtime_seconds: 28800 # 8 hours; abort if a spec stalls
-
-# ─────────────────────────────────────────────────────────────────────────────
-# Additional Configuration (uncomment to customize)
-# ─────────────────────────────────────────────────────────────────────────────
-
-# core:
-#   scratchpad: ".ralph/agent/scratchpad.md"
-#   specs_dir: ".ralph/specs/"
-
-# Custom hats for multi-agent workflows:
-# hats:
-#   builder:
-#     name: "Builder"
-#     triggers: ["build.task"]
-#     publishes: ["build.done", "build.blocked"]
-#
-#   reviewer:
-#     name: "Reviewer"
-#     triggers: ["review.request"]
-#     publishes: ["review.approved", "review.changes_requested"]
-
-# Create PROMPT.md with your task, then run: ralph run
diff --git a/apps/copilot/README.md b/apps/copilot/README.md
new file mode 100644
index 00000000..99f647e8
--- /dev/null
+++ b/apps/copilot/README.md
@@ -0,0 +1,5 @@
+# GitHub Copilot Plugins
+
+This directory is reserved for future Plugins targeting the GitHub Copilot agent runtime. No Copilot-targeting plugins live here yet — this README is the only tracked file.
+
+The monorepo is structured to host Plugins for multiple agent ecosystems under `apps/<agent>/<plugin>/` — see the root [`AGENTS.md`](../../AGENTS.md) and [`CONTEXT.md`](../../CONTEXT.md) for the cross-cutting Plugin / Workspace Package / Release / Consumer vocabulary.
diff --git a/docs/adr/0020-per-plugin-ralph-loops.md b/docs/adr/0020-per-plugin-ralph-loops.md
index 20a13d83..bb9c9090 100644
--- a/docs/adr/0020-per-plugin-ralph-loops.md
+++ b/docs/adr/0020-per-plugin-ralph-loops.md
@@ -1,22 +1,5 @@
 # 0020. Each plugin has its own Ralph loop with its own ralph.yml and PROMPT.md
 
-**Status:** Accepted (2025-04)
-
-## Context
-
-Ralph Orchestrator implements specs one at a time in a loop. A single monorepo-wide loop would mix plugin-specific specs with workspace specs, making iteration focus unclear.
-
-## Decision
-
-Two levels of Ralph loops exist:
-
-1. **Monorepo root loop** (`pnpm ralph` at repo root): implements `docs/plans/` specs (workspace structure, shared packages, CI).
-2. **Per-plugin loop** (`pnpm --filter <name> ralph`): implements `apps/claude-code/<plugin>/docs/plans/` specs.
-
-Each level has its own `ralph.yml` and `PROMPT.md`.
-
-## Consequences
-
-- Adding a new plugin requires creating its own `ralph.yml` and `PROMPT.md`.
-- Running the root loop does not touch plugin-specific specs and vice versa.
-- Status markers (`**Status: done**`) are scoped per loop level.
+> **Moved** to [apps/claude-code/unic-archon-dlc/docs/adr/0002-per-plugin-ralph-loops.md](../../apps/claude-code/unic-archon-dlc/docs/adr/0002-per-plugin-ralph-loops.md) (2026-05).
+>
+> This ADR is harness-internal and now lives with the unic-archon-dlc plugin. The redirect preserves external links; new references should use the new path.
diff --git a/docs/adr/0023-spec-template-format.md b/docs/adr/0023-spec-template-format.md
index dc727630..4dbcf871 100644
--- a/docs/adr/0023-spec-template-format.md
+++ b/docs/adr/0023-spec-template-format.md
@@ -1,25 +1,5 @@
 # 0023. Spec template format for Ralph-executable specs
 
-**Status:** Accepted (2025-04)
-
-## Context
-
-Ralph Orchestrator needs specs to be unambiguous enough to execute deterministically. An ad hoc format leads to missing context, Ralph pausing for clarification, or incorrect implementations.
-
-## Decision
-
-All specs follow `docs/process/spec-template.md`, which mandates:
-
-- A header block: `Priority`, `Effort`, `Version impact`, `Depends on`, `Touches`
-- `## Current behaviour` with exact before-state snapshots (code or CLI output)
-- `## Target behaviour` with exact after-state
-- `## Implementation steps` with explicit before→after diffs or full file content
-- `## Verification` with shell commands and their expected output
-- `## Acceptance criteria` as checkboxes
-- `## Out of scope` bounding the change
-
-## Consequences
-
-- Specs are longer to write but faster for Ralph to execute without back-and-forth.
-- The "Out of scope" section prevents scope creep during automated implementation.
-- Specs that deviate from the template may cause Ralph to pause or implement incorrectly.
+> **Moved** to [apps/claude-code/unic-archon-dlc/docs/adr/0003-spec-template-format.md](../../apps/claude-code/unic-archon-dlc/docs/adr/0003-spec-template-format.md) (2026-05).
+>
+> This ADR is harness-internal and now lives with the unic-archon-dlc plugin. The redirect preserves external links; new references should use the new path.
diff --git a/docs/adr/0024-ralph-atomic-iteration.md b/docs/adr/0024-ralph-atomic-iteration.md
index bc74b6d1..ff8a8994 100644
--- a/docs/adr/0024-ralph-atomic-iteration.md
+++ b/docs/adr/0024-ralph-atomic-iteration.md
@@ -1,17 +1,5 @@
 # 0024. Ralph implements one spec per iteration, then commits and stops
 
-**Status:** Accepted (2025-04)
-
-## Context
-
-Batching multiple specs into one Ralph iteration makes commits hard to review, complicates revert, and makes it harder to resume after a failure.
-
-## Decision
-
-Ralph's loop implements exactly one spec per run, creates one commit, and stops. Loop resumption reads `**Status: done**` markers in spec files to determine the next spec to implement. The loop never modifies `PROMPT.md` or `ralph.yml` to batch specs.
-
-## Consequences
-
-- Each PR from a Ralph run contains one logical unit of work.
-- The commit message references the spec number and title.
-- Resuming after a failed Ralph run is mechanical: remove the `**Status: done**` marker from the partially-implemented spec and restart.
+> **Moved** to [apps/claude-code/unic-archon-dlc/docs/adr/0004-ralph-atomic-iteration.md](../../apps/claude-code/unic-archon-dlc/docs/adr/0004-ralph-atomic-iteration.md) (2026-05).
+>
+> This ADR is harness-internal and now lives with the unic-archon-dlc plugin. The redirect preserves external links; new references should use the new path.
diff --git a/docs/adr/0026-tdd-dispatch-by-version-impact.md b/docs/adr/0026-tdd-dispatch-by-version-impact.md
index 75be6f6a..0f830a59 100644
--- a/docs/adr/0026-tdd-dispatch-by-version-impact.md
+++ b/docs/adr/0026-tdd-dispatch-by-version-impact.md
@@ -1,30 +1,5 @@
 # 0026. Ralph uses `/tdd` for behavioral specs, direct implementation for structural ones, dispatched by `**Version impact:**`
 
-**Status:** Accepted (2026-05)
-
-## Context
-
-The Ralph loop previously told the agent to "follow the Implementation steps exactly" for every spec. This produced horizontal slicing (building layer by layer as prescribed) rather than vertical slicing (a thin end-to-end slice that proves value immediately). It also left no explicit feedback loop before committing — the agent could implement and commit without ever running tests.
-
-Adding `/tdd` unconditionally would be awkward for structural specs (config files, release tooling, docs) where there is no meaningful unit of behavior to test first. A test asserting that `plugin.json` contains a field adds noise with no signal.
-
-## Decision
-
-`PROMPT.md` dispatches on the spec's `**Version impact:**` field:
-
-- **`none`** — implement directly; follow the Implementation steps as a recipe. No `/tdd`.
-- **`patch` / `minor` / `major`** — invoke `/tdd`; treat the Implementation steps as guidance (key files, rough order), not a recipe. The spec's Acceptance criteria are the target; the red-green-refactor cycle drives the path.
-
-`**Version impact:**` was chosen as the dispatch signal because it is already present in every spec, it reliably separates behavioral changes (which touch runtime code) from structural ones (which touch config, docs, or tooling), and it requires no new metadata.
-
-## Considered options
-
-- **Effort-gated dispatch** — switch to `/tdd` after N failed attempts. Rejected: this rewards thrashing rather than preventing it, and gives the agent no clear signal upfront.
-- **Universal `/tdd`** — apply to all specs including `none`. Rejected: a test asserting a config field exists adds noise and would fight the prescriptive implementation steps that structural specs genuinely need.
-- **New metadata field in spec frontmatter** — e.g. `**Implementation style: tdd|direct**`. Rejected: `**Version impact:**` already encodes the same distinction; a new field would duplicate information and drift.
-
-## Consequences
-
-- Specs with `**Version impact: none**` remain recipe-style; their Implementation steps should stay prescriptive.
-- Specs with a semver impact should be written acceptance-criteria-first going forward; detailed step-by-step instructions are demoted to guidance.
-- The always-on `pnpm check` block in Step 4 serves as the post-`/tdd` hygiene gate, catching formatting regressions the TDD loop does not cover.
+> **Moved** to [apps/claude-code/unic-archon-dlc/docs/adr/0005-tdd-dispatch-by-version-impact.md](../../apps/claude-code/unic-archon-dlc/docs/adr/0005-tdd-dispatch-by-version-impact.md) (2026-05).
+>
+> This ADR is harness-internal and now lives with the unic-archon-dlc plugin. The redirect preserves external links; new references should use the new path.
diff --git a/docs/adr/0027-feature-runner-context-bundle.md b/docs/adr/0027-feature-runner-context-bundle.md
index 79366449..56581e01 100644
--- a/docs/adr/0027-feature-runner-context-bundle.md
+++ b/docs/adr/0027-feature-runner-context-bundle.md
@@ -1,43 +1,5 @@
 # 0027. Feature Runner injects a scoped context bundle into every `/tdd` sub-agent invocation
 
-**Status:** Accepted (2026-05)
-
-## Context
-
-The Feature Runner invokes `/tdd` as a non-interactive sub-agent (via the Agent tool) for each issue in a feature. The first draft of the Feature Runner PRD specified "the issue file content as context" as the sole input to each `/tdd` invocation.
-
-This is insufficient. `/tdd` is designed to be interactive: its planning phase asks the user to confirm interface changes and approve which behaviours to test before writing any code. In AFK mode there is no user to ask. Without additional context, `/tdd` reasons from a single vertical slice with no access to the "why" behind the feature, the architectural constraints that apply, or the vocabulary the codebase uses. This creates a risk of a correct-but-wrong implementation — code that satisfies the issue's literal description but diverges from the intent established during the grilling and PRD sessions.
-
-Matt Pocock's reference AFK loop (`afk.sh`) injects all issue files plus recent commits as a single prompt string before invoking `/tdd`, establishing the precedent that the agent needs broader context than a single work item.
-
-## Decision
-
-The Feature Runner assembles a **context bundle** for each `/tdd` sub-agent invocation:
-
-1. **Issue file** — `## What to build` and `## Acceptance criteria` serve as the pre-answered planning conversation (see ADR-0029).
-2. **PRD** — read from the feature's `docs/issues/<slug>/PRD.md` (the slug is the feature directory name; the PRD lives at a fixed path within it). Carries the shared vision from the grilling session and the "why" behind the feature. Without it, the agent lacks the context needed to judge correctness beyond the literal issue description.
-3. **Sibling issue files** — all other `NN-*.md` files in the feature directory. Provides dependency awareness and a "what is already resolved" signal without requiring the runner to summarise prior work.
-4. **Scoped CONTEXT.md** — the domain glossary for the feature's domain (see scoping rule below). Ensures test names and interface vocabulary match the project's language.
-5. **Scoped ADRs** — the architectural decisions constraining the implementation (see scoping rule below).
-6. **Recent commits** — the last 5 git commits. The grilling and PRD process typically produces changes to CONTEXT.md and ADRs that land in commits before the Feature Runner runs. These commits carry the ideation trail that informed the PRD.
-
-### ADR scoping rule
-
-ADRs and CONTEXT.md are scoped to the domain of the feature, not the monorepo:
-
-- **Plugin feature** (PRD references paths under `apps/claude-code/<plugin>/`) → inject that plugin's `docs/adr/` and `CONTEXT.md`.
-- **Repo/tooling feature** (PRD references paths outside `apps/`, e.g. `.claude/`, `docs/`, `packages/`) → inject the root `docs/adr/` and root `CONTEXT.md`.
-
-Scope is inferred by scanning the PRD for `apps/claude-code/<plugin>` path references. Root ADRs cover versioning, tagging, and CI tooling — they are noise for plugin implementation work and must not be injected into plugin feature runs.
-
-## Considered options
-
-- **Lazy discovery** — let `/tdd` explore the codebase and find ADRs and CONTEXT.md on its own. Rejected: `/tdd` does instruct the agent to use the domain glossary and respect ADRs, but in non-interactive sub-agent mode this exploration is unreliable. Injection is guaranteed; discovery is not.
-- **Issue file only** — the minimal approach from the first PRD draft. Rejected: `/tdd` loses the PRD's "why", the sibling issues' dependency signal, and the architectural constraints, all of which are needed to produce implementations that match the grilled intent.
-- **Inject everything** — all ADRs from all directories. Rejected: root ADRs are irrelevant to plugin work and add context noise without signal.
-
-## Consequences
-
-- The Feature Runner skill reads the PRD from the fixed slug-derived path `docs/issues/<slug>/PRD.md`. The `## Parent` link on issue files is informational for human readers; the runner does not parse it. Features without a `PRD.md` at that path cannot be run by the Feature Runner without manual intervention.
-- The Feature Runner skill must scan the PRD for `apps/claude-code/<plugin>` references to determine which CONTEXT.md and ADR directory to inject.
-- The context bundle grows with the number of sibling issues; for features with many issues, later invocations carry more sibling context than earlier ones. This is acceptable — it mirrors the growing "what is done" signal available in real commits.
+> **Moved** to [apps/claude-code/unic-archon-dlc/docs/adr/0006-feature-runner-context-bundle.md](../../apps/claude-code/unic-archon-dlc/docs/adr/0006-feature-runner-context-bundle.md) (2026-05).
+>
+> This ADR is harness-internal and now lives with the unic-archon-dlc plugin. The redirect preserves external links; new references should use the new path.
diff --git a/docs/adr/0028-blocked-by-canonical-sequencing.md b/docs/adr/0028-blocked-by-canonical-sequencing.md
index 1f82faaa..d33c7cc7 100644
--- a/docs/adr/0028-blocked-by-canonical-sequencing.md
+++ b/docs/adr/0028-blocked-by-canonical-sequencing.md
@@ -1,32 +1,5 @@
 # 0028. `## Blocked by` is the canonical sequencing signal for Feature Runner issue execution
 
-**Status:** Accepted (2026-05)
-
-## Context
-
-Issues in `docs/issues/<slug>/` are named with a numeric prefix (`NN-*.md`) produced by the `to-issues` skill, which publishes issues in dependency order so blockers get lower numbers. This makes numerical filename order a reliable proxy for execution order in practice.
-
-However, `to-issues` also records explicit dependency information in each issue's `## Blocked by` field. The numeric prefix is a UX convenience — it makes the dependency graph human-readable at a glance in a file browser. It is not an execution contract. The `to-issues` skill's "Blocked by" field is the canonical representation of the dependency graph: it can express non-linear dependencies that numerical order cannot (e.g. issue 03 blocking issue 02 after a user reorders slices during review).
-
-Treating numerical order as the execution contract would make the Feature Runner silently incorrect whenever `## Blocked by` and filename order diverge — a failure mode that would be invisible until a downstream issue ran on a broken foundation.
-
-## Decision
-
-The Feature Runner builds a **topological order** from `## Blocked by` references before executing any issue. Numerical filename order is used only as a tiebreaker when two issues have no dependency relationship between them.
-
-If `## Blocked by` references conflict with numerical filename order (i.e. a lower-numbered issue declares a blocker that is a higher-numbered issue), the Feature Runner halts with an error and surfaces the conflict to the user rather than proceeding in the wrong order. Silent execution on a potentially wrong order is not acceptable.
-
-Issues with `## Blocked by: None` (or equivalent) have no predecessors and may be placed anywhere in the topological order consistent with their number.
-
-## Considered options
-
-- **Numerical order only** — simpler to implement; no graph parsing required. Rejected: not an execution contract; silently wrong when user reorders slices or when `to-issues` produces a non-linear dependency graph.
-- **`## Blocked by` order, silent fallback to numerical on conflict** — avoids halting. Rejected: a conflict between the two signals indicates a malformed feature (either the issue was hand-edited or `to-issues` produced unexpected output); proceeding silently would compound the error.
-- **`## Blocked by` order, halt on conflict** — chosen. Forces the human to resolve ambiguity before the autonomous run begins, preventing downstream issues from inheriting a broken foundation.
-
-## Consequences
-
-- The Feature Runner skill must parse `## Blocked by` fields and construct a dependency graph before beginning execution.
-- Features where `## Blocked by` and numerical order disagree will not run until the conflict is resolved by the developer.
-- The `to-issues` skill's practice of publishing issues in dependency order (blockers first) remains a useful convention that keeps numerical order and the dependency graph aligned in the common case.
-- The dependency graph also reveals which issues are parallelisable (those with `## Blocked by: None` and no dependents). The Feature Runner serialises all execution regardless — see the Feature Runner PRD Out of Scope for the rationale.
+> **Moved** to [apps/claude-code/unic-archon-dlc/docs/adr/0007-blocked-by-canonical-sequencing.md](../../apps/claude-code/unic-archon-dlc/docs/adr/0007-blocked-by-canonical-sequencing.md) (2026-05).
+>
+> This ADR is harness-internal and now lives with the unic-archon-dlc plugin. The redirect preserves external links; new references should use the new path.
diff --git a/docs/adr/0029-feature-runner-afk-invocation.md b/docs/adr/0029-feature-runner-afk-invocation.md
index 59e12d64..5f9b89d0 100644
--- a/docs/adr/0029-feature-runner-afk-invocation.md
+++ b/docs/adr/0029-feature-runner-afk-invocation.md
@@ -1,36 +1,5 @@
 # 0029. Feature Runner invokes `/tdd` non-interactively; issue acceptance criteria replace the planning phase
 
-**Status:** Accepted (2026-05)
-
-## Context
-
-`/tdd`'s planning phase is interactive: before writing any code it asks the user to confirm interface changes, prioritise which behaviours to test, and approve the plan. This is by design — it prevents the agent from outrunning its headlights on ambiguous requirements.
-
-The Feature Runner's core use case is autonomous, overnight execution (composable with `/loop`). There is no user present to answer planning questions. Invoking `/tdd` as a sub-agent via the Agent tool means there is no TTY — interactive prompts cannot be issued and the planning phase cannot execute as designed.
-
-Matt Pocock's reference AFK loop (`afk.sh`) resolves this by running Claude with `--print` (non-interactive mode) and injecting issue files as the implicit plan. The Agent tool in Claude Code is the equivalent mechanism: sub-agents run without a TTY and must infer their plan from the provided context.
-
-## Decision
-
-The Feature Runner invokes `/tdd` via the Agent tool (non-interactive). The issue's `## Acceptance criteria` section serves as the pre-answered planning conversation:
-
-- `## What to build` answers "what interface changes are needed"
-- `## Acceptance criteria` answers "which behaviours to test" and "what does done look like"
-
-Because issues are produced by `to-issues` (which slices the PRD vertically and quizzes the user on the breakdown) and then reviewed by the user before reaching `ready-for-agent`, the acceptance criteria represent a human-approved definition of done. The interactive planning phase is not bypassed in substance — it was completed during the grilling and issue-writing pipeline; the Feature Runner simply does not repeat it at runtime.
-
-`/tdd` is not modified. No AFK flag or non-interactive variant is introduced.
-
-## Considered options
-
-- **Fork `/tdd` into a non-interactive variant** (`/tdd-afk` or similar) — would allow explicit suppression of planning prompts. Rejected: creates a maintenance burden; the two variants would diverge over time; the Agent tool already provides non-interactive execution without any skill changes.
-- **Add an AFK flag to `/tdd`** — e.g. a frontmatter option or a prompt prefix that skips planning. Rejected: couples the `/tdd` skill to the Feature Runner's invocation model; `to-issues` already produces the information that the planning phase would gather.
-- **Non-interactive Agent tool invocation with injected context bundle** — chosen. No skill modifications required; the planning information is supplied via the context bundle (ADR-0027) rather than elicited at runtime.
-
-## Consequences
-
-- Issues that reach the Feature Runner must have specific, human-reviewed `## Acceptance criteria`. Vague criteria (e.g. "the feature works correctly") remove the planning substitute and leave `/tdd` without a concrete definition of done.
-- The `to-issues` + `/triage` → `ready-for-agent` pipeline is load-bearing: it is the point at which the planning conversation occurs. The Feature Runner depends on that pipeline having been followed correctly.
-- `/tdd` remains unchanged and continues to work interactively when invoked directly by the user.
-- When the queue is empty (no `ready-for-agent` features remain), the Feature Runner outputs `LOOP_COMPLETE` as its stop signal. This is what makes `/loop /implement-feature` composable for overnight draining: the `/loop` skill catches `LOOP_COMPLETE` and terminates the loop cleanly rather than spinning on an empty queue. This mirrors the Spec Runner's `completion_promise: LOOP_COMPLETE` in `ralph.yml` and Matt Pocock's `<promise>NO MORE TASKS</promise>` in `afk.sh`.
-- The Feature Runner's prompt template explicitly instructs the sub-agent to invoke the `tdd` skill via the Skill tool, rather than relying on auto-discovery via trigger phrases. This makes the procedural-guidance load deterministic across Claude Code surfaces.
+> **Moved** to [apps/claude-code/unic-archon-dlc/docs/adr/0008-feature-runner-afk-invocation.md](../../apps/claude-code/unic-archon-dlc/docs/adr/0008-feature-runner-afk-invocation.md) (2026-05).
+>
+> This ADR is harness-internal and now lives with the unic-archon-dlc plugin. The redirect preserves external links; new references should use the new path.
diff --git a/docs/adr/0030-retire-ralph-adopt-archon-runner.md b/docs/adr/0030-retire-ralph-adopt-archon-runner.md
new file mode 100644
index 00000000..fcd2d6fb
--- /dev/null
+++ b/docs/adr/0030-retire-ralph-adopt-archon-runner.md
@@ -0,0 +1,5 @@
+# 0030. Retire ralph-orchestrator; adopt unic-archon-dlc as the Feature Runner
+
+> **Moved** to [apps/claude-code/unic-archon-dlc/docs/adr/0009-retire-ralph-adopt-archon-runner.md](../../apps/claude-code/unic-archon-dlc/docs/adr/0009-retire-ralph-adopt-archon-runner.md) (2026-05).
+>
+> This ADR is harness-internal and now lives with the unic-archon-dlc plugin. The redirect preserves external links; new references should use the new path.
diff --git a/docs/adr/0031-retire-implement-feature-skill.md b/docs/adr/0031-retire-implement-feature-skill.md
new file mode 100644
index 00000000..9b6e2261
--- /dev/null
+++ b/docs/adr/0031-retire-implement-feature-skill.md
@@ -0,0 +1,5 @@
+# 0031. Retire the `/implement-feature` skill; Feature Runner backed solely by `unic-dlc-build`
+
+> **Moved** to [apps/claude-code/unic-archon-dlc/docs/adr/0010-retire-implement-feature-skill.md](../../apps/claude-code/unic-archon-dlc/docs/adr/0010-retire-implement-feature-skill.md) (2026-05).
+>
+> This ADR is harness-internal and now lives with the unic-archon-dlc plugin. The redirect preserves external links; new references should use the new path.
diff --git a/docs/agents/branching.md b/docs/agents/branching.md
new file mode 100644
index 00000000..375005a5
--- /dev/null
+++ b/docs/agents/branching.md
@@ -0,0 +1,30 @@
+<!--
+  AUTO-GENERATED by unic-archon-dlc — DO NOT EDIT BY HAND.
+  Source: apps/claude-code/unic-archon-dlc/lib/agent-docs-writer.mjs
+  Regenerate: re-run `/unic-archon-dlc:setup` (or call `runInstall()` from `lib/install-runner.mjs`).
+  This file describes the unic-archon-dlc target workflow; the current driver
+  in any given Consumer repo may differ.
+-->
+
+# Branching Strategy
+
+Configured by unic-archon-dlc.
+
+## Strategy: Gitflow
+
+| Branch type | Pattern          | PR target          |
+| ----------- | ---------------- | ------------------ |
+| Production  | `main`           | —                  |
+| Integration | `develop`        | —                  |
+| Feature     | `feature/<name>` | `develop`          |
+| Hotfix      | `hotfix/<name>`  | `main` + `develop` |
+
+## Default branch names
+
+- **Main branch:** `main`
+- **Integration branch:** `develop`
+- **Feature branch prefix:** `feature/`
+
+## PR conventions
+
+All PRs target `develop`. Merge strategy: `merge`.
diff --git a/docs/agents/domain.md b/docs/agents/domain.md
index 967c09ec..5fc42a08 100644
--- a/docs/agents/domain.md
+++ b/docs/agents/domain.md
@@ -1,42 +1,22 @@
-# Domain Docs
+<!--
+  AUTO-GENERATED by unic-archon-dlc — DO NOT EDIT BY HAND.
+  Source: apps/claude-code/unic-archon-dlc/lib/agent-docs-writer.mjs
+  Regenerate: re-run `/unic-archon-dlc:setup` (or call `runInstall()` from `lib/install-runner.mjs`).
+  This file describes the unic-archon-dlc target workflow; the current driver
+  in any given Consumer repo may differ.
+-->
 
-How the engineering skills should consume this repo's domain documentation.
+# Domain
 
-## Before exploring, read these
+Configured by unic-archon-dlc.
 
-- **`CONTEXT-MAP.md`** at the repo root — it points at one `CONTEXT.md` per plugin context. Read each one relevant to the topic.
-- **`docs/adr/`** — read ADRs that touch the area you're about to work in.
-- Per-plugin ADRs live at `apps/claude-code/<plugin>/docs/adr/`.
+## Repository layout: multi-context
 
-If any of these files don't exist, **proceed silently**. Don't flag their absence; don't suggest creating them upfront. The producer skill (`/grill-with-docs`) creates them lazily when terms or decisions actually get resolved.
+This repository uses **multi-context** layout. Each package/app has its own `CONTEXT.md` file. A `CONTEXT-MAP.md` at the repo root maps each context to its location.
 
-## File structure
+- **Context map:** `CONTEXT-MAP.md`
+- **ADRs:** monorepo-wide decisions live in root `docs/adr/`; each context may also keep its own `docs/adr/` for decisions scoped to that context.
 
-```
-/
-├── CONTEXT-MAP.md                          ← points to per-plugin contexts
-├── docs/adr/                               ← monorepo-wide decisions
-└── apps/
-    └── claude-code/
-        ├── pr-review/
-        │   ├── CONTEXT.md
-        │   └── docs/adr/
-        ├── auto-format/
-        │   ├── CONTEXT.md
-        │   └── docs/adr/
-        └── unic-confluence/
-            ├── CONTEXT.md
-            └── docs/adr/
-```
+## How agents use this
 
-## Use the glossary's vocabulary
-
-When your output names a domain concept, use the term as defined in the relevant `CONTEXT.md`. Don't drift to synonyms the glossary explicitly avoids.
-
-If the concept isn't in the glossary yet, note it for `/grill-with-docs` rather than inventing new language.
-
-## Flag ADR conflicts
-
-If your output contradicts an existing ADR, surface it explicitly:
-
-> _Contradicts ADR-0007 (…) — but worth reopening because…_
+Every agent working in this repo should read the relevant `CONTEXT.md` (located via `CONTEXT-MAP.md`) and the ADRs in root `docs/adr/` plus any context-scoped `docs/adr/` before proposing terminology changes or architectural decisions.
diff --git a/docs/agents/feature-runner.md b/docs/agents/feature-runner.md
index e013e5a7..3c4307ba 100644
--- a/docs/agents/feature-runner.md
+++ b/docs/agents/feature-runner.md
@@ -1,124 +1,23 @@
 # Feature Runner
 
-The Feature Runner is the `/implement-feature` skill. It automates the implementation side of the AI-development cycle: given a Feature slug, it creates an isolated branch, works through all `ready-for-agent` issues in dependency order using `/tdd`, opens a pull request, and marks each issue `resolved`. It is the issue-tracker-driven counterpart to the Spec Runner.
+The **Feature Runner** is the concept of a runner that implements a Feature's issues end-to-end in one worktree, branch, and pull request — see root `CONTEXT.md`.
 
-Invoke it with `/implement-feature <slug>` (named Feature) or `/implement-feature` (auto-select). Compose it with `/loop` for overnight queue draining.
+## Current state
 
-## Lifecycle
+- **Current default — manual `/tdd` per issue.** Until the AFK runner is wired into this repo, the developer drives `/tdd` against each `ready-for-agent` issue, respects `## Blocked by` order ([ADR-0007](../../apps/claude-code/unic-archon-dlc/docs/adr/0007-blocked-by-canonical-sequencing.md)), marks the issue `resolved`, and opens a PR targeting `develop` once the feature's issues are done.
+- **Long-term AFK runner — `unic-dlc-build`.** Shipped by `unic-archon-dlc` (also developed in this monorepo). See [ADR-0009](../../apps/claude-code/unic-archon-dlc/docs/adr/0009-retire-ralph-adopt-archon-runner.md) for the retirement of `ralph-orchestrator` and [ADR-0010](../../apps/claude-code/unic-archon-dlc/docs/adr/0010-retire-implement-feature-skill.md) for the retirement of the interim `/implement-feature` skill that briefly filled this role.
 
-```
-feature selected → worktree created → issues implemented in topological order → PR opened → issues closed on merge
-```
+## What survives across runners
 
-### 1. Feature selection
+Regardless of which runner executes a Feature, these conventions hold:
 
-- **Named**: `/implement-feature <slug>` targets `docs/issues/<slug>/` directly.
-- **Auto-select**: `/implement-feature` with no argument scans `docs/issues/` and picks the first Feature (alphabetically by slug) where at least one `NN-*.md` file is `ready-for-agent` and no file is in an unprepped state (`needs-triage`, `needs-info`, `needs-specs`). Partially-completed Features (mix of `resolved` and `ready-for-agent`) are included — the runner resumes from where it stopped. Features where every issue is `resolved`, `closed`, or `rejected` are skipped (nothing left to run). Issues with status `ready-for-human` or `rejected` do not disqualify a feature.
-- **Empty queue**: when no qualifying Feature exists, the runner emits `LOOP_COMPLETE` on its own line and exits cleanly. This is the stop signal that `/loop` uses to terminate an overnight run.
+- Issues live at `docs/issues/<slug>/NN-*.md` with the `## What to build` / `## Acceptance criteria` format.
+- `## Blocked by` is the canonical execution-order signal ([ADR-0007](../../apps/claude-code/unic-archon-dlc/docs/adr/0007-blocked-by-canonical-sequencing.md)). Numeric filename prefixes are a UX convenience, not a contract.
+- Each Feature ships as a single PR targeting `develop`. Issues are marked `resolved` on implementation and `closed` after the PR merges.
 
-### 2. Worktree creation
+## Related
 
-The runner creates (or reuses) a git worktree and branch:
-
-- Branch: `feature/afk/<slug>`
-- Worktree path: `.claude/worktrees/<slug>`
-
-If `.claude/worktrees/<slug>` already exists (prior failed run), the runner reuses it — `git worktree add` is skipped and the existing branch retains its committed work. If it does not exist, the runner creates it from `develop`. On success, the worktree is removed after the PR is opened. On failure, it is left in place for inspection.
-
-### 3. Issue implementation
-
-Issues are executed via `/tdd` sub-agent invocations in **topological order** derived from `## Blocked by` references (see [Dependency ordering](#dependency-ordering) below). Only `ready-for-agent` issues are executed. `resolved`, `closed`, and `rejected` issues satisfy dependencies but are skipped. `ready-for-human` issues are unsatisfied — if a `ready-for-agent` issue depends on one, the runner halts before executing anything.
-
-Before each invocation the runner outputs:
-
-```
-Implementing issue N of M: <issue title>
-```
-
-After a successful `/tdd` invocation, the issue file is updated: `Status: ready-for-agent` → `Status: resolved`.
-
-On failure, a note is appended to the failing issue under `## Comments` (see [Failure behaviour](#failure-behaviour)) and the runner stops. No subsequent issues run.
-
-### 4. PR and cleanup
-
-When all issues are resolved, the runner:
-
-1. Pushes `feature/afk/<slug>` to origin.
-2. Opens a pull request targeting `develop` via `gh pr create`. The PR title is `feat(<slug>): <PRD title>` and the body references the PRD and lists all resolved issues.
-3. Removes the worktree.
-
-Issues remain at `Status: resolved` until the PR is merged, at which point they are manually marked `closed` (or via a future hook).
-
-## Context bundle
-
-Each `/tdd` sub-agent invocation receives a six-part context bundle assembled by the runner:
-
-| Part                  | What it contains                                                    | Why                                                                                      |
-| --------------------- | ------------------------------------------------------------------- | ---------------------------------------------------------------------------------------- |
-| **Issue file**        | `## What to build` and `## Acceptance criteria`                     | Replaces the interactive planning phase in AFK mode                                      |
-| **PRD**               | `docs/issues/<slug>/PRD.md`                                         | Carries the "why" and shared vision from the grilling session                            |
-| **Sibling issues**    | All other `NN-*.md` files in the feature directory at current state | Shows what is already resolved and what is still pending                                 |
-| **Scoped CONTEXT.md** | Plugin or root `CONTEXT.md` (see ADR scope below)                   | Ensures interface vocabulary and test names match the domain glossary                    |
-| **Scoped ADRs**       | All `*.md` files in the scoped ADR directory                        | Communicates the architectural constraints that bind the implementation                  |
-| **Recent commits**    | `git log --oneline -5`                                              | Carries the ideation trail from grilling and PRD work that landed just before the runner |
-
-### ADR scope
-
-ADRs are scoped to the domain of the Feature:
-
-- **Plugin Feature**: the PRD references one or more paths under `apps/claude-code/<plugin>/` → inject `apps/claude-code/<plugin>/CONTEXT.md` and `apps/claude-code/<plugin>/docs/adr/`. Root ADRs are **not** injected.
-- **Repo/tooling Feature**: no `apps/claude-code/<plugin>/` references in the PRD → inject root `CONTEXT.md` and root `docs/adr/`.
-
-## Dependency ordering
-
-`## Blocked by` is the canonical dependency signal, not numeric filename order. Numeric order is a UX convenience produced by `/to-issues` (it publishes blockers first so numbers usually match), but it is not an execution contract.
-
-The runner builds a topological execution order from `## Blocked by` references across all `NN-*.md` files before executing anything.
-
-**Conflict detection**: if an issue A lists issue B in `## Blocked by`, and B has a higher numeric prefix than A, the dependency contradicts the numeric convention. The runner halts before executing any issue and reports:
-
-```
-Feature Runner error: dependency conflict detected.
-  Issue NN-<A> is blocked by NN-<B>, but NN-<B> has a higher number than NN-<A>.
-  This conflicts with the numerical ordering convention. Resolve the ordering manually before re-running.
-```
-
-`## Blocked by: None`, `## Blocked by: None — can start immediately`, or a missing `## Blocked by` section all mean the issue has no predecessors.
-
-## Failure behaviour
-
-When a `/tdd` sub-agent cannot complete an issue:
-
-1. The issue status is changed to `needs-info`. This prevents the auto-selection path from picking up this Feature on subsequent `/loop` iterations until the developer investigates and restores `ready-for-agent` (or closes/rejects the issue).
-2. The runner appends a failure note to the issue file under `## Comments`:
-
-```markdown
-## Comments
-
-> _This was generated by AI during triage._
-
-**Feature Runner failure** — `/tdd` could not complete this issue. Status has been set to `needs-info`.
-
-The worktree at `.claude/worktrees/<slug>` has been left in place for inspection. Once the issue is resolved manually, restore `**Status:** ready-for-agent` and re-run `/implement-feature <slug>` to resume. Alternatively, close or reject the issue if it should not be retried.
-```
-
-3. The runner stops. No subsequent issues in the Feature are executed.
-4. The worktree is left at `.claude/worktrees/<slug>` on `feature/afk/<slug>` for inspection.
-
-Re-running `/implement-feature <slug>` after a manual fix resumes from the first `ready-for-agent` issue (already-resolved issues are skipped via the topological filter).
-
-## Historical cleanup convention
-
-When a Spec in `docs/plans/` is marked `done` and a corresponding `docs/issues/<slug>/` directory exists, the issues in that directory were implemented via the Spec Runner, not the Feature Runner. They will never be processed by `/implement-feature` and should not remain at `ready-for-agent`.
-
-**Convention**: manually mark all `NN-*.md` files in `docs/issues/<slug>/` as `closed` and append a note:
-
-```markdown
-## Comments
-
-> _This was generated by AI during triage._
-
-Marked `closed` — implemented via the Spec Runner (see `docs/plans/<spec-file>.md`, marked `done`). The Feature Runner was not used for this Feature.
-```
-
-This prevents the auto-selection path from picking up stale Features and keeps the issue tracker accurate.
+- [`docs/process/development-workflow.md`](../process/development-workflow.md) — phase-by-phase workflow
+- [`docs/process/ai-development.md`](../process/ai-development.md) — deep guide: mental model, context quality, AFK trust chain
+- [`docs/agents/issue-tracker.md`](issue-tracker.md) — issue file conventions
+- [`docs/agents/triage-labels.md`](triage-labels.md) — 8-state triage vocabulary
diff --git a/docs/agents/issue-tracker.md b/docs/agents/issue-tracker.md
index 1fe8cd58..404d77f5 100644
--- a/docs/agents/issue-tracker.md
+++ b/docs/agents/issue-tracker.md
@@ -1,21 +1,34 @@
-# Issue tracker: Local Markdown
+<!--
+  AUTO-GENERATED by unic-archon-dlc — DO NOT EDIT BY HAND.
+  Source: apps/claude-code/unic-archon-dlc/lib/agent-docs-writer.mjs
+  Regenerate: re-run `/unic-archon-dlc:setup` (or call `runInstall()` from `lib/install-runner.mjs`).
+  This file describes the unic-archon-dlc target workflow; the current driver
+  in any given Consumer repo may differ.
+-->
 
-Pre-scope ideas live in `docs/inbox/` (see `docs/inbox/README.md`). Once an idea has been grilled and a feature slug agreed, it moves here.
+# Issue Tracker: github
 
-Issues and PRDs for this repo live as markdown files in `docs/issues/`.
+Configured by unic-archon-dlc.
 
-## Conventions
+## Backend
+
+**Tracker:** `github`
+**PR strategy:** `merge`
 
-- One feature per directory: `docs/issues/<feature-slug>/`
-- The PRD is `docs/issues/<feature-slug>/PRD.md`
-- Implementation issues are `docs/issues/<feature-slug>/<NN>-<slug>.md`, numbered from `01`
-- Triage state is recorded as a `Status:` line near the top of each issue file (see `triage-labels.md` for the label strings)
-- Comments and conversation history append to the bottom of the file under a `## Comments` heading
+## Create a new issue
 
-## When a skill says "publish to the issue tracker"
+```sh
+gh issue create --title "<title>" --label "<label>"
+```
 
-Create a new file under `docs/issues/<feature-slug>/` (creating the directory if needed).
+## Update issue state
 
-## When a skill says "fetch the relevant ticket"
+```sh
+gh issue edit <number> --add-label "<label>"
+```
+
+## Conventions
 
-Read the file at the referenced path. The user will normally pass the full path (e.g. `docs/issues/<feature-slug>/<NN>-<slug>.md`). Issue numbers are scoped per feature directory, not globally unique — always use the full path to locate a file unambiguously.
+- Issue state is tracked via labels matching the canonical triage vocabulary (see `docs/agents/labels.md`).
+- Dependency links use the tracker's native "blocked by" field; for local-markdown, use a `## Blocked by` heading.
+- The tracker adapter (`lib/tracker-adapter.mjs`) translates canonical label names to tracker strings at write time.
diff --git a/docs/agents/labels.md b/docs/agents/labels.md
new file mode 100644
index 00000000..0b549b6a
--- /dev/null
+++ b/docs/agents/labels.md
@@ -0,0 +1,43 @@
+<!--
+  AUTO-GENERATED by unic-archon-dlc — DO NOT EDIT BY HAND.
+  Source: apps/claude-code/unic-archon-dlc/lib/agent-docs-writer.mjs
+  Regenerate: re-run `/unic-archon-dlc:setup` (or call `runInstall()` from `lib/install-runner.mjs`).
+  This file describes the unic-archon-dlc target workflow; the current driver
+  in any given Consumer repo may differ.
+-->
+
+# Labels
+
+Three-tier taxonomy for `github`. Canonical names are used inside workflows; the tracker adapter maps them to tracker strings at write time.
+
+## State labels
+
+| Canonical       | Tracker string  |
+| --------------- | --------------- |
+| needs-triage    | needs-triage    |
+| needs-info      | needs-info      |
+| needs-specs     | needs-specs     |
+| ready-for-agent | ready-for-agent |
+| ready-for-human | ready-for-human |
+| resolved        | resolved        |
+| closed          | closed          |
+| rejected        | rejected        |
+
+## Type labels
+
+| Canonical | Tracker string |
+| --------- | -------------- |
+| feature   | feature        |
+| bug       | bug            |
+| spike     | spike          |
+| tech-debt | tech-debt      |
+| docs      | docs           |
+
+## Priority labels
+
+| Canonical | Tracker string |
+| --------- | -------------- |
+| p0        | p0             |
+| p1        | p1             |
+| p2        | p2             |
+| p3        | p3             |
diff --git a/docs/agents/workflow.md b/docs/agents/workflow.md
new file mode 100644
index 00000000..529496be
--- /dev/null
+++ b/docs/agents/workflow.md
@@ -0,0 +1,29 @@
+<!--
+  AUTO-GENERATED by unic-archon-dlc — DO NOT EDIT BY HAND.
+  Source: apps/claude-code/unic-archon-dlc/lib/agent-docs-writer.mjs
+  Regenerate: re-run `/unic-archon-dlc:setup` (or call `runInstall()` from `lib/install-runner.mjs`).
+  This file describes the unic-archon-dlc target workflow; the current driver
+  in any given Consumer repo may differ.
+-->
+
+# Workflow Phases
+
+unic-archon-dlc ships seven Archon workflow YAML DAGs. Six are lifecycle phases (explore → plan → build → qa → cleanup → triage) that produce persistent artifacts committed to `docs/workflow/<slug>/`. The `review` workflow is on-demand and posts (or updates) a structured review on the current PR via the configured tracker backend.
+
+| Phase   | Command                    | Artifact outputs                                                                                              |
+| ------- | -------------------------- | ------------------------------------------------------------------------------------------------------------- |
+| explore | `/unic-dlc-explore <slug>` | `docs/workflow/<slug>/findings.md`                                                                            |
+| plan    | `/unic-dlc-plan <slug>`    | `docs/workflow/<slug>/PRD.md`, `issues.json`, `build-<slug>.yaml`                                             |
+| build   | `/unic-dlc-build <slug>`   | `docs/workflow/<slug>/report.md`                                                                              |
+| qa      | `/unic-dlc-qa <slug>`      | merged PR                                                                                                     |
+| cleanup | `/unic-dlc-cleanup <slug>` | `docs/workflow/<slug>/arch-review.md`                                                                         |
+| triage  | `/unic-dlc-triage`         | `HANDOFF.md`, `docs/workflow/ROADMAP.md`                                                                      |
+| review  | `/unic-dlc-review`         | PR comment (github/ado/jira) or `docs/workflow/<slug>/review-comment.md` (local-markdown); idempotent re-runs |
+
+## State separation
+
+| Layer                        | Storage                          | Who owns it       |
+| ---------------------------- | -------------------------------- | ----------------- |
+| Transient workflow state     | `$ARTIFACTS_DIR` (Archon native) | Archon runtime    |
+| Persistent project artifacts | `docs/workflow/<slug>/`          | Committed to repo |
+| Issue / ticket tracking      | Configured tracker               | Tracker backend   |
diff --git a/docs/inbox/README.md b/docs/inbox/README.md
deleted file mode 100644
index 8f22680e..00000000
--- a/docs/inbox/README.md
+++ /dev/null
@@ -1,43 +0,0 @@
-# Inbox
-
-A low-friction capture zone for raw ideas, bugs, questions, and anything else that pops up mid-conversation and shouldn't be lost. Items here have not yet been grilled, scoped, or assigned a feature slug.
-
-## How to add an item
-
-Run `/inbox <one-liner>` in Claude Code. That's it.
-
-```
-/inbox pr-review should support GitLab
-/inbox unic-confluence image upload support
-/inbox investigate windows path issue in auto-format
-```
-
-A file is created at `docs/inbox/<slug>.md`. No follow-up needed.
-
-## File format
-
-```markdown
----
-title: pr-review should support GitLab
-created: 2026-05-03
----
-
-pr-review should support GitLab
-```
-
-No `status` field — the location is the status. Every file here is awaiting triage.
-
-## Graduation
-
-When you're ready to act on an item:
-
-1. Open the file and run `/grill-with-docs` or `/grill-me` to refine it
-2. Run `/to-prd` to synthesize into a PRD → creates `docs/issues/<slug>/PRD.md`
-3. Manually delete the inbox file (or let the skill handle it)
-
-Items do **not** move to `docs/issues/<slug>/` until a scope and slug have been agreed via grilling.
-
-## Related
-
-- `docs/agents/issue-tracker.md` — conventions for structured issues post-grilling
-- `docs/agents/triage-labels.md` — 8-state vocabulary used in `docs/issues/` (not here)
diff --git a/docs/issues/adr-reshelving-harness-internal/PRD.md b/docs/issues/adr-reshelving-harness-internal/PRD.md
new file mode 100644
index 00000000..6c5c04bd
--- /dev/null
+++ b/docs/issues/adr-reshelving-harness-internal/PRD.md
@@ -0,0 +1,239 @@
+---
+title: Reshelve harness-internal ADRs into the unic-archon-dlc plugin
+created: 2026-05-23
+---
+
+**Status:** ready-for-agent
+**Plugin:** `apps/claude-code/unic-archon-dlc` (plus monorepo-wide doc edits)
+**Category:** documentation
+
+## Problem Statement
+
+The root `docs/adr/` directory currently holds 31 ADRs. Nine of them are scoped to the `unic-archon-dlc` plugin's internals — the Feature Runner, the spec template format, the retired Ralph orchestrator, the retired `/implement-feature` skill — yet they live alongside genuinely monorepo-wide decisions about pnpm, Biome, CI, releases, GPG signing, and the like.
+
+That mixing has three concrete costs:
+
+- The root ADR index is harder to scan because half of the post-0019 entries describe a single plugin's harness rather than cross-cutting monorepo conventions.
+- The scope of each decision is miscommunicated. A reader of root `docs/adr/` reasonably expects every entry to be a repo-level decision; the harness ADRs are not.
+- Supersession chains (0030 supersedes 0020 + 0024; 0031 supersedes 0027 + 0029) are read as monorepo history rather than as the harness's evolution.
+
+This Feature reshelves the nine harness-internal ADRs into the plugin's own `docs/adr/` directory, sweeps inbound references, leaves redirect stubs for external links, and updates the auto-generated `docs/agents/domain.md` so the convention is documented at the surface where agents read it.
+
+## Solution
+
+Nine ADR files move from root `docs/adr/` to `apps/claude-code/unic-archon-dlc/docs/adr/` and are renumbered to a fresh sequence (0002–0010, continuing after the existing 0001-setup-as-slash-command). Each renumbered ADR carries a single-line "Renumbered from monorepo-root ADR-NNNN" trace under its Status header so git archaeology continues to lead somewhere useful.
+
+At each old root path, a one-line redirect stub replaces the original file. The stub preserves external links (GitHub permalinks shared in Slack, PR descriptions, issue comments) and honours the existing "Never delete an ADR" rule from `docs/adr/README.md`. Internal inbound links across four monorepo docs are swept to point at the new locations directly.
+
+The `buildDomainDoc` template in `lib/agent-docs-writer.mjs` is generalised in its multi-context branch to mention that each context may keep its own `docs/adr/`. The wording is generic — no hardcoded path to this monorepo's structure leaks into the harness. A regenerated `docs/agents/domain.md` (produced by re-running `/unic-archon-dlc:setup` against this repo) lands in the same PR so the dogfood state matches the new generator output. The per-plugin ADR index README is refreshed to list all ten files. A single line is added to `CONTEXT-MAP.md` under "Relationships" describing the ADR-location split.
+
+## User Stories
+
+1. As a maintainer scanning root `docs/adr/`, I want only monorepo-wide decisions to appear, so that the index communicates scope correctly.
+2. As a contributor opening the `unic-archon-dlc` plugin directory, I want every harness-internal ADR to live under `apps/claude-code/unic-archon-dlc/docs/adr/`, so that I can read the plugin's design history without cross-jumping to the root.
+3. As a contributor reading a renumbered ADR, I want a "Renumbered from monorepo-root ADR-NNNN" trace at the top, so that I can correlate the file with old commit messages that refer to the original number.
+4. As a maintainer chasing a GitHub permalink shared months ago to `docs/adr/0030-retire-ralph-adopt-archon-runner.md`, I want a redirect stub at that path pointing to the new home, so that the external link still leads somewhere actionable.
+5. As a contributor reading `CONTRIBUTING.md`, I want references to ADR-0030 / ADR-0031 to resolve directly to their new locations under the plugin, so that I do not pay an extra redirect hop.
+6. As a contributor reading `docs/process/ai-development.md` or `docs/process/development-workflow.md`, I want every ADR link to point at the new harness path with the new number, so that the process docs reflect current structure.
+7. As a contributor reading `docs/agents/feature-runner.md`, I want the ADR-0028 / 0030 / 0031 links to resolve directly to the new harness paths, so that the agent-facing doc stays accurate after reshelving.
+8. As a contributor following an ADR supersession chain (0020 superseded by 0030, etc.), I want both ends of the chain to live in the same directory, so that I do not have to cross monorepo / plugin boundaries to read the history.
+9. As an agent reading `docs/agents/domain.md`, I want the multi-context section to acknowledge that per-context `docs/adr/` directories are a permitted pattern, so that I do not assume root `docs/adr/` is the only location for architectural decisions.
+10. As a maintainer of `unic-archon-dlc` running setup against a Consumer that uses a multi-context layout, I want the harness to describe the per-context ADR pattern generically — without baking in this monorepo's plugin path — so that the generated `docs/agents/domain.md` is portable.
+11. As a contributor opening `CONTEXT-MAP.md`, I want a one-line note that monorepo-wide decisions live in root `docs/adr/` and per-context decisions live in each context's `docs/adr/`, so that the routing rule is discoverable from the structural index.
+12. As a contributor opening `apps/claude-code/unic-archon-dlc/docs/adr/README.md`, I want the index table to list all ten ADRs in the directory, so that I can navigate the plugin's design history without `ls`-ing the folder.
+13. As a maintainer running `pnpm --filter unic-archon-dlc test` after the generator change, I want a node:test assertion that `buildDomainDoc` emits the per-context-ADR phrase in multi-context mode and omits it in single-context mode, so that future edits to the generator wording fail CI rather than silently regressing.
+14. As a maintainer verifying the regenerated `docs/agents/domain.md`, I want the file to be produced by a real `/unic-archon-dlc:setup` run — not hand-edited — so that the dogfood state matches what the harness would write in any other multi-context Consumer.
+15. As a maintainer reviewing the resulting PR, I want a single grep guard (`docs/adr/00\(20\|23\|24\|26\|27\|28\|29\|30\|31\)`) to return only the redirect stubs and historical-only references, so that I can confirm no live link points at the old paths.
+16. As a contributor reading the dogfood-banner PRD or the custom-spec-runner-grill-02 conversation, I want those name-only references (`ADR-0030`, `ADR-0031`, etc.) to be left alone, so that historical wording is preserved.
+17. As a maintainer following Gitflow, I want this work to land via a feature branch targeting `develop`, so that the release flow remains predictable.
+
+## Implementation Decisions
+
+### Renumbering policy
+
+ADRs move with a fresh sequence starting at `0002` (the existing `0001-setup-as-slash-command.md` keeps its slot). The mapping is fixed by the move order — original ascending order maps to new ascending order:
+
+| Old number | New number | Slug                             |
+| ---------- | ---------- | -------------------------------- |
+| 0020       | 0002       | per-plugin-ralph-loops           |
+| 0023       | 0003       | spec-template-format             |
+| 0024       | 0004       | ralph-atomic-iteration           |
+| 0026       | 0005       | tdd-dispatch-by-version-impact   |
+| 0027       | 0006       | feature-runner-context-bundle    |
+| 0028       | 0007       | blocked-by-canonical-sequencing  |
+| 0029       | 0008       | feature-runner-afk-invocation    |
+| 0030       | 0009       | retire-ralph-adopt-archon-runner |
+| 0031       | 0010       | retire-implement-feature-skill   |
+
+Slugs are preserved verbatim. Each renumbered file gains a single trace line immediately under the Status header, of the form `Renumbered from monorepo-root ADR-NNNN (2026-05).`
+
+Per-directory numbering is already the documented convention in root `docs/adr/README.md` ("Files are named NNNN-slug.md, zero-padded to 4 digits, per directory"), so no policy text needs changing.
+
+### Redirect stub format
+
+At each of the nine original root paths, the file is replaced with a one-line redirect stub. The stub is intentionally short and clearly marked so it is not mistaken for an active ADR:
+
+```markdown
+# NNNN. <original title>
+
+> **Moved** to [apps/claude-code/unic-archon-dlc/docs/adr/00MM-<slug>.md](../../apps/claude-code/unic-archon-dlc/docs/adr/00MM-<slug>.md) (2026-05).
+>
+> This ADR is harness-internal and now lives with the unic-archon-dlc plugin. The redirect preserves external links; new references should use the new path.
+```
+
+The stub keeps the original NNNN title for index searchability but carries no Status / Context / Decision / Consequences sections. The body is a single blockquote callout so a reader scanning the file sees the redirect immediately and does not read further looking for content that is no longer there.
+
+### Supersession chains stay intact
+
+Both cohorts move together:
+
+- **Active** (still load-bearing): 0023, 0026, 0028, 0030, 0031 — become 0003, 0005, 0007, 0009, 0010.
+- **Superseded** (historical only): 0020 (→ 0002), 0024 (→ 0004), 0027 (→ 0006), 0029 (→ 0008).
+
+Inside the moved set, every ADR-to-ADR reference (e.g. "Superseded by ADR-0030", "see ADR-0027 / ADR-0029") is updated to the new harness-local number. All references are intra-directory after the move, so relative paths simplify (e.g. `0030-retire-ralph-adopt-archon-runner.md` becomes `0009-retire-ralph-adopt-archon-runner.md`).
+
+### Internal link sweep
+
+Four monorepo docs contain inbound links to the moved set. Each is updated so every reference points at the new harness path with the new number. No old-path link survives in these files:
+
+- `CONTRIBUTING.md` — two links (0030, 0031) → (0009, 0010).
+- `docs/agents/feature-runner.md` — three links (0028, 0030, 0031) → (0007, 0009, 0010).
+- `docs/process/ai-development.md` — five links (0027, 0028, 0029, 0030, 0031) → (0006, 0007, 0008, 0009, 0010).
+- `docs/process/development-workflow.md` — three links (0028, 0030, 0031) → (0007, 0009, 0010).
+
+Relative path prefixes adjust accordingly (e.g. `docs/agents/feature-runner.md` previously used `../adr/0028-…`; the new link becomes `../../apps/claude-code/unic-archon-dlc/docs/adr/0007-…`).
+
+Two name-only references are explicitly left untouched:
+
+- `docs/issues/unic-archon-dlc-dogfood-banner/PRD.md` mentions "ADR-0030 and ADR-0031" by name only (no link). It is an already-merged PRD describing past context; updating the wording would muddy the historical record.
+- `docs/conversations/custom-spec-runner-grill-02.md` is a verbatim historical conversation transcript.
+
+### Generator change in `lib/agent-docs-writer.mjs`
+
+The `buildDomainDoc(c, projectDir)` function's multi-context branch currently emits two bullet points:
+
+```
+- **Context map:** `CONTEXT-MAP.md`
+- **ADRs:** `docs/adr/` (repo-level decisions)
+```
+
+The second bullet is generalised to acknowledge that each context may keep its own `docs/adr/`. The wording stays portable — no hardcoded plugin path, no mention of `unic-archon-dlc` specifically. Suggested wording (final form is owned by the implementer subject to the constraint):
+
+```
+- **Context map:** `CONTEXT-MAP.md`
+- **ADRs:** monorepo-wide decisions live in root `docs/adr/`; each context may also keep its own `docs/adr/` for decisions scoped to that context.
+```
+
+The single-context branch is untouched.
+
+No other lib module changes. The function signature, return shape, and call sites stay identical.
+
+### Regenerated `docs/agents/domain.md`
+
+After the generator change ships, the existing `docs/agents/domain.md` in this repo is regenerated by running `/unic-archon-dlc:setup` against the repo. The dogfood banner already present at the top of the file remains in place (it is reapplied by `prependBanner(AGENT_DOC_BANNER, …)` in `writeAgentDocs`). Hand-editing the file is explicitly forbidden by the banner; the implementation must use the setup re-run path so the diff matches what the updated generator would produce in any Consumer.
+
+### `CONTEXT-MAP.md` relationship note
+
+A single line is added to `CONTEXT-MAP.md` under the existing "Relationships" section:
+
+```
+- Architectural decisions are split by scope: monorepo-wide decisions live in root `docs/adr/`; decisions scoped to a single context live in that context's own `docs/adr/`.
+```
+
+The `CONTEXT.md` glossary is **not** touched. Adding a "Plugin ADR" term would violate its strict ubiquitous-language-only role; the ADR location is a structural fact, not domain vocabulary.
+
+### Plugin ADR README refresh
+
+`apps/claude-code/unic-archon-dlc/docs/adr/README.md` currently lists only 0001 in its index table. After the move it is refreshed to list all ten ADRs in numeric order. The existing pointer at the top of the file ("Plugin-scoped architectural decisions. Repo-wide decisions live in `docs/adr/` at the monorepo root.") stays unchanged.
+
+### Release impact
+
+The generator change to `lib/agent-docs-writer.mjs` is a behavioural change to the on-disk output of generated files in any multi-context Consumer using `unic-archon-dlc`. Version impact: **patch**.
+
+- Run `pnpm --filter unic-archon-dlc bump patch`.
+- Add a dated CHANGELOG entry under the new version describing the generator change.
+
+The remaining edits (ADR moves, stubs, link sweep, README refresh, CONTEXT-MAP line, regenerated `docs/agents/domain.md`) are documentation only and require no further version bump. `verify:changelog` runs PR-only (ADR-0017) and will be satisfied by the patch bump entry.
+
+### Commits and branch
+
+- Branch: `feature/adr-reshelving-harness-internal`, targeting `develop`.
+- Commit scopes follow Conventional Commits (ADR-0021):
+  - `docs(adr): …` for the ADR moves, stubs, link sweep, plugin ADR README refresh, and `CONTEXT-MAP.md` edit.
+  - `chore(unic-archon-dlc): …` for the `lib/agent-docs-writer.mjs` change, the new test assertions, the `bump patch`, the CHANGELOG entry, and the regenerated `docs/agents/domain.md`.
+- Per repo doctrine: never touch `LICENSE` files.
+
+## Testing Decisions
+
+The change splits cleanly into a mechanical doc move (verified by review + grep guard) and a small generator change (verified by node:test assertions + the setup re-run).
+
+### Extend `test/install-agent-docs.test.mjs`
+
+Prior art: this file already exercises `writeAgentDocs(projectDir, config)` end-to-end by writing into an `os.tmpdir()` sandbox and asserting file contents. The new behaviour is added by extending the existing tests rather than creating a new file, mirroring how the dogfood-banner work extended the same test in #124.
+
+New assertions:
+
+- When `config.repo_layout === 'multi-context'`, the generated `docs/agents/domain.md` contains the per-context-ADR phrase introduced in the multi-context branch of `buildDomainDoc`.
+- When `config.repo_layout` is `'single-context'` (or unset, defaulting to single-context), the generated `docs/agents/domain.md` does **not** contain the per-context-ADR phrase.
+- Both branches continue to include the `AGENT_DOC_BANNER` prefix (regression guard for the dogfood-banner contract).
+
+Tests assert observable file contents — not the internal call shape of `buildDomainDoc`. They run under `node:test` + `assert/strict`, consistent with the rest of the test suite.
+
+### Grep guard
+
+Mechanical regression guard run by the implementer (and the reviewer) before merge:
+
+```sh
+grep -rn "docs/adr/00\(20\|23\|24\|26\|27\|28\|29\|30\|31\)" \
+  --include="*.md" --include="*.mjs" \
+  --include="*.js" --include="*.ts" \
+  -- .
+```
+
+Expected results:
+
+- The nine redirect stubs at `docs/adr/00XX-…md` (each contains its own old path in the file's first heading line — that is fine).
+- The historical conversation doc `docs/conversations/custom-spec-runner-grill-02.md` (name-only mentions, left alone).
+
+Any other result indicates a missed link.
+
+### Visual diff review
+
+The reviewer confirms:
+
+- All nine moved files exist in `apps/claude-code/unic-archon-dlc/docs/adr/` with their new numbers and slugs.
+- Each renumbered file carries the "Renumbered from monorepo-root ADR-NNNN" trace under Status.
+- ADR-to-ADR supersession references inside the moved set use the new numbers.
+- The nine redirect stubs at root use the stub format above and link to the correct new paths.
+- The four sweep-target files (`CONTRIBUTING.md`, `docs/agents/feature-runner.md`, `docs/process/ai-development.md`, `docs/process/development-workflow.md`) link to the new paths.
+- `apps/claude-code/unic-archon-dlc/docs/adr/README.md` index table lists all ten ADRs.
+- `CONTEXT-MAP.md` Relationships section contains the one-line ADR split note.
+- `docs/agents/domain.md` matches what `/unic-archon-dlc:setup` produces against this repo (i.e. is not hand-edited).
+
+### Build / lint gates
+
+- `pnpm check` — Biome + Prettier on the whole tree.
+- `pnpm typecheck` — `tsc --checkJs --noEmit` over `.mjs` sources.
+- `pnpm --filter unic-archon-dlc test` — node:test for the plugin (covers the extended `install-agent-docs.test.mjs`).
+- `pnpm test` from root — runs the full per-package test sweep.
+- `pnpm --filter unic-archon-dlc verify:changelog` — guards the CHANGELOG entry for the new patch version.
+
+## Out of Scope
+
+- Any change to the retained root ADRs (0001–0019, 0021, 0022, 0025). These are genuine monorepo-wide decisions and stay at root.
+- Any change to `LICENSE` files anywhere in the tree (root doctrine — maintainer-managed).
+- Touching `docs/conversations/custom-spec-runner-grill-02.md`, the dogfood-banner PRD, or any other historical-only reference.
+- Renaming or restructuring the `apps/claude-code/unic-archon-dlc/docs/adr/` directory itself.
+- Adding any new vocabulary term to `CONTEXT.md` (glossary stays untouched).
+- Touching the per-plugin `docs/plans/` directories — handled by the cleanup chore PR.
+- Refreshing per-plugin `CLAUDE.md` / `AGENTS.md` files — tracked separately in #125.
+- The dogfood-banner change — already shipped (#124).
+- Any implementation against this PRD itself — the PRD is the deliverable; implementation follows via `/tdd` once the issue is picked up.
+
+## Further Notes
+
+- The reshelving is reversible: the moves are renames + content edits in one PR, and a follow-up PR could swap the directions if the convention is later revised. The redirect stubs guarantee external links keep resolving regardless.
+- The `unic-archon-dlc` plugin is still not operational as a Feature Runner in this repo (manual `/tdd` remains the path per the soon-to-be-renumbered ADR-0009 and ADR-0010). The reshelving makes the harness-internal scope of those decisions structurally explicit without changing operational practice.
+- Future harness ADRs go straight into `apps/claude-code/unic-archon-dlc/docs/adr/` from inception. There is no expected back-pressure to revisit this split.
+- The generator change is intentionally light: it adds one informational bullet, not a structural shift. A future Consumer using a multi-context layout will see the same wording without any plugin-specific assumption leaking through.
+- Tag for the patch release will be `unic-archon-dlc@<new-version>`, created automatically by the release workflow on merge to `main` (ADR-0018).
diff --git a/docs/issues/auto-format-config/01-config-module.md b/docs/issues/auto-format-config/01-config-module.md
deleted file mode 100644
index 76b08e66..00000000
--- a/docs/issues/auto-format-config/01-config-module.md
+++ /dev/null
@@ -1,39 +0,0 @@
-# Extract `lib/config.mjs` with `DEFAULTS`, `loadConfig`, and tests
-
-**Status:** closed
-**Category:** refactor
-
-## Parent
-
-`docs/issues/auto-format-config/PRD.md`
-
-## What to build
-
-Create `scripts/lib/config.mjs` and `scripts/lib/config.test.mjs`.
-
-`config.mjs` exports:
-
-- `DEFAULTS` — the `ProjectConfig` constant currently at lines 25–57 of `format-hook.mjs`
-- `loadConfig(projectDir: string) → ProjectConfig` — the merge logic currently in
-  `loadProjectConfig()`, with `PROJECT_DIR` replaced by the `projectDir` parameter
-
-`config.test.mjs` covers ten scenarios via real temp directories (see PRD for the full list).
-
-`package.json#scripts.test` is updated to include `scripts/lib/config.test.mjs`.
-
-`format-hook.mjs` is not modified in this issue.
-
-Full implementation details in `apps/claude-code/auto-format/docs/plans/29-config-module.md`.
-
-## Acceptance criteria
-
-- [ ] `scripts/lib/config.mjs` exports `DEFAULTS` and `loadConfig`
-- [ ] `loadConfig` takes `projectDir: string`, does not close over any module-level constant
-- [ ] All 10 tests in `scripts/lib/config.test.mjs` pass
-- [ ] `pnpm test` passes (existing tests unaffected)
-- [ ] `pnpm typecheck` passes
-- [ ] `format-hook.mjs` is not modified
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/auto-format-config/02-wire-up.md b/docs/issues/auto-format-config/02-wire-up.md
deleted file mode 100644
index f14b0eab..00000000
--- a/docs/issues/auto-format-config/02-wire-up.md
+++ /dev/null
@@ -1,31 +0,0 @@
-# Update `format-hook.mjs` to use `lib/config.mjs`
-
-**Status:** closed
-**Category:** refactor
-
-## Parent
-
-`docs/issues/auto-format-config/PRD.md`
-
-## What to build
-
-Update `scripts/format-hook.mjs` to import `loadConfig` from `./lib/config.mjs` and remove the
-now-moved code:
-
-1. Add `import { loadConfig } from './lib/config.mjs'`
-2. Delete the `DEFAULTS` constant (lines 25–57)
-3. Delete the `loadProjectConfig` function (lines 65–93)
-4. Replace `const CONFIG = loadProjectConfig()` with `const CONFIG = loadConfig(PROJECT_DIR)`
-
-External behaviour is identical. `tests/format-hook.test.mjs` must pass without modification.
-
-## Acceptance criteria
-
-- [ ] `grep -n "const DEFAULTS\|function loadProjectConfig" scripts/format-hook.mjs` → 0 matches
-- [ ] `grep -n "loadConfig" scripts/format-hook.mjs` → match on import and call site
-- [ ] `pnpm test` passes without any changes to `tests/format-hook.test.mjs`
-- [ ] `pnpm typecheck` passes
-
-## Blocked by
-
-01-config-module
diff --git a/docs/issues/auto-format-config/03-version-bump.md b/docs/issues/auto-format-config/03-version-bump.md
deleted file mode 100644
index 5959abb7..00000000
--- a/docs/issues/auto-format-config/03-version-bump.md
+++ /dev/null
@@ -1,28 +0,0 @@
-# Version bump and CHANGELOG entry
-
-**Status:** closed
-**Category:** release
-
-## Parent
-
-`docs/issues/auto-format-config/PRD.md`
-
-## What to build
-
-Bump the auto-format plugin version and add a dated CHANGELOG entry.
-
-```sh
-pnpm --filter auto-format bump patch
-```
-
-CHANGELOG entry should describe the config loading extraction as an internal refactor.
-
-## Acceptance criteria
-
-- [ ] `pnpm --filter auto-format verify:changelog` passes
-- [ ] CHANGELOG entry present for the new version
-- [ ] `plugin.json` and `marketplace.json` versions are in sync
-
-## Blocked by
-
-02-wire-up
diff --git a/docs/issues/auto-format-runners/01-formatter-descriptor-type.md b/docs/issues/auto-format-runners/01-formatter-descriptor-type.md
deleted file mode 100644
index b1848a8c..00000000
--- a/docs/issues/auto-format-runners/01-formatter-descriptor-type.md
+++ /dev/null
@@ -1,39 +0,0 @@
-# Add `FormatterDescriptor` typedef to `lib/types.mjs`
-
-**Status:** closed
-**Category:** refactor
-
-## Parent
-
-`docs/issues/auto-format-runners/PRD.md`
-
-## What to build
-
-Add a `FormatterDescriptor` JSDoc typedef to `scripts/lib/types.mjs`, after the existing
-`ProjectConfig` definition:
-
-```js
-/**
- * Descriptor for a single formatter invocation.
- *
- * @typedef {{
- *   name: string,
- *   bin: string,
- *   args: (filePath: string) => string[],
- *   warnIfMissing?: boolean,
- *   toleratedStatuses?: number[],
- * }} FormatterDescriptor
- */
-```
-
-No runtime code changes. No other files touched.
-
-## Acceptance criteria
-
-- [ ] `grep -n "FormatterDescriptor" scripts/lib/types.mjs` → match
-- [ ] All five fields present with correct types
-- [ ] `pnpm typecheck` passes
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/auto-format-runners/02-runner-module.md b/docs/issues/auto-format-runners/02-runner-module.md
deleted file mode 100644
index 975faed7..00000000
--- a/docs/issues/auto-format-runners/02-runner-module.md
+++ /dev/null
@@ -1,38 +0,0 @@
-# Extract `lib/runners.mjs` with `runFormatter` and tests
-
-**Status:** closed
-**Category:** refactor
-
-## Parent
-
-`docs/issues/auto-format-runners/PRD.md`
-
-## What to build
-
-Create two new files: `scripts/lib/runners.mjs` and `scripts/lib/runners.test.mjs`.
-
-`runners.mjs` exports a single `runFormatter(descriptor, filePath, cwd, timeoutMs)` function that
-owns the subprocess contract — binary existence check, `spawnSync` invocation, SIGTERM/null
-detection, tolerated-exit-code filtering, stderr reporting. It imports `FormatterDescriptor` from
-`./types.mjs` via JSDoc `@import`. It has no knowledge of Prettier, ESLint, or Biome.
-
-`runners.test.mjs` tests the contract with real child processes (stub Node.js scripts in temp
-directories). Six tests covering: missing binary silent, missing binary warn, timeout, tolerated
-status, non-tolerated status, args passed correctly.
-
-`package.json#scripts.test` is updated to include `scripts/lib/runners.test.mjs` so `pnpm test`
-covers the new file.
-
-Full implementation details in `apps/claude-code/auto-format/docs/plans/26-runner-module.md`.
-
-## Acceptance criteria
-
-- [ ] `scripts/lib/runners.mjs` exists and exports `runFormatter`
-- [ ] All 6 tests in `scripts/lib/runners.test.mjs` pass
-- [ ] `pnpm test` passes (existing tests unaffected)
-- [ ] `pnpm typecheck` passes
-- [ ] `format-hook.mjs` is not modified in this issue
-
-## Blocked by
-
-01-formatter-descriptor-type
diff --git a/docs/issues/auto-format-runners/03-replace-runner-functions.md b/docs/issues/auto-format-runners/03-replace-runner-functions.md
deleted file mode 100644
index 48c1530b..00000000
--- a/docs/issues/auto-format-runners/03-replace-runner-functions.md
+++ /dev/null
@@ -1,38 +0,0 @@
-# Replace runner functions with descriptors in `format-hook.mjs`
-
-**Status:** closed
-**Category:** refactor
-
-## Parent
-
-`docs/issues/auto-format-runners/PRD.md`
-
-## What to build
-
-Replace the three inline runner functions (`runPrettier`, `runEslint`, `runBiome`) in
-`scripts/format-hook.mjs` with three `FormatterDescriptor` constants and a dispatch through
-`runFormatter` from `lib/runners.mjs`.
-
-Changes:
-
-1. Add `import { runFormatter } from './lib/runners.mjs'`
-2. Add `/** @import { FormatterDescriptor } from './lib/types.mjs' */` type-import tag
-3. Define `PRETTIER_DESCRIPTOR`, `ESLINT_DESCRIPTOR`, `BIOME_DESCRIPTOR` constants
-4. Update `main()` dispatch to bind `cwd` and `timeoutMs` once and call `runFormatter`
-5. Delete `runPrettier`, `runEslint`, `runBiome` functions
-
-External behaviour is identical. `tests/format-hook.test.mjs` must pass without modification.
-
-Full implementation details in
-`apps/claude-code/auto-format/docs/plans/27-replace-runner-functions.md`.
-
-## Acceptance criteria
-
-- [ ] `grep -n "function runPrettier\|function runEslint\|function runBiome" scripts/format-hook.mjs` → 0 matches
-- [ ] `grep -n "PRETTIER_DESCRIPTOR\|ESLINT_DESCRIPTOR\|BIOME_DESCRIPTOR" scripts/format-hook.mjs` → 3 matches
-- [ ] `pnpm test` passes without any changes to `tests/format-hook.test.mjs`
-- [ ] `pnpm typecheck` passes
-
-## Blocked by
-
-02-runner-module
diff --git a/docs/issues/auto-format-runners/04-version-bump.md b/docs/issues/auto-format-runners/04-version-bump.md
deleted file mode 100644
index be38de12..00000000
--- a/docs/issues/auto-format-runners/04-version-bump.md
+++ /dev/null
@@ -1,28 +0,0 @@
-# Version bump and CHANGELOG entry
-
-**Status:** closed
-**Category:** release
-
-## Parent
-
-`docs/issues/auto-format-runners/PRD.md`
-
-## What to build
-
-Bump the auto-format plugin version and add a dated CHANGELOG entry.
-
-```sh
-pnpm --filter auto-format bump patch
-```
-
-CHANGELOG entry should describe the formatter runner extraction as an internal refactor.
-
-## Acceptance criteria
-
-- [ ] `pnpm --filter auto-format verify:changelog` passes
-- [ ] CHANGELOG entry present for the new version
-- [ ] `plugin.json` and `marketplace.json` versions are in sync
-
-## Blocked by
-
-03-replace-runner-functions
diff --git a/docs/issues/feature-runner/01-skill-scaffold.md b/docs/issues/feature-runner/01-skill-scaffold.md
deleted file mode 100644
index fdd896c5..00000000
--- a/docs/issues/feature-runner/01-skill-scaffold.md
+++ /dev/null
@@ -1,29 +0,0 @@
-# Skill scaffold and minimal execution loop
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Create `.claude/skills/implement-feature/SKILL.md`. Implement the full happy path for a named-slug invocation on a feature with one or more issues: create a `feature/afk/<slug>` worktree from `develop`, read all `NN-*.md` files in the feature directory, filter to those with `Status: ready-for-agent` (skip `resolved` and `closed`), sort numerically, invoke `/tdd` as a non-interactive sub-agent for each issue in order with the issue file and PRD as context, mark each issue `Status: resolved` on successful completion.
-
-All file and git operations must use Claude tool calls — no shell scripts, no POSIX paths. This is the tracer bullet: the entire end-to-end happy path without failure handling, PR creation, or progress output.
-
-## Acceptance criteria
-
-- [ ] `.claude/skills/implement-feature/SKILL.md` exists and is invocable as `/implement-feature <slug>`
-- [ ] Running `/implement-feature <slug>` creates a `feature/afk/<slug>` branch and worktree from `develop`
-- [ ] Only issues at `Status: ready-for-agent` are processed; `resolved` and `closed` issues are skipped
-- [ ] Each issue is handed to `/tdd` as a non-interactive sub-agent invocation with the issue file and PRD as context
-- [ ] Each issue file is updated to `Status: resolved` after successful `/tdd` completion
-- [ ] All file and git operations use Claude tool calls, not shell scripts (cross-platform requirement)
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/feature-runner/02-failure-handling.md b/docs/issues/feature-runner/02-failure-handling.md
deleted file mode 100644
index f3fde0a0..00000000
--- a/docs/issues/feature-runner/02-failure-handling.md
+++ /dev/null
@@ -1,26 +0,0 @@
-# Failure handling
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Extend the skill to handle `/tdd` sub-agent failures. When `/tdd` fails on an issue, append a failure note to that issue file under a `## Comments` heading, leave the issue at `Status: ready-for-agent`, stop the runner, and leave the worktree in place for inspection. Subsequent issues in the feature must not run — they could inherit a broken foundation.
-
-## Acceptance criteria
-
-- [ ] When `/tdd` fails on an issue, a failure note is appended to that issue file under `## Comments` (prefixed with the AI-generated disclaimer)
-- [ ] The failing issue remains at `Status: ready-for-agent` after the failure note is appended
-- [ ] The runner stops after the first failure; no subsequent issues in the feature are executed
-- [ ] The worktree is left in place (not removed) on failure, for inspection
-- [ ] The runner surfaces a clear failure message to the user naming which issue failed
-
-## Blocked by
-
-`docs/issues/feature-runner/01-skill-scaffold.md`
diff --git a/docs/issues/feature-runner/03-pr-creation.md b/docs/issues/feature-runner/03-pr-creation.md
deleted file mode 100644
index 84c03cbd..00000000
--- a/docs/issues/feature-runner/03-pr-creation.md
+++ /dev/null
@@ -1,28 +0,0 @@
-# PR creation and worktree cleanup
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Extend the skill so that when all issues in a feature are `resolved`, it opens a pull request targeting `develop` using the `gh` CLI. The PR title is derived from the feature slug and the PRD's `title` frontmatter field. The PR body references the feature PRD and lists all resolved issues. After the PR is opened successfully, the worktree is removed.
-
-Blocked on failure handling (02) so the runner never reaches PR creation when an issue has failed.
-
-## Acceptance criteria
-
-- [ ] When all issues in the feature are `resolved`, `gh pr create` is called targeting `develop`
-- [ ] PR title is derived from the feature slug and the PRD's `title` frontmatter field
-- [ ] PR body includes a reference to `docs/issues/<slug>/PRD.md` and a list of all resolved issues
-- [ ] The worktree is removed after the PR is opened successfully
-- [ ] No PR is opened if the runner stopped due to a `/tdd` failure (failure handling from 02 prevents this)
-
-## Blocked by
-
-`docs/issues/feature-runner/02-failure-handling.md`
diff --git a/docs/issues/feature-runner/04-progress-reporting.md b/docs/issues/feature-runner/04-progress-reporting.md
deleted file mode 100644
index 6f3ce247..00000000
--- a/docs/issues/feature-runner/04-progress-reporting.md
+++ /dev/null
@@ -1,24 +0,0 @@
-# Progress reporting
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Extend the skill to emit a progress line before each `/tdd` invocation so the user can judge when to check back. The line should name the current issue number, the total count, and the issue title.
-
-## Acceptance criteria
-
-- [ ] Before each `/tdd` invocation, the runner outputs a line in the format: `Implementing issue N of M: <issue title>`
-- [ ] N counts from 1 and M is the total number of `ready-for-agent` issues in the feature (excluding already `resolved` or `closed` ones at run start)
-- [ ] The progress line is emitted regardless of whether failure handling or PR creation are in place
-
-## Blocked by
-
-`docs/issues/feature-runner/01-skill-scaffold.md`
diff --git a/docs/issues/feature-runner/05-full-context-bundle.md b/docs/issues/feature-runner/05-full-context-bundle.md
deleted file mode 100644
index 5b9b26c9..00000000
--- a/docs/issues/feature-runner/05-full-context-bundle.md
+++ /dev/null
@@ -1,27 +0,0 @@
-# Full context bundle for /tdd sub-agent invocations
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Extend the `/tdd` sub-agent invocation to use the full context bundle defined in ADR-0027, replacing the minimal bundle (issue file + PRD only) from the scaffold slice. The bundle adds: all sibling issue files in the feature directory, domain-scoped `CONTEXT.md`, domain-scoped ADRs, and the last 5 git commits.
-
-Scope is inferred by scanning the PRD for `apps/claude-code/<plugin>` path references. If found, inject that plugin's `docs/adr/` and `CONTEXT.md`. If not found, inject the root `docs/adr/` and root `CONTEXT.md`.
-
-## Acceptance criteria
-
-- [ ] Each `/tdd` invocation receives the full bundle: issue file, PRD, all sibling issue files, scoped `CONTEXT.md`, scoped ADRs, last 5 git commits
-- [ ] For a feature whose PRD references paths under `apps/claude-code/<plugin>/`, the plugin's `docs/adr/` and `CONTEXT.md` are injected
-- [ ] For a feature whose PRD does not reference any `apps/claude-code/<plugin>/` path, the root `docs/adr/` and root `CONTEXT.md` are injected
-- [ ] Root ADRs are not injected into plugin feature runs
-
-## Blocked by
-
-`docs/issues/feature-runner/01-skill-scaffold.md`
diff --git a/docs/issues/feature-runner/06-dependency-graph.md b/docs/issues/feature-runner/06-dependency-graph.md
deleted file mode 100644
index 8de3d229..00000000
--- a/docs/issues/feature-runner/06-dependency-graph.md
+++ /dev/null
@@ -1,28 +0,0 @@
-# Dependency graph and topological ordering
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Extend the skill to parse `## Blocked by` references from all issue files in the feature, build a topological execution order, and replace the numerical sort from the scaffold slice. If any `## Blocked by` reference conflicts with numerical filename order, halt with a descriptive error before executing any issue. Issues with `## Blocked by: None` (or equivalent) have no predecessors.
-
-See ADR-0028 for the rationale: `## Blocked by` is the canonical dependency signal; numerical order is a UX convenience, not an execution contract.
-
-## Acceptance criteria
-
-- [ ] `## Blocked by` references are parsed from all `NN-*.md` files in the feature directory before execution begins
-- [ ] Issues are executed in topological order derived from `## Blocked by`, not numerical filename order
-- [ ] Issues with `## Blocked by: None` or no `## Blocked by` section have no predecessors in the graph
-- [ ] If a `## Blocked by` reference points to an issue with a higher number than the blocking issue (conflict with numerical order), the runner halts with a descriptive error before executing any issue
-- [ ] The error message names the conflicting issues so the developer can resolve the ordering
-
-## Blocked by
-
-`docs/issues/feature-runner/01-skill-scaffold.md`
diff --git a/docs/issues/feature-runner/07-auto-selection.md b/docs/issues/feature-runner/07-auto-selection.md
deleted file mode 100644
index 91102879..00000000
--- a/docs/issues/feature-runner/07-auto-selection.md
+++ /dev/null
@@ -1,26 +0,0 @@
-# Auto-selection and LOOP_COMPLETE
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Extend the skill to handle no-argument invocation. When invoked without a slug, scan `docs/issues/` for features where every `NN-*.md` file is at `Status: ready-for-agent`. Select the first such feature alphabetically and run it. When no qualifying feature exists, emit `LOOP_COMPLETE` on its own line and exit cleanly. This makes `/loop /implement-feature` composable for overnight queue draining — the `/loop` skill catches `LOOP_COMPLETE` and terminates.
-
-## Acceptance criteria
-
-- [ ] Running `/implement-feature` with no argument scans `docs/issues/` and selects the first feature (alphabetically by slug) where all `NN-*.md` files are `Status: ready-for-agent`
-- [ ] The selected feature is then executed using the same logic as the named-slug path
-- [ ] When no qualifying feature exists, the skill outputs `LOOP_COMPLETE` on its own line and exits cleanly with no error
-- [ ] Running `/loop /implement-feature` terminates cleanly after `LOOP_COMPLETE` is emitted
-- [ ] Features with a mix of `ready-for-agent` and `resolved`/`closed` issues are not selected (partial features are skipped)
-
-## Blocked by
-
-`docs/issues/feature-runner/01-skill-scaffold.md`
diff --git a/docs/issues/feature-runner/08-feature-runner-docs.md b/docs/issues/feature-runner/08-feature-runner-docs.md
deleted file mode 100644
index cd5029a2..00000000
--- a/docs/issues/feature-runner/08-feature-runner-docs.md
+++ /dev/null
@@ -1,38 +0,0 @@
-# `docs/agents/feature-runner.md` reference document
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Write `docs/agents/feature-runner.md` — the agent reference document for the Feature Runner. It must describe the full lifecycle as implemented (not as planned), and document the historical cleanup convention. It must not duplicate content from `docs/agents/issue-tracker.md`, which is overwritten by `setup-matt-pocock-skills`.
-
-The document should cover:
-
-- The Feature Runner lifecycle: feature selected → worktree created → issues implemented in topological order → PR opened → issues marked `closed` on merge
-- The context bundle injected into each `/tdd` invocation (what it contains and why)
-- The `## Blocked by` sequencing rule and what happens on conflict
-- The historical cleanup convention: when a Spec in `docs/plans/` is marked `done` and a corresponding `docs/issues/<slug>/` folder exists, manually mark all issue files `closed` and append a note referencing the Spec
-
-## Acceptance criteria
-
-- [ ] `docs/agents/feature-runner.md` exists
-- [ ] It documents the full Feature Runner lifecycle with all phases named correctly using domain vocabulary from `CONTEXT.md`
-- [ ] It documents the context bundle contents (issue, PRD, siblings, scoped CONTEXT.md, scoped ADRs, recent commits)
-- [ ] It documents the `## Blocked by` sequencing rule and conflict behaviour
-- [ ] It documents the historical cleanup convention with a concrete example of the note to append
-- [ ] It does not duplicate content from `docs/agents/issue-tracker.md`
-
-## Blocked by
-
-- `docs/issues/feature-runner/03-pr-creation.md`
-- `docs/issues/feature-runner/04-progress-reporting.md`
-- `docs/issues/feature-runner/05-full-context-bundle.md`
-- `docs/issues/feature-runner/06-dependency-graph.md`
-- `docs/issues/feature-runner/07-auto-selection.md`
diff --git a/docs/issues/feature-runner/09-retry-on-tdd-failure.md b/docs/issues/feature-runner/09-retry-on-tdd-failure.md
deleted file mode 100644
index 75ef81fc..00000000
--- a/docs/issues/feature-runner/09-retry-on-tdd-failure.md
+++ /dev/null
@@ -1,21 +0,0 @@
-# Retry mechanism on /tdd failure
-
-**Status:** rejected
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## Summary
-
-Add a retry mechanism to `implement-feature` so the runner tries the `/tdd` invocation again before stopping on failure.
-
-## Rejection reasoning
-
-Grilled 2026-05-09. Both failure modes that could trigger a retry are not worth automating:
-
-- **Agent tool errors** (safety classifier, quota exceeded) have unpredictable resolution windows (minutes to hours). An immediate or short-delay retry is a coin flip and provides no reliable value.
-- **Sub-agent gave up** — `/tdd` already has a built-in red-green loop that exhausts retries internally. If it exits with failure, a second invocation hits the same wall. This is a signal for human review, not automation.
-
-The existing "stop + note + leave worktree" behaviour from issue 02 is already the correct response for every failure mode that can occur in practice.
diff --git a/docs/issues/feature-runner/10-references-split.md b/docs/issues/feature-runner/10-references-split.md
deleted file mode 100644
index a03ccffd..00000000
--- a/docs/issues/feature-runner/10-references-split.md
+++ /dev/null
@@ -1,72 +0,0 @@
-# Extract protocol strings to references/runner-output-formats.md
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Create `.claude/skills/implement-feature/references/runner-output-formats.md` containing all verbatim output strings that the runner emits or embeds. Remove those strings from `SKILL.md` and replace each with a reference to the file by name.
-
-The verbatim strings to extract are:
-
-1. **Progress line** — `Implementing issue N of M: <issue title>`
-2. **Dependency conflict error block** — the four-line `Feature Runner error: dependency conflict detected.` message (including the two indented detail lines and the resolution instruction)
-3. **Failure note** — the full `## Comments` markdown block appended to a failing issue file (including the blockquote attribution and the bold `Feature Runner failure —` sentence)
-4. **PR body template** — the heredoc body passed to `gh pr create` (Feature section, Resolved issues list, and the 🤖 attribution line)
-5. **LOOP_COMPLETE signal** — the bare `LOOP_COMPLETE` string with a note that it must appear on its own line
-
-Each entry in `references/runner-output-formats.md` should have a short heading, the exact string (in a fenced code block where multiline), and one sentence explaining when the runner emits it.
-
-`SKILL.md` procedural steps should reference the file by name rather than repeating the strings inline, reducing SKILL.md by approximately 50 lines.
-
-## Acceptance criteria
-
-- [ ] `.claude/skills/implement-feature/references/runner-output-formats.md` exists and contains all five strings listed above
-- [ ] Each entry has a heading, a fenced code block with the exact string, and a one-sentence context note
-- [ ] `SKILL.md` no longer contains the verbatim strings inline; each step references `references/runner-output-formats.md` by name instead
-- [ ] The procedural logic in `SKILL.md` is unchanged — only the literal strings move
-- [ ] `SKILL.md` line count is reduced to approximately 160 lines or fewer
-
-## Blocked by
-
-`docs/issues/feature-runner/01-skill-scaffold.md`
-
-## Comments
-
-> _This was generated by AI during triage._
-
-## Agent Brief
-
-**Category:** enhancement
-**Summary:** Extract verbatim runner output strings from `implement-feature` SKILL.md into a dedicated `references/runner-output-formats.md` file.
-
-**Current behavior:**
-Five verbatim output strings are embedded inline inside the procedural steps of the `implement-feature` skill: the progress line, the dependency conflict error block, the failure note markdown block, the PR body heredoc, and the LOOP_COMPLETE signal. Updating any of these strings requires editing the step prose rather than a named reference file.
-
-**Desired behavior:**
-A `references/runner-output-formats.md` file exists inside the `implement-feature` skill directory alongside `SKILL.md`. It contains all five strings, each under a short heading, in a fenced code block, with one sentence explaining when the runner emits it. The procedural steps in `SKILL.md` reference the file by name instead of repeating the strings inline. Procedural logic is unchanged — only the literal strings move.
-
-**Key interfaces:**
-
-- The `implement-feature` skill directory structure — gains a `references/` subdirectory, consistent with the `new-plugin` and `verify-spec` skills which also use `references/`
-- `references/runner-output-formats.md` — new file; its five sections must exactly match the strings currently in SKILL.md (no editorial changes to the strings themselves)
-
-**Acceptance criteria:**
-
-- [ ] `references/runner-output-formats.md` exists in the `implement-feature` skill directory and contains all five strings listed in `## What to build`
-- [ ] Each entry has a heading, a fenced code block with the exact string, and a one-sentence context note
-- [ ] `SKILL.md` no longer contains the verbatim strings inline; each relevant step references `references/runner-output-formats.md` by name
-- [ ] The procedural logic in `SKILL.md` is unchanged — only the literal strings move
-- [ ] `SKILL.md` line count is reduced to approximately 160 lines or fewer
-
-**Out of scope:**
-
-- Changing the content of any output string (this is a structural move only)
-- Adding new output strings or removing existing ones
-- Updating `docs/agents/feature-runner.md` (it already documents the strings in prose)
diff --git a/docs/issues/feature-runner/11-quick-start.md b/docs/issues/feature-runner/11-quick-start.md
deleted file mode 100644
index 31656b3b..00000000
--- a/docs/issues/feature-runner/11-quick-start.md
+++ /dev/null
@@ -1,67 +0,0 @@
-# Add Quick start section to SKILL.md
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Add a `## Quick start` section near the top of `.claude/skills/implement-feature/SKILL.md`, immediately after the opening paragraph and before `## Steps`. The section should show the three realistic invocation patterns as annotated bullet points, so a user can orient themselves before reading the full step-by-step.
-
-The three patterns:
-
-1. **Named run** — `/implement-feature pr-review-doc-context-enrichment` — targets a specific feature slug directly
-2. **Auto-select** — `/implement-feature` with no argument — scans `docs/issues/` and picks the first fully `ready-for-agent` feature alphabetically
-3. **Overnight loop** — `/loop /implement-feature` — composes with the `/loop` skill to drain the queue unattended; the runner emits `LOOP_COMPLETE` when no qualifying feature remains, which terminates the loop
-
-No separate `EXAMPLES.md` file is needed — `docs/agents/feature-runner.md` already covers lifecycle and mechanics in prose.
-
-## Acceptance criteria
-
-- [ ] A `## Quick start` section exists in `SKILL.md` between the opening paragraph and `## Steps`
-- [ ] The section contains exactly the three invocation patterns listed above, each as a bullet with a concrete example command and a one-line description
-- [ ] No new files are created (no `EXAMPLES.md`)
-- [ ] The rest of `SKILL.md` is unchanged
-
-## Blocked by
-
-`docs/issues/feature-runner/01-skill-scaffold.md`
-
-## Comments
-
-> _This was generated by AI during triage._
-
-## Agent Brief
-
-**Category:** enhancement
-**Summary:** Add a `## Quick start` section to the `implement-feature` skill so users can orient themselves before reading the full 7-step workflow.
-
-**Current behavior:**
-The `implement-feature` SKILL.md opens with a one-paragraph description and jumps directly into `## Steps`. A user invoking the skill for the first time must read all 7 steps to understand the three basic usage patterns.
-
-**Desired behavior:**
-A `## Quick start` section appears between the opening paragraph and `## Steps`. It contains exactly three bullet points — one per invocation pattern — each with a concrete example command and a one-line description of what it does. No new files are created.
-
-**Key interfaces:**
-
-- The `implement-feature` SKILL.md top-level structure: frontmatter → opening paragraph → `## Quick start` → `## Steps`
-- The three invocation patterns (content defined in `## What to build`) must match the behavior already implemented in Step 0 of SKILL.md
-
-**Acceptance criteria:**
-
-- [ ] A `## Quick start` section exists in SKILL.md between the opening paragraph and `## Steps`
-- [ ] The section contains exactly three bullet points covering: named run, auto-select, and overnight loop via `/loop`
-- [ ] Each bullet includes a concrete example command and a one-line description
-- [ ] No new files are created (no `EXAMPLES.md` or other companion files)
-- [ ] All other sections of SKILL.md are unchanged
-
-**Out of scope:**
-
-- Changing any step logic or the 7-step structure
-- Creating an `EXAMPLES.md` file
-- Updating `docs/agents/feature-runner.md`
diff --git a/docs/issues/feature-runner/12-heredoc-note-in-references.md b/docs/issues/feature-runner/12-heredoc-note-in-references.md
deleted file mode 100644
index 2c448ca9..00000000
--- a/docs/issues/feature-runner/12-heredoc-note-in-references.md
+++ /dev/null
@@ -1,74 +0,0 @@
-# Add heredoc wrapping note to PR body template in runner-output-formats.md
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-The PR body template in `references/runner-output-formats.md` shows the multiline body content but gives no instruction on how to pass it to `gh pr create`. Step 7 of SKILL.md currently shows `--body "<PR body template with substitutions>"` as a placeholder, but an agent executing the step has no guidance on quoting a multiline string for the `--body` flag.
-
-Add a "how to use" note directly under the PR body template in `references/runner-output-formats.md` showing the bash heredoc wrapper:
-
-```
-Pass via a bash heredoc:
-
-gh pr create \
-  --base develop \
-  --title "feat(<slug>): <PRD title>" \
-  --body "$(cat <<'EOF'
-<substituted body content>
-EOF
-)"
-```
-
-No changes to SKILL.md are needed — step 7 already directs the agent to `references/runner-output-formats.md` for the body template.
-
-## Acceptance criteria
-
-- [ ] The PR body template section in `references/runner-output-formats.md` includes a note explaining that the body must be passed via a bash heredoc
-- [ ] The note shows a complete, correct `gh pr create` command with the heredoc wrapper
-- [ ] SKILL.md step 7 is unchanged
-- [ ] No other sections of `references/runner-output-formats.md` are modified
-
-## Blocked by
-
-`docs/issues/feature-runner/10-references-split.md`
-
-## Comments
-
-> _This was generated by AI during triage._
-
-## Agent Brief
-
-**Category:** enhancement
-**Summary:** Add heredoc wrapping instruction to the PR body template in `references/runner-output-formats.md` so agents know how to pass multiline content to `gh pr create`.
-
-**Current behavior:**
-The PR body template section shows the body content in a fenced code block but gives no instruction on how to embed it in a `gh pr create` call. Step 7 of SKILL.md shows `--body "<PR body template with substitutions>"` as a placeholder — an agent executing this step has no guidance on quoting a multiline string for `--body`.
-
-**Desired behavior:**
-The PR body template section in `references/runner-output-formats.md` includes a follow-on note showing the complete `gh pr create` command with the body content wrapped in a bash heredoc (`--body "$(cat <<'EOF' ... EOF)"`). An agent reading the section finds both the template content and the exact quoting pattern needed to use it.
-
-**Key interfaces:**
-
-- The PR body template section in `references/runner-output-formats.md` — gains a "how to use" note after the fenced code block
-- SKILL.md step 7 — unchanged; it already directs the agent to the references file
-
-**Acceptance criteria:**
-
-- [ ] The PR body template section in `references/runner-output-formats.md` includes a note explaining that the body must be passed via a bash heredoc
-- [ ] The note shows a complete, correct `gh pr create` command with the heredoc wrapper
-- [ ] SKILL.md step 7 is unchanged
-- [ ] No other sections of `references/runner-output-formats.md` are modified
-
-**Out of scope:**
-
-- Changing the PR body template content itself
-- Adding cross-platform alternatives to the heredoc (the cross-platform concern pre-dates this issue and is not being addressed here)
-- Modifying SKILL.md
diff --git a/docs/issues/feature-runner/13-tdd-prompt-template-in-references.md b/docs/issues/feature-runner/13-tdd-prompt-template-in-references.md
deleted file mode 100644
index 34a59187..00000000
--- a/docs/issues/feature-runner/13-tdd-prompt-template-in-references.md
+++ /dev/null
@@ -1,91 +0,0 @@
-# Extract /tdd prompt template to references/tdd-prompt-template.md
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-The 23-line `/tdd` prompt template in step 4 of SKILL.md is the last large inline block. It is a structured template (not a verbatim output string) that the agent fills in dynamically at runtime, but its presence mid-step interrupts the prose flow and keeps SKILL.md at 177 lines — above the ≤160 acceptance criterion set in issue 10.
-
-Create `references/tdd-prompt-template.md` containing the full prompt template. Update step 4 of SKILL.md to reference the file by name instead of repeating the template inline.
-
-The template to extract (currently in step 4 of SKILL.md, inside a fenced code block):
-
-```
-You are running /tdd in AFK mode. The planning phase is complete — do not ask for confirmation. Use the acceptance criteria below as the pre-approved plan and proceed directly to the red→green→refactor loop.
-
-Working directory: .claude/worktrees/<slug>
-
---- ISSUE ---
-<full content of the current issue file>
-
---- PRD (parent context) ---
-<full content of docs/issues/<slug>/PRD.md>
-
---- SIBLING ISSUES ---
-<full content of each sibling issue file, separated by the filename as a header>
-
---- CONTEXT.md ---
-<full content of the scoped CONTEXT.md>
-
---- ADRs ---
-<full content of each scoped ADR file, separated by the filename as a header>
-
---- RECENT COMMITS (last 5) ---
-<output of git log --oneline -5>
-```
-
-The new file should have a short heading, the template in a fenced code block, and a one-sentence note explaining that placeholders are filled at runtime.
-
-Step 4 in SKILL.md should replace the fenced block with: "Construct the prompt using the template in `references/tdd-prompt-template.md`, substituting all `<placeholder>` values at runtime."
-
-## Acceptance criteria
-
-- [ ] `references/tdd-prompt-template.md` exists in the `implement-feature` skill directory and contains the full prompt template
-- [ ] The template in the new file is identical to the template currently in SKILL.md step 4 (no content changes)
-- [ ] Step 4 of SKILL.md no longer contains the inline fenced block; it references `references/tdd-prompt-template.md` by name
-- [ ] SKILL.md line count drops to approximately 154 lines or fewer
-- [ ] No other steps in SKILL.md are modified
-
-## Blocked by
-
-`docs/issues/feature-runner/10-references-split.md`
-
-## Comments
-
-> _This was generated by AI during triage._
-
-## Agent Brief
-
-**Category:** enhancement
-**Summary:** Extract the 23-line `/tdd` prompt template from SKILL.md step 4 into `references/tdd-prompt-template.md` to bring SKILL.md under the ≤160 line target.
-
-**Current behavior:**
-Step 4 of the `implement-feature` skill contains a 23-line fenced code block with the `/tdd` prompt template inline. SKILL.md is 177 lines — above the ≤160 target set in issue 10. The template is the last remaining large inline block.
-
-**Desired behavior:**
-`references/tdd-prompt-template.md` exists alongside `references/runner-output-formats.md` in the `implement-feature` skill directory. It contains the prompt template (with a heading, fenced code block, and one-sentence note about runtime substitution). Step 4 of SKILL.md replaces the inline block with a single sentence referencing the file by name. SKILL.md drops to approximately 154 lines.
-
-**Key interfaces:**
-
-- `references/tdd-prompt-template.md` — new file; template content must be identical to what is currently inline in SKILL.md step 4
-- SKILL.md step 4 — loses the inline fenced block, gains a one-sentence reference to the new file; all other prose in step 4 is unchanged
-
-**Acceptance criteria:**
-
-- [ ] `references/tdd-prompt-template.md` exists and contains the full prompt template, identical to the current inline version
-- [ ] Step 4 of SKILL.md references `references/tdd-prompt-template.md` by name instead of repeating the template
-- [ ] SKILL.md line count drops to approximately 154 lines or fewer
-- [ ] No other steps in SKILL.md are modified
-
-**Out of scope:**
-
-- Changing the prompt template content (structural move only)
-- Modifying `references/runner-output-formats.md`
-- Updating `docs/agents/feature-runner.md` or ADR 0029
diff --git a/docs/issues/feature-runner/14-smarter-auto-select.md b/docs/issues/feature-runner/14-smarter-auto-select.md
deleted file mode 100644
index 99069b22..00000000
--- a/docs/issues/feature-runner/14-smarter-auto-select.md
+++ /dev/null
@@ -1,62 +0,0 @@
-# Smarter auto-select: resume partial features and reuse existing worktrees
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Relax the auto-selection qualification rule so that partially-completed features (e.g. after a failure mid-run) are included alongside fresh ones. Update the worktree creation step to reuse an existing worktree rather than failing or recreating it.
-
-### Auto-select qualification (Step 0)
-
-**Current rule:** a feature qualifies only if every `NN-*.md` file is exactly `ready-for-agent`.
-
-**New rule:** a feature qualifies if:
-
-- Every `NN-*.md` file has a status in `{ready-for-agent, resolved, closed, rejected, ready-for-human}`
-- At least one `NN-*.md` file has status `ready-for-agent`
-
-Any issue in `{needs-triage, needs-info, needs-specs}` (or any unrecognised state) disqualifies the whole feature — it is not fully prepped for autonomous execution.
-
-Alphabetical slug ordering is unchanged.
-
-### Worktree creation (Step 2)
-
-Before running `git worktree add`, check whether `.claude/worktrees/<slug>` already exists:
-
-```
-ls .claude/worktrees/<slug>
-```
-
-- **Exists** (prior failed run) → skip `git worktree add` and reuse the existing worktree. The branch `feature/afk/<slug>` already contains the committed work from that run.
-- **Does not exist** → create it as before:
-
-```
-git worktree add .claude/worktrees/<slug> -b feature/afk/<slug> develop
-```
-
-### Quick start note
-
-Add a note to the Quick start section explaining that a failed partial feature requires named invocation (`/implement-feature <slug>`) only if the loop is not running — if the loop is running, the new auto-select rule will pick it up automatically once the developer fixes the failing issue and leaves it at `ready-for-agent`.
-
-## Acceptance criteria
-
-- [ ] Auto-select picks up a feature where some issues are `resolved` and at least one is `ready-for-agent`
-- [ ] Auto-select skips a feature that has any issue in `needs-triage`, `needs-info`, or `needs-specs`
-- [ ] Auto-select still skips a feature where every issue is `resolved`, `closed`, or `rejected` (nothing left to run)
-- [ ] Issues with status `ready-for-human` or `rejected` do not disqualify a feature
-- [ ] When `.claude/worktrees/<slug>` already exists, the runner reuses it and does not call `git worktree add`
-- [ ] When `.claude/worktrees/<slug>` does not exist, the runner creates it as before
-- [ ] The execution queue (Step 3) continues to filter to `ready-for-agent` only — `resolved`, `closed`, `rejected`, and `ready-for-human` issues act only as satisfied dependency nodes
-- [ ] `docs/agents/feature-runner.md` — Feature selection section updated to reflect the new qualification rule
-- [ ] SKILL.md Quick start updated to note loop auto-resume behaviour after failure fix
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/feature-runner/15-ready-for-human-unsatisfied-dependency.md b/docs/issues/feature-runner/15-ready-for-human-unsatisfied-dependency.md
deleted file mode 100644
index 97dbba96..00000000
--- a/docs/issues/feature-runner/15-ready-for-human-unsatisfied-dependency.md
+++ /dev/null
@@ -1,71 +0,0 @@
-# Halt when a ready-for-agent issue depends on a ready-for-human blocker
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Extend the step 3 dependency validation so that `ready-for-human` blockers are treated as **unsatisfied** dependencies. If a `ready-for-agent` issue in the execution queue has a `## Blocked by` dependency whose status is `ready-for-human`, the runner must halt before executing anything.
-
-Also clarify the "satisfied" language in step 3 and `docs/agents/feature-runner.md` to be explicit about which statuses satisfy dependencies and which do not.
-
-### Dependency satisfaction model
-
-| Status            | Satisfies dependents? | Executed?                                                 |
-| ----------------- | --------------------- | --------------------------------------------------------- |
-| `resolved`        | Yes                   | No                                                        |
-| `closed`          | Yes                   | No                                                        |
-| `rejected`        | Yes                   | No — rejection is a terminal decision; dependents proceed |
-| `ready-for-human` | **No**                | No — human work may not be done; dependents are blocked   |
-| `ready-for-agent` | —                     | Yes                                                       |
-
-### New validation check (Step 3)
-
-After building the topological execution order, before executing anything, add a second check:
-
-For each `ready-for-agent` issue in the execution queue, inspect its `## Blocked by` list. If any listed blocker has status `ready-for-human`, halt immediately with the **unsatisfied dependency error** (see `references/runner-output-formats.md`).
-
-### New error format (`references/runner-output-formats.md`)
-
-Add an **unsatisfied dependency error** section:
-
-```
-Feature Runner error: unsatisfied dependency.
-  Issue NN-<A> is blocked by NN-<B>, but NN-<B> has status ready-for-human.
-  Complete issue NN-<B> manually (or update its status) before re-running /implement-feature <slug>.
-```
-
-### Language fix (Step 3 and `docs/agents/feature-runner.md`)
-
-Replace:
-
-> "`resolved` and `closed` issues are already satisfied and act only as satisfied dependencies, not as items to execute."
-
-With:
-
-> "`resolved`, `closed`, and `rejected` issues are satisfied and act as satisfied dependency nodes — not items to execute. `ready-for-human` issues are not satisfied: if a `ready-for-agent` issue depends on one, the runner halts before executing anything."
-
-## Rationale
-
-`ready-for-human` means the work still needs to be done — just not by the runner. If a `ready-for-agent` issue depends on that work, executing it on an incomplete foundation risks a correct-but-wrong implementation.
-
-`rejected` is treated as satisfied because it is a terminal, intentional decision: the team consciously ruled the work out. A dependent issue proceeding without a rejected blocker is exactly what triage intended. (Typical scenario: team rejects issue 01 after initial triage; issue 02 which was blocked by 01 should still run.)
-
-## Acceptance criteria
-
-- [ ] Step 3 of SKILL.md checks each `ready-for-agent` issue's blockers for `ready-for-human` status before executing anything
-- [ ] If any such blocker is found, the runner halts with the unsatisfied dependency error before executing any issue
-- [ ] `references/runner-output-formats.md` contains the new unsatisfied dependency error format
-- [ ] Step 3 "satisfied" language lists `resolved`, `closed`, and `rejected` as satisfied; notes `ready-for-human` as unsatisfied
-- [ ] `docs/agents/feature-runner.md` updated to reflect the same model
-- [ ] A `rejected` blocker does not halt the runner (treated as satisfied)
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/feature-runner/16-explicit-tdd-skill-invocation.md b/docs/issues/feature-runner/16-explicit-tdd-skill-invocation.md
deleted file mode 100644
index ccc58462..00000000
--- a/docs/issues/feature-runner/16-explicit-tdd-skill-invocation.md
+++ /dev/null
@@ -1,49 +0,0 @@
-# Explicit `/tdd` skill invocation and pinned `subagent_type` in step 4
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-The Feature Runner's step 4 invokes `/tdd` via the Agent tool but does not pin `subagent_type` and does not instruct the sub-agent to explicitly load the `/tdd` skill. The current prompt template frames the sub-agent as "running /tdd in AFK mode" and relies on trigger-phrase auto-discovery to load `/tdd`'s procedural guidance — an unreliable mechanism.
-
-ADR-0029 ratifies replacing `/tdd`'s _planning_ phase with acceptance criteria. The rest of the skill (anti-horizontal-slicing, tracer-bullet loop, refactor-only-when-green, deep-modules / interface-design / refactoring sub-references) is still load-bearing and must be loaded deterministically.
-
-### Changes
-
-**1. `SKILL.md` step 4** — pin the subagent type.
-
-Replace the current "invoke `/tdd` as a non-interactive sub-agent using the Agent tool" sentence with one that names the subagent type explicitly:
-
-> Invoke the sub-agent using the Agent tool with `subagent_type: general-purpose` — the only stock type with access to both the `Skill` tool (to load `/tdd`) and `Edit`/`Write` tools (to write code).
-
-**2. `references/tdd-prompt-template.md`** — prepend an explicit skill invocation instruction.
-
-At the top of the prompt body (before the AFK framing), add:
-
-```
-Begin by invoking the `tdd` skill via the Skill tool to load its full procedural guidance (red→green→refactor, vertical-slice rule, deep-modules / interface-design / refactoring sub-references). Then follow it using the acceptance criteria below as the pre-approved plan — the planning phase is complete; do not ask for confirmation.
-```
-
-**3. `docs/adr/0029-feature-runner-afk-invocation.md`** — amend the Consequences section.
-
-Append:
-
-> The Feature Runner's prompt template explicitly instructs the sub-agent to invoke the `tdd` skill via the Skill tool, rather than relying on auto-discovery via trigger phrases. This makes the procedural-guidance load deterministic across Claude Code surfaces.
-
-## Acceptance criteria
-
-- [ ] `SKILL.md` step 4 names `subagent_type: general-purpose` for the `Agent` call
-- [ ] `references/tdd-prompt-template.md` opens with an explicit Skill-tool invocation instruction for `tdd`, placed before the AFK framing
-- [ ] `docs/adr/0029-feature-runner-afk-invocation.md` Consequences section reflects the explicit-invocation policy
-- [ ] The existing AFK framing ("The planning phase is complete — do not ask for confirmation") is preserved after the new instruction
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/feature-runner/17-prd-title-extraction-step-1.md b/docs/issues/feature-runner/17-prd-title-extraction-step-1.md
deleted file mode 100644
index c027c7c2..00000000
--- a/docs/issues/feature-runner/17-prd-title-extraction-step-1.md
+++ /dev/null
@@ -1,32 +0,0 @@
-# Extract PRD `title:` frontmatter in step 1
-
-**Status:** closed
-**Category:** bug
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Step 7 of SKILL.md says: _"Derive the PR title from the PRD's `title` frontmatter field (already read in step 1)"_. But step 1 only says to read the PRD and scan it for `apps/claude-code/<plugin>/` path references. It never says to extract the `title:` value from the YAML frontmatter.
-
-A re-implementer or a sub-agent following step 1 verbatim would complete the step without capturing the PRD title, then encounter the step 7 parenthetical with no value to substitute.
-
-### Change
-
-In `SKILL.md` step 1, under "**Read the PRD:**", add one explicit instruction after the plugin-path scan:
-
-> Also extract the `title:` field from the PRD's YAML frontmatter (the value between the opening `---` and closing `---` at the top of the file). Retain it for use in step 7's PR title derivation.
-
-## Acceptance criteria
-
-- [ ] Step 1 of SKILL.md explicitly instructs the runner to extract the PRD's `title:` YAML frontmatter value
-- [ ] Step 7's parenthetical "(already read in step 1)" remains accurate — the PRD title is gathered in step 1, not on-demand in step 7
-- [ ] No other changes to SKILL.md
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/feature-runner/18-failure-loop-protection.md b/docs/issues/feature-runner/18-failure-loop-protection.md
deleted file mode 100644
index 6636297f..00000000
--- a/docs/issues/feature-runner/18-failure-loop-protection.md
+++ /dev/null
@@ -1,73 +0,0 @@
-# Protect `/loop` from re-picking a failed Feature (flip status to `needs-info`)
-
-**Status:** closed
-**Category:** enhancement
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-When `/tdd` fails on an issue, the Feature Runner appends a failure note and stops — but leaves the failing issue at `ready-for-agent`. Under `/loop /implement-feature`, auto-select then re-picks the same Feature on the next iteration (because it still has a `ready-for-agent` issue) and retries the same failing issue, burning loop iterations indefinitely.
-
-The fix: on `/tdd` failure, flip the failing issue's `**Status:**` line from `ready-for-agent` to `needs-info`. `needs-info` is an existing triage label ("waiting on reporter for more information") and is already in the auto-select disqualification list at step 0 — no vocabulary expansion is required.
-
-### Changes
-
-**1. `SKILL.md` step 4 "On failure"**
-
-After appending the failure note, add a second edit:
-
-> Using the Edit tool, change the `**Status:** ready-for-agent` line to `**Status:** needs-info`. This prevents auto-select from picking up this Feature on subsequent loop iterations.
-
-**2. `references/runner-output-formats.md` — failure note**
-
-Update the failure note to mention the status flip and the recovery procedure:
-
-```markdown
-## Comments
-
-> _This was generated by AI during triage._
-
-**Feature Runner failure** — `/tdd` could not complete this issue. Status has been set to `needs-info`.
-
-The worktree at `.claude/worktrees/<slug>` has been left in place for inspection. Once the issue is resolved manually, restore `**Status:** ready-for-agent` and re-run `/implement-feature <slug>` to resume. Alternatively, close or reject the issue if it should not be retried.
-```
-
-**3. `docs/agents/feature-runner.md` — "Failure behaviour" section**
-
-Update step 2 of the numbered list:
-
-Replace:
-
-> The issue remains at `Status: ready-for-agent`.
-
-With:
-
-> The issue status is changed to `needs-info`. This prevents the auto-selection path from picking up this Feature on subsequent `/loop` iterations until the developer investigates and restores `ready-for-agent` (or closes/rejects the issue).
-
-**4. `SKILL.md` Quick start — "Safe to interrupt" bullet**
-
-Distinguish Ctrl+C from `/tdd` failure explicitly. Update the bullet:
-
-> **Safe to interrupt** — Ctrl+C during any issue leaves that issue at `ready-for-agent`; re-running resumes from the first unresolved issue. Note: a **/tdd failure** (as opposed to a Ctrl+C interrupt) sets the failing issue to `needs-info`, preventing auto-select from retrying until the developer intervenes.
-
-## Rationale
-
-The overnight-loop use case is the primary composition for the Feature Runner. An auto-select that retries the same broken Feature undermines the purpose of `/loop`. `needs-info` already signals "blocked on a human" in the triage vocabulary — exactly the right state for a Feature Runner failure.
-
-## Acceptance criteria
-
-- [ ] On `/tdd` failure, the failing issue's `**Status:**` is changed to `needs-info` (not left at `ready-for-agent`)
-- [ ] Subsequent auto-select runs skip the Feature (because `needs-info` already disqualifies a Feature in step 0)
-- [ ] The failure note in `references/runner-output-formats.md` documents the `needs-info` flip and the recovery procedure (restore `ready-for-agent` and re-run, or close/reject)
-- [ ] `docs/agents/feature-runner.md` "Failure behaviour" section reflects the status flip to `needs-info`
-- [ ] SKILL.md Quick start distinguishes Ctrl+C (leaves `ready-for-agent`) from `/tdd` failure (sets `needs-info`)
-- [ ] Manual interrupt (`Ctrl+C`) behaviour is unchanged: issue stays `ready-for-agent`
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/feature-runner/19-skill-agents-doc-crosslink.md b/docs/issues/feature-runner/19-skill-agents-doc-crosslink.md
deleted file mode 100644
index 0597a546..00000000
--- a/docs/issues/feature-runner/19-skill-agents-doc-crosslink.md
+++ /dev/null
@@ -1,39 +0,0 @@
-# Add cross-link from SKILL.md to `docs/agents/feature-runner.md`
-
-**Status:** closed
-**Category:** documentation
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-`SKILL.md` and `docs/agents/feature-runner.md` serve distinct audiences and should remain separate:
-
-- `SKILL.md` — runtime guidance for the orchestrating Claude instance.
-- `docs/agents/feature-runner.md` — human-facing reference: lifecycle diagram, dependency-satisfaction matrix, ADR scope, historical cleanup convention.
-
-Currently there is no link from SKILL.md to the agents doc. A human reading the skill (or a developer extending it) has no signal that a fuller reference exists.
-
-### Change
-
-In `SKILL.md` Supporting Documentation section, add one line pointing to the agents doc:
-
-```markdown
-- **`docs/agents/feature-runner.md`** — human-facing reference: lifecycle diagram, dependency-satisfaction matrix, ADR scope, historical cleanup convention. Maintained in parallel; consult for broader context, not for runtime instructions.
-```
-
-No prose is removed from either file.
-
-## Acceptance criteria
-
-- [ ] `SKILL.md` Supporting Documentation section includes a pointer to `docs/agents/feature-runner.md` with a one-line description
-- [ ] No prose is removed from SKILL.md or `docs/agents/feature-runner.md`
-- [ ] SKILL.md word count remains under 2,000 words after the addition
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/feature-runner/20-prompt-template-and-step-cleanup.md b/docs/issues/feature-runner/20-prompt-template-and-step-cleanup.md
deleted file mode 100644
index 127c0e3e..00000000
--- a/docs/issues/feature-runner/20-prompt-template-and-step-cleanup.md
+++ /dev/null
@@ -1,60 +0,0 @@
-# Prompt-template deduplication and step-1/step-4 wording cleanup
-
-**Status:** closed
-**Category:** documentation
-
-> _This was generated by AI during triage._
-
-## Parent
-
-`docs/issues/feature-runner/PRD.md`
-
-## What to build
-
-Three small cleanups surfaced in the second-pass review of issues 16–19.
-
-### 20.a — Deduplicate prompt-template framing
-
-`references/tdd-prompt-template.md` currently opens with two paragraphs that both say "the planning phase is complete; do not ask for confirmation" and "use the acceptance criteria as the pre-approved plan." This is a side-effect of issue 16 prepending the explicit `tdd` Skill-tool-load instruction while preserving the original AFK framing verbatim. Every `/tdd` sub-agent invocation receives the duplicate.
-
-Replace both paragraphs with one consolidated paragraph that retains all unique signal: AFK identity, planning-phase-complete, explicit `tdd` Skill tool load, acceptance-criteria-as-plan, red→green→refactor loop entry.
-
-**Replacement** (everything before the `Working directory:` line):
-
-```
-You are running `/tdd` in AFK mode. The interactive planning phase is complete — do not ask for confirmation. Begin by invoking the `tdd` skill via the Skill tool to load its full procedural guidance (red→green→refactor, vertical-slice rule, deep-modules / interface-design / refactoring sub-references), then follow it using the acceptance criteria below as the pre-approved plan and proceed directly to the red→green→refactor loop.
-```
-
-### 20.b — Tighten step 1 closing line
-
-`SKILL.md` step 1 closing line (currently "These four items (PRD, CONTEXT.md, ADRs, recent commits) are static") was not updated when issue 17 added the PRD `title:` frontmatter parse. The PRD entry should reflect that both content and title are captured.
-
-**Replacement:**
-
-> These items (PRD content + title, CONTEXT.md, ADRs, recent commits) are static — gather them once before the issue loop begins.
-
-### 20.c — Reorder step 4 "On failure" sub-steps
-
-`SKILL.md` step 4 "On failure" currently appends the failure note (whose text says "Status has been set to `needs-info`") _before_ flipping the status to `needs-info`. This briefly leaves the file self-contradicting: the note says the status is `needs-info` while the status line still reads `ready-for-agent`.
-
-Swap the order so the status flip happens first:
-
-1. Using the Edit tool, change the `**Status:** ready-for-agent` line to `**Status:** needs-info`. This prevents auto-select from picking up this Feature on subsequent loop iterations.
-2. Append the **failure note** (see `references/runner-output-formats.md`) to the issue file using the Edit tool, substituting `<slug>`.
-3. Stop the runner immediately. Do not execute any subsequent issues — they may depend on a foundation this issue was meant to lay.
-4. Report to the user: which issue failed, that the worktree is at `.claude/worktrees/<slug>` on branch `feature/afk/<slug>`, and that no subsequent issues were run.
-
-`docs/agents/feature-runner.md` "Failure behaviour" should reflect the same numbered order (status flip is step 2 in the list, failure note is step 1 currently — update to match).
-
-## Acceptance criteria
-
-- [ ] `references/tdd-prompt-template.md` contains exactly one paragraph framing the AFK invocation, before the `Working directory:` line
-- [ ] The single paragraph mentions: AFK mode, planning-phase-complete, `tdd` Skill tool load via Skill tool, acceptance criteria as pre-approved plan, red→green→refactor loop
-- [ ] No information from the previous two-paragraph version is lost
-- [ ] `SKILL.md` step 1 closing line names "PRD content + title" (or equivalent) alongside CONTEXT.md, ADRs, and recent commits
-- [ ] `SKILL.md` step 4 "On failure" performs the status flip (sub-step 1) before appending the failure note (sub-step 2)
-- [ ] `docs/agents/feature-runner.md` "Failure behaviour" numbered list reflects the same ordering (status flip before note)
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/inbox-collision-check/01-fix-exit-code.md b/docs/issues/inbox-collision-check/01-fix-exit-code.md
deleted file mode 100644
index 7c76305d..00000000
--- a/docs/issues/inbox-collision-check/01-fix-exit-code.md
+++ /dev/null
@@ -1,42 +0,0 @@
-# Fix inverted exit code in `/inbox` collision check
-
-**Status:** closed
-**Category:** bug
-
-> _This was generated by AI during triage._
-
-## Agent Brief
-
-**Category:** bug
-**Summary:** The `/inbox` command's collision-check script exits with code 1 on the happy path (no collision), causing Claude Code's Bash tool to display a false "Error: Exit code 1" on every successful capture.
-
-**Current behavior:**
-The collision check in `.claude/commands/inbox.md` uses:
-
-```js
-process.exit(require('fs').existsExists(p) ? 0 : 1)
-```
-
-Exit 0 (tool-success) means "file exists" (collision). Exit 1 (tool-error) means "file does not exist" (no collision). Since the no-collision path is the common case, every normal inbox capture shows a red "Error: Exit code 1" in the Bash tool output, even though the file is created successfully.
-
-**Desired behavior:**
-The no-collision path should exit 0 (tool-success) and the collision path should exit 1 (tool-error). This way the Bash tool shows no error on a normal capture, and Claude correctly detects a collision (non-zero exit) and appends `-2`, `-3`, etc.
-
-**Key interfaces:**
-
-- The collision-check `node -e` snippet in `.claude/commands/inbox.md` Step 2 — swap the exit code values
-
-**Acceptance criteria:**
-
-- [ ] Running `/inbox <new-idea>` produces no "Error: Exit code 1" in tool output
-- [ ] Running `/inbox <existing-slug>` still detects the collision and appends `-2` to the slug
-- [ ] The same fix is applied to the copy at the installed path (`.claude/plugins/marketplaces/unic-agent-plugins/.claude/commands/inbox.md`) if it is managed in this repo
-
-**Out of scope:**
-
-- Rewriting the collision-check with a different mechanism (fs.existsSync is fine)
-- Changes to other parts of the inbox skill
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/plugin-claude-md-monorepo-refresh/PRD.md b/docs/issues/plugin-claude-md-monorepo-refresh/PRD.md
new file mode 100644
index 00000000..d108db91
--- /dev/null
+++ b/docs/issues/plugin-claude-md-monorepo-refresh/PRD.md
@@ -0,0 +1,165 @@
+---
+title: Refresh per-plugin CLAUDE.md / AGENTS.md for monorepo context
+created: 2026-05-23
+---
+
+**Status:** ready-for-agent
+**Plugin:** all four Claude Code Plugins (`apps/claude-code/{auto-format,pr-review,unic-confluence,unic-archon-dlc}`)
+**Category:** docs
+
+## Problem Statement
+
+The per-plugin agent-guidance files in this monorepo predate the migration and still speak as if each plugin lives in its own standalone repo.
+
+- `apps/claude-code/auto-format/CLAUDE.md`, `apps/claude-code/pr-review/CLAUDE.md`, and `apps/claude-code/unic-confluence/CLAUDE.md` all open with "This file provides guidance to Claude Code when working in this repository" and proceed to duplicate content that already lives in the root `AGENTS.md`: pnpm scripts, SemVer policy, Conventional Commits scope, code conventions, tag scheme, Gitflow rules.
+- `unic-confluence/CLAUDE.md` carries an obsolete naming-convention table (`unic-claude-code-<service>`) that referred to the pre-migration repo layout.
+- `auto-format/CLAUDE.md` calls the plugin by its retired identifier (`unic-claude-code-format`).
+- `pr-review/CLAUDE.md` references `commands/review-pr.md` and a "spec 02" that the file does not link out to or pin down.
+- `apps/claude-code/unic-archon-dlc/` has no agent-guidance file at all. It has `CONTEXT.md` (domain vocabulary), but a contributor opening that directory has nothing equivalent to the other plugins' files.
+
+The root `AGENTS.md`/`CLAUDE.md` symlink convention is not mirrored at the plugin level: each plugin has a single `CLAUDE.md` file. This works for Claude Code but does not advertise the same cross-agent compatibility the root uses.
+
+The backlog item in `docs/plans/README.md` — "Update plugin `CLAUDE.md` files for monorepo context" — was deferred during the plugin migration Specs 05–07 and never picked back up.
+
+## Solution
+
+Each Plugin gets an `AGENTS.md` written in monorepo voice, with a `CLAUDE.md` symlink alongside (mirroring the root pattern). The file contains only what is _specific_ to that Plugin; everything generic links to the root `AGENTS.md`. The four files share a common section structure so a reader who has seen one knows where to look in the others.
+
+`unic-archon-dlc` gets its first `AGENTS.md` in the same style. Its `CONTEXT.md` continues to own the domain vocabulary and is referenced from the new file, not duplicated.
+
+## User Stories
+
+1. As a new contributor opening `apps/claude-code/pr-review/AGENTS.md`, I want a one-paragraph statement of what the Plugin does so that I do not have to read source code to orient myself.
+2. As a new contributor opening any plugin's `AGENTS.md`, I want it to link out to the root `AGENTS.md` for cross-cutting conventions, so that I do not get conflicting guidance from the two files.
+3. As a new contributor reading a plugin file, I want generic monorepo rules (pnpm scripts, Gitflow, SemVer policy, code conventions) to live in one place — the root — so that drift between files cannot create contradictions.
+4. As a maintainer renaming a plugin command or changing the bot signature, I want the canonical wording to live in the plugin's own `AGENTS.md`, so that I do not have to remember to update the root file too.
+5. As a contributor working on `pr-review`, I want the bot signature format and the soft-dependency on `pr-review-toolkit` to be visible in the plugin file, so that I do not break observability or runtime preconditions accidentally.
+6. As a contributor working on `auto-format`, I want the "always exits zero" doctrine and the "consumer owns Formatters" rule to be prominent in the plugin file, so that I do not accidentally introduce a blocking Format Hook or bundled Formatter.
+7. As a contributor working on `unic-confluence`, I want the marker-injection priority order and the long "do not add" list to remain in the plugin file, so that I do not propose features that have already been decided against.
+8. As a contributor working on `unic-archon-dlc`, I want the file to state the Setup idempotence contract and the dogfooding note explicitly, so that I understand both how the Plugin behaves in a Consumer repo and that this monorepo is also a Consumer.
+9. As a contributor opening any plugin's directory, I want a pointer to that plugin's own `docs/adr/` so that I can find the historical design decisions specific to that Plugin without reading the root `docs/adr/`.
+10. As an AI agent operating with the AGENTS.md convention, I want every Plugin directory to expose an `AGENTS.md` file so that I can discover guidance without falling back to Claude-specific filenames.
+11. As an AI agent operating in Claude Code, I want a `CLAUDE.md` to still be present in every Plugin directory so that the existing tooling continues to find it.
+12. As a maintainer reviewing a diff, I want all four plugin `AGENTS.md` files to share the same top-level section structure, so that I can compare them at a glance.
+13. As a maintainer auditing for accuracy, I want references to obsolete identifiers (`unic-claude-code-format`, `unic-claude-code-<service>`, the retired `spec-NN` commit scope) removed from every plugin file, so that the documentation reflects current state.
+14. As a contributor reading the auto-format file, I want the standalone-repo "Project Overview" framing replaced with a plugin-inside-workspace framing, so that I do not misunderstand the surrounding context.
+15. As a contributor reading the unic-confluence file, I want the "no `pnpm run` wrapper is defined" note revisited or removed, so that the file does not contradict the monorepo's actual `pnpm bump` / `pnpm verify:changelog` script entries.
+16. As a maintainer, I want each plugin file to point to the same canonical set of root docs (root `AGENTS.md`, root `CONTEXT.md`, root `docs/process/`, root `docs/adr/`, root `CONTEXT-MAP.md`) so that the linking surface is predictable.
+17. As an agent or contributor following a link from a plugin file to root docs, I want those links to be relative paths that work from the plugin directory, so that they resolve correctly in editors and on GitHub.
+18. As a maintainer, I want a brief acknowledgement in each file that the per-plugin `docs/plans/` directory is historical (no longer the intake path) so that contributors do not treat retired specs as live work.
+
+## Implementation Decisions
+
+### File naming and convention
+
+Each Plugin directory gets two files at its root:
+
+- `AGENTS.md` — the canonical agent guidance file.
+- `CLAUDE.md` — a relative symlink pointing to `AGENTS.md`, matching the root convention (`CLAUDE.md -> AGENTS.md`).
+
+The existing `CLAUDE.md` regular file is replaced by the symlink in each affected Plugin. The new `AGENTS.md` carries the refreshed content. The symlink direction must be `CLAUDE.md → AGENTS.md` (not the reverse), so the canonical file name in `git log` and in the file tree is `AGENTS.md`.
+
+### Uniform section template (all four files)
+
+Every plugin `AGENTS.md` uses the same top-level structure. Sections may be omitted when not applicable (for example, a Plugin with no external runtime dependencies omits that section).
+
+1. **What this Plugin is** — one paragraph. States the Plugin's role inside the monorepo and points to its `CONTEXT.md` for vocabulary.
+2. **Where to start** — link list to root `AGENTS.md`, root `CONTEXT.md`, root `CONTEXT-MAP.md`, root `docs/adr/`, root `docs/process/`, and to this Plugin's own `docs/adr/`. Notes that root `AGENTS.md` is the source of truth for cross-cutting conventions (pnpm, Gitflow, SemVer, Conventional Commits, code conventions, LICENSE policy, cross-platform requirement).
+3. **Commands** — only the pnpm scripts unique to this Plugin (typically `bump`, `sync-version`, `tag`, `verify:changelog`, plugin-specific `test` / `typecheck`). Root-wide commands are not duplicated.
+4. **Layout** — short tree of the directories that exist in this Plugin (`.claude-plugin/`, `commands/`, `agents/`, `hooks/`, `lib/`, `scripts/`, `tests/`, `test/`, `docs/`). One line per directory describing its role.
+5. **Plugin doctrines** — the load-bearing invariants that are not obvious from the code or the root rules. These are the contents most worth preserving from the existing files.
+6. **External dependencies** — runtime tools or Consumer-side packages this Plugin assumes (e.g. `az devops` CLI for `pr-review`; the Archon runtime for `unic-archon-dlc`).
+7. **Do not add** — Plugin-specific scope guard, listing things that have been considered and explicitly excluded.
+8. **Plugin ADRs** — single line linking to `docs/adr/` for this Plugin.
+
+A sentence at the end of section 8 acknowledges that `docs/plans/` (per-plugin) is historical and not the intake path for new work; pointers to the issue tracker and root `docs/process/` cover the live workflow.
+
+### Per-plugin content decisions
+
+What survives, plugin by plugin (content list, not full prose):
+
+**auto-format**
+
+- Doctrines: Format Hook always exits zero (ADR-0001); consumer owns Formatters (ADR-0003); zero runtime dependencies bar (ADR-0002); per-project config merge (ADR-0004); POSIX path normalisation on Windows (ADR-0006).
+- Layout: `hooks/`, `scripts/`, `tests/`, `.claude-plugin/`.
+- External deps: none.
+- Do not add: bundled Formatters, MCP server, bash hooks, pre-commit hooks.
+
+**pr-review**
+
+- Doctrines: canonical bot signature (ADR-0001) — exact wording lives in the file; signature-based prior-review detection (ADR-0002); target latest iteration (ADR-0003); incremental diff baseline (ADR-0004); four-state thread classification (ADR-0005); reply-not-duplicate auto-resolve (ADR-0006); summary rewritten not appended (ADR-0007); soft dependency on `pr-review-toolkit` (ADR-0008); orchestrator-thin rule (≤ 200 lines).
+- Layout: `commands/`, `agents/`, `scripts/`, `tests/`, `.claude-plugin/`.
+- External deps: Azure CLI with `azure-devops` extension; `pr-review-toolkit` plugin from `anthropics/claude-plugins-official`.
+- Do not add: GitHub PR support (until ADR), vote-on-PR action, PR description generation.
+
+**unic-confluence**
+
+- Doctrines: refuse publish without markers (ADR-0001); three-strategy injection priority (ADR-0002); structured macro for code blocks (ADR-0003); dry-run read-only (ADR-0004); ping-check auth over per-page-verify (ADR-0005); hard HTTP timeout (ADR-0006); `CliError` class (ADR-0007); pure-functions lib with tests (ADR-0008); bare-integer page-id schema (ADR-0009); no catalog for runtime deps (ADR-0010); alias auto-population (ADR-0011); do-not-add-scope-guard (ADR-0012).
+- Layout: `commands/`, `scripts/`, `.claude-plugin/`.
+- External deps: Confluence v2 REST API endpoint; `~/.unic-confluence.json` credentials file.
+- Do not add: the existing eight-item list (image upload, create-page, multi-space, MCP server, agents/sub-agents, recursive publishing, changesets, watch mode) survives intact — it is the most valuable content in the current file.
+
+**unic-archon-dlc** (new file)
+
+- Doctrines: Setup is the sole entry point and is idempotent (ADR-0001 in this Plugin's own `docs/adr/`); Session is scoped by Slug; HANDOFF.md / ROADMAP.md are written exclusively by the triage workflow; the marker-delimited `## Agent skills` block in a Consumer's `CLAUDE.md` is auto-managed; this monorepo dogfoods the Plugin against itself (cross-link to the separate dogfood-banner PRD).
+- Layout: `commands/`, `lib/`, `hooks/`, `test/`, `docs/`, `.claude-plugin/`.
+- External deps: Archon workflow engine version ≥ 0.10 in the Consumer project.
+- Do not add: parallel-runner support before the linear path is operational; per-plugin variants of the Slug scheme; Consumer-side opt-out flags for individual workflow phases (until a real need surfaces).
+- Dogfooding note: explicitly states that this monorepo has had Setup run against it, points the reader at `docs/agents/*.md` as the generated output, and warns that those files describe the target workflow rather than current local practice.
+
+### What gets dropped (across all four files)
+
+- pnpm script listings that mirror the root (`pnpm install`, `pnpm test` (root-level), `pnpm check`, `pnpm format`, `pnpm ci:check`, `pnpm typecheck` (root)). The plugin-specific `bump`/`sync-version`/`tag`/`verify:changelog` scripts remain.
+- SemVer policy tables — these duplicate ADR-0022 and root `AGENTS.md`.
+- Conventional Commits scope explanation — duplicates root `AGENTS.md` and ADR-0021.
+- Tag scheme explanation — duplicates ADR-0008 and root `AGENTS.md`.
+- Generic "code conventions" sections (tabs vs spaces, single quotes, ESM) — root `AGENTS.md` owns these.
+- Cross-platform requirement statements — root owns this.
+- Standalone-repo install instructions ("Add to `~/.claude/settings.json`") — moved to each Plugin's `README.md` if not already there; out of scope of this PRD beyond noting it.
+- `unic-confluence`'s "Naming convention" table — obsolete since plugin migration; the canonical Plugin identifier now lives in `.claude-plugin/plugin.json`.
+
+### Symlink mechanics
+
+The symlink is created with a relative target so it survives a `git clone` on any OS that supports symlinks. On Windows, repository owners with the relevant Git config (`core.symlinks=true`) get a working symlink; without it, the symlink resolves as a text file containing the target path. This matches how the root `CLAUDE.md -> AGENTS.md` symlink already behaves and so requires no new policy.
+
+### No version bump
+
+This is a documentation-only change to Plugin directories. No Plugin's behaviour changes; no `plugin.json` field changes; no new commands or hooks ship. `pnpm --filter <name> bump` is not invoked for any Plugin. The CHANGELOGs are not touched. `verify:changelog` ignores doc-only PRs by design (PR-only gate, see ADR-0017) — confirm this still holds for the diff before merging.
+
+### One PR or four?
+
+Single PR is acceptable because the changes are mechanically uniform across all four Plugins and reviewing the four files side-by-side is easier than tracking four separate PRs. If the diff grows unwieldy, split per Plugin. Default: one PR targeting `develop`.
+
+## Testing Decisions
+
+This is a documentation refresh with no code changes; there is no test suite to update.
+
+The verification surface is human review of the four `AGENTS.md` files plus the four `CLAUDE.md` symlinks. Reviewer checks:
+
+- Each `AGENTS.md` follows the eight-section template; sections absent only when the Plugin genuinely lacks content for that section.
+- No content from the dropped list (root-mirroring pnpm scripts, SemVer, Conventional Commits, code conventions) appears in any plugin file.
+- Per-plugin doctrines listed above are present and link to the relevant Plugin ADR by number.
+- Each `CLAUDE.md` is a symlink to `AGENTS.md`, not a regular file with duplicated content.
+- Links from each plugin file resolve correctly on GitHub (relative paths up to the repo root).
+- `pnpm check` and `pnpm format` pass against the new Markdown files (Prettier handles `.md`).
+- No reference remains to the retired identifiers `unic-claude-code-format`, `unic-claude-code-<service>`, or `spec-NN` commit scopes.
+
+If automated assertions are desired in the future (e.g. a script that asserts every plugin directory has an `AGENTS.md` and a matching `CLAUDE.md` symlink), that can be added later as a separate Feature. Not included here.
+
+## Out of Scope
+
+- Any code changes inside the Plugins (no command behaviour, no hook contract, no lib refactor).
+- Changes to the root `AGENTS.md` / `CLAUDE.md` symlink.
+- Changes to per-plugin `CONTEXT.md` — they own the domain glossary and remain as-is.
+- The dogfood-banner work tracked separately (`docs/issues/unic-archon-dlc-dogfood-banner/`).
+- ADR reshelving from root `docs/adr/` into `apps/claude-code/unic-archon-dlc/docs/adr/`, tracked separately.
+- Archival or deletion of per-plugin `docs/plans/` directories — handled by the cleanup chore PR.
+- Refreshing each Plugin's `README.md` (Consumer-facing). Out of scope; if the standalone-repo install instructions that get dropped from `CLAUDE.md` are missing from the README, address in a follow-up.
+- An assertion script that enforces the `AGENTS.md` + `CLAUDE.md` symlink convention across all Plugins.
+
+## Further Notes
+
+- The pr-review file currently references "spec 02" without linking. The refreshed file replaces this with the corresponding ADR link (`apps/claude-code/pr-review/docs/adr/0002-signature-based-prior-review-detection.md`).
+- `unic-archon-dlc`'s file is the only fully new artifact; the other three are rewrites. Treat the new file as the canonical example of the template and align the other three to its structure.
+- The PRD assumes the `AGENTS.md` + `CLAUDE.md` symlink pattern is the right answer for the monorepo. The root already follows this. If a Consumer's tooling required Plugin directories to contain a regular `CLAUDE.md` file rather than a symlink, this would change; no such Consumer is known.
+- Re-running `unic-archon-dlc` setup against this monorepo after the new `AGENTS.md` lands would update the marker-delimited `## Agent skills` block in root `AGENTS.md` only. The per-Plugin `AGENTS.md` files are not touched by Setup, so no regeneration risk.
diff --git a/docs/issues/pr-review-agents-not-discovered/01-rename-agents-dir-and-fix-frontmatter.md b/docs/issues/pr-review-agents-not-discovered/01-rename-agents-dir-and-fix-frontmatter.md
deleted file mode 100644
index a35c0e19..00000000
--- a/docs/issues/pr-review-agents-not-discovered/01-rename-agents-dir-and-fix-frontmatter.md
+++ /dev/null
@@ -1,66 +0,0 @@
----
-title: 'pr-review: rename .agents/ → agents/ and add name: field to agent frontmatter'
-created: 2026-05-19
----
-
-**Status:** resolved
-**Category:** bug
-**Plugin:** `apps/claude-code/pr-review`
-**Depends on:** —
-
-## Problem Statement
-
-`pr-review:ado-fetcher` (and all other `pr-review` plugin agents) fail at runtime with:
-
-```
-Error: Agent type 'pr-review:ado-fetcher' not found.
-```
-
-**Root cause:** Claude Code's agent discovery scans an `agents/` directory inside each plugin. The `pr-review` plugin stores its agent files in `.agents/` (dot-prefixed, hidden on Unix), which is never scanned. The `pr-review-toolkit` plugin — which works correctly — uses `agents/` (no dot).
-
-**Secondary issue:** The `pr-review-toolkit` agent files declare a `name:` field in their YAML frontmatter (e.g. `name: code-reviewer`). The `pr-review` agent files only have `allowed-tools` and `description`. Whether the missing `name:` field is independently load-bearing is uncertain; it is added here defensively to match the working convention.
-
-**Evidence:**
-
-- Working: `~/.claude/plugins/cache/claude-plugins-official/pr-review-toolkit/unknown/agents/code-reviewer.md`
-- Broken: `~/.claude/plugins/cache/unic-agent-plugins/pr-review/1.2.10/.agents/ado-fetcher.md`
-
-## Affected files
-
-| File                                                                           | Change                                                                                      |
-| ------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------- |
-| `apps/claude-code/pr-review/.agents/` (directory)                              | Rename to `agents/`                                                                         |
-| `apps/claude-code/pr-review/.agents/ado-fetcher.md`                            | Add `name: ado-fetcher` to frontmatter                                                      |
-| `apps/claude-code/pr-review/.agents/ado-writer.md`                             | Add `name: ado-writer` to frontmatter                                                       |
-| `apps/claude-code/pr-review/.agents/re-review-coordinator.md`                  | Add `name: re-review-coordinator` to frontmatter                                            |
-| `apps/claude-code/pr-review/.agents/doc-context-orchestrator.md`               | Add `name: doc-context-orchestrator` to frontmatter                                         |
-| `apps/claude-code/pr-review/.agents/doc-context-synthesizer.md`                | Add `name: doc-context-synthesizer` to frontmatter                                          |
-| `apps/claude-code/pr-review/CLAUDE.md`                                         | Update repository layout section: `.agents/` → `agents/`                                    |
-| `apps/claude-code/pr-review/docs/adr/0013-orchestrator-split-for-review-pr.md` | Amend in place: fix `.agents/` → `agents/` in the "Three focused agents live in…" paragraph |
-| `CHANGELOG.md`                                                                 | Add patch entry                                                                             |
-| `.claude-plugin/plugin.json` + `marketplace.json`                              | Bump patch version                                                                          |
-
-Do **not** update historical plan files (`docs/plans/`) — those are already-executed specs and are left as written.
-
-## Implementation steps
-
-1. Rename the directory: `git mv apps/claude-code/pr-review/.agents apps/claude-code/pr-review/agents`
-2. Add `name: <slug>` as the first frontmatter key in each of the five agent files (use the filename stem as the value, e.g. `name: ado-fetcher`).
-3. Update the repository layout table in `CLAUDE.md`: change `.agents/` → `agents/` in the directory tree and in any prose that names the directory.
-4. Amend ADR 0013 in place: change the one occurrence of `the plugin's \`.agents/\` directory`to`the plugin's \`agents/\` directory`. No status or consequence lines need changing.
-5. Add a `CHANGELOG.md` entry and bump the patch version in `plugin.json` + `marketplace.json`.
-
-## Verification
-
-After implementing, verify the fix end-to-end:
-
-1. **Cache invalidation check** — it is unknown whether the directory-source marketplace re-reads from the repo live or serves from the `~/.claude/plugins/cache/` snapshot. After renaming, check whether `~/.claude/plugins/cache/unic-agent-plugins/pr-review/<version>/.agents/` still exists. If it does, manually remove it (or delete and re-add the `pr-review@unic` entry in `enabledPlugins`) to force a fresh cache read.
-2. **Agent discovery** — restart Claude Code and confirm `pr-review:ado-fetcher` appears in the available agent list (the list shown in the error message is a reliable source of truth).
-3. **End-to-end smoke test** — run `/pr-review:review-pr` against a real ADO PR URL and confirm the ADO Fetcher agent launches without the "not found" error.
-
-## Acceptance criteria
-
-- `pr-review:ado-fetcher`, `pr-review:ado-writer`, `pr-review:re-review-coordinator`, `pr-review:doc-context-orchestrator`, and `pr-review:doc-context-synthesizer` all appear in Claude Code's available agent list after the rename.
-- Running `/pr-review:review-pr <ADO-PR-URL>` reaches at least Step 5 (ADO Fetcher launch) without an "Agent type not found" error.
-- No references to `.agents/` remain in `CLAUDE.md` or `docs/adr/0013-*`.
-- ADR 0013 retains its original status and all consequence lines; only the directory name is corrected.
diff --git a/docs/issues/pr-review-doc-context-enrichment/01-confluence-page-client.md b/docs/issues/pr-review-doc-context-enrichment/01-confluence-page-client.md
deleted file mode 100644
index 205c159a..00000000
--- a/docs/issues/pr-review-doc-context-enrichment/01-confluence-page-client.md
+++ /dev/null
@@ -1,83 +0,0 @@
-# Confluence page client script + tests
-
-**Status:** closed
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-doc-context-enrichment/PRD.md`
-
-## What to build
-
-Create `scripts/confluence-client.mjs` — a self-contained Node.js script (zero external runtime dependencies, `// @ts-check` + JSDoc) that loads Confluence credentials and fetches a page by URL via the Confluence v2 REST API.
-
-**Credential loading** follows the shared Unic precedence: environment variables (`CONFLUENCE_URL`, `CONFLUENCE_USER`, `CONFLUENCE_TOKEN`) take precedence; `~/.unic-confluence.json` is the fallback (same format as `unic-confluence`). Throws a descriptive error if neither source yields valid credentials.
-
-**Page fetching**: given a Confluence page URL, extract the numeric page ID (pattern: `/pages/{id}/` or `/pages/{id}` with optional trailing slug), call `GET /wiki/api/v2/pages/{pageId}?body-format=storage` with Basic auth, and return the `body.storage.value` string (Confluence storage XML). Throw a descriptive error on non-2xx responses or network failures — the caller handles warnings and fallback.
-
-**CLI entry point**: when invoked directly (`node scripts/confluence-client.mjs <url>`), print the storage-format body to stdout and exit non-zero on any error. Also support `--check-creds` (exit 0 if credentials are available, non-zero otherwise).
-
-**Test suite** (`node:test`, no live Confluence connection): mock the HTTPS layer with fixtures. Cover:
-
-- Credential loading: env vars take precedence over `~/.unic-confluence.json`
-- Credential loading: throws with actionable message when neither source is configured
-- Page ID extraction from URL variants (`/pages/{id}/slug`, `/pages/{id}`, no trailing slash)
-- Returns storage-format body on 200 response
-- Throws descriptive error on 401, 403, 404
-- Throws descriptive error on network failure / timeout
-
-Follow the fixture pattern established in `apps/claude-code/pr-review/docs/plans/09-test-harness.md`.
-
-## Acceptance criteria
-
-- [ ] `scripts/confluence-client.mjs` exists with zero external runtime dependencies
-- [ ] `node scripts/confluence-client.mjs <valid-url>` fetches and prints page content (manual smoke test)
-- [ ] `node scripts/confluence-client.mjs --check-creds` exits 0 when credentials are configured, non-zero otherwise
-- [ ] All `node:test` cases pass
-- [ ] Credential lookup order mirrors `unic-confluence` exactly
-
-## Blocked by
-
-None — can start immediately.
-
-## Comments
-
-> _This was generated by AI during triage._
-
-## Agent Brief
-
-**Category:** enhancement
-**Summary:** Create a self-contained Confluence page client script with credential loading, page fetching, and a node:test suite — all with zero external runtime dependencies.
-
-**Current behavior:**
-The pr-review plugin has no mechanism to fetch Confluence pages. No client script exists.
-
-**Desired behavior:**
-A standalone Node.js script (`// @ts-check`, ESM) is available in the plugin's `scripts/` directory. It loads Confluence credentials following the Unic precedence order (env vars → `~/.unic-confluence.json`), accepts a Confluence page URL, extracts the numeric page ID from the URL path, fetches the page body in storage format via the Confluence v2 REST API, and returns the raw `body.storage.value` string. It exposes a `--check-creds` CLI flag that exits 0 when credentials are present and non-zero otherwise. On any error (bad credentials, non-2xx response, network failure) it throws a descriptive message — the caller handles warnings and fallback.
-
-The test suite uses `node:test` with a mocked HTTPS layer (no live Confluence connection). It covers the credential precedence, all URL ID extraction patterns, the happy path, and the main error conditions.
-
-**Key interfaces:**
-
-- `loadCredentials()` — reads env vars first, falls back to `~/.unic-confluence.json`; throws a descriptive `Error` if neither source is configured. Mirror the implementation in `unic-confluence`'s push script exactly.
-- `fetchPageText(pageUrl, credentials)` — extracts numeric page ID from URL (handles `/pages/{id}/slug`, `/pages/{id}`, no trailing slash), calls Confluence v2 `GET /wiki/api/v2/pages/{id}?body-format=storage` with Basic auth, returns `body.storage.value`; throws on non-2xx or network error.
-- CLI `--check-creds` flag — exits 0 if credentials load without error, non-zero otherwise.
-- `~/.unic-confluence.json` shape — `{ url, username, token }` (same as `unic-confluence`).
-
-**Acceptance criteria:**
-
-- [ ] Script exists with zero external runtime dependencies
-- [ ] `--check-creds` exits 0 when credentials are configured, non-zero otherwise
-- [ ] Page ID extracted correctly from all three URL variants
-- [ ] Returns storage-format body on 200 response
-- [ ] Throws descriptive error on 401, 403, 404, and network failure
-- [ ] Credential loading: env vars take precedence over `~/.unic-confluence.json`
-- [ ] Credential loading: throws actionable message when neither source is configured
-- [ ] All `node:test` cases pass without a live Confluence connection
-- [ ] Credential lookup order is identical to `unic-confluence`'s push script
-
-**Out of scope:**
-
-- Parsing or stripping Confluence storage XML — callers receive the raw value
-- Caching page content across invocations
-- Any changes to `commands/review-pr.md` (that is issues 02 and 03)
diff --git a/docs/issues/pr-review-doc-context-enrichment/02-work-item-doc-context-enrichment.md b/docs/issues/pr-review-doc-context-enrichment/02-work-item-doc-context-enrichment.md
deleted file mode 100644
index abc18508..00000000
--- a/docs/issues/pr-review-doc-context-enrichment/02-work-item-doc-context-enrichment.md
+++ /dev/null
@@ -1,81 +0,0 @@
-# Work item Doc Context enrichment
-
-**Status:** closed
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-doc-context-enrichment/PRD.md`
-
-## What to build
-
-Add a new **step 4a** to `commands/review-pr.md` that runs after the changed files list (step 4) and in parallel with the local diff and file reading steps (5–7). Step 8 waits for step 4a to complete before launching review agents.
-
-Step 4a fetches the ADO work items linked to the PR via the `pullRequestWorkItems` API endpoint. If the list is empty, it exits silently with no change to behaviour. For each linked work item, it fetches the work item details (`az boards work-item show`) and spawns one **Doc Context Sub-agent** per work item in parallel (single message).
-
-Each Doc Context Sub-agent receives the work item ID, title, and description (HTML), plus the changed files list and the local diff if already available. It produces a diff-aware summary of the work item description — focused only on what is relevant to the changed files — and returns a structured block:
-
-```
-### Work item: [{ID}] {Title}
-{diff-aware summary}
-```
-
-Step 4a collects all sub-agent outputs, prepends a `## Business context for this PR` header, and stores the result as `DOC_CONTEXT`. Step 8 injects `DOC_CONTEXT` as a preamble before the diff content in every review agent prompt. When `DOC_CONTEXT` is empty (no work items, or all sub-agents returned nothing useful), the preamble is omitted and step 8 behaves exactly as today.
-
-## Acceptance criteria
-
-- [ ] A PR with one or more linked ADO work items: review agents receive a `## Business context for this PR` preamble with a summarised work item section per item
-- [ ] A PR with no linked work items: step 4a exits silently, review agent prompts are unchanged
-- [ ] Multiple work items are all included in the Doc Context block
-- [ ] Step 4a runs in parallel with steps 5–7 (not sequentially before them)
-- [ ] When the local diff is not yet available at step 4a time, sub-agents work from the changed files list only
-- [ ] No new comments or threads are posted to the PR
-
-## Blocked by
-
-None — can start immediately.
-
-## Comments
-
-> _This was generated by AI during triage._
-
-## Agent Brief
-
-**Category:** enhancement
-**Summary:** Add step 4a to the review command that fetches linked ADO work items, spawns parallel Doc Context Sub-agents to summarise them, and injects the result as a business context preamble into every review agent's prompt.
-
-**Current behavior:**
-The review command (steps 1–12) passes the PR title, description, diff, and changed file contents to review agents. No work item data is fetched. Review agents have no visibility into the business intent of the PR.
-
-**Desired behavior:**
-A new step 4a runs after the changed files list (step 4) and in parallel with steps 5–7. It calls the ADO `pullRequestWorkItems` API to get the IDs of all work items linked to the PR. If none are linked, step 4a exits silently and behaviour is unchanged. For each linked work item it fetches the full work item record (`az boards work-item show`) and spawns one Doc Context Sub-agent per work item in a single parallel message. Each sub-agent receives the work item ID, title, and HTML description, plus the changed files list and local diff (if already available). It returns a diff-aware summary as:
-
-```
-### Work item: [{ID}] {Title}
-{diff-aware summary focused on what is relevant to the changed files}
-```
-
-Step 4a collects all sub-agent outputs, prepends a `## Business context for this PR` header, and stores the assembled block as `DOC_CONTEXT`. Step 8 injects `DOC_CONTEXT` as a preamble before the diff content in every review agent prompt. If `DOC_CONTEXT` is empty, step 8 behaves exactly as before.
-
-**Key interfaces:**
-
-- ADO `pullRequestWorkItems` resource — called via `az devops invoke` with `repositoryId`, `pullRequestId`, and `org` route parameters; returns `{ value: [{ id }] }`
-- `az boards work-item show --id {id} --org {ORG_URL}` — fields needed: `System.Title`, `System.Description`
-- Doc Context block header: `## Business context for this PR`
-- Work item section format: `### Work item: [{ID}] {Title}` followed by the summary
-- Step 8 prompt preamble injection: `DOC_CONTEXT` precedes the diff section
-
-**Acceptance criteria:**
-
-- [ ] PR with linked work items: review agents receive a `## Business context for this PR` preamble with one section per work item
-- [ ] PR with no linked work items: step 4a exits silently, review agent prompts unchanged
-- [ ] Multiple linked work items: all appear in the Doc Context block
-- [ ] Step 4a spawns sub-agents in parallel with steps 5–7, not before them
-- [ ] Sub-agents function correctly when local diff is not yet available (work from changed files list only)
-- [ ] No new comments or threads are posted to the PR
-
-**Out of scope:**
-
-- Fetching Confluence pages (that is issue 03)
-- Any credential handling (no Confluence dependency in this issue)
-- Changes to the confluence-client script (that is issue 01)
diff --git a/docs/issues/pr-review-doc-context-enrichment/03-confluence-page-doc-context-enrichment.md b/docs/issues/pr-review-doc-context-enrichment/03-confluence-page-doc-context-enrichment.md
deleted file mode 100644
index a0152915..00000000
--- a/docs/issues/pr-review-doc-context-enrichment/03-confluence-page-doc-context-enrichment.md
+++ /dev/null
@@ -1,116 +0,0 @@
-# Confluence page Doc Context enrichment
-
-**Status:** closed
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-doc-context-enrichment/PRD.md`
-
-## What to build
-
-Extend the work item Doc Context Sub-agent (from issue 02) to follow Confluence page links found in work item descriptions.
-
-Each work item Doc Context Sub-agent, after summarising the work item description, extracts all Confluence page URLs from the description text. It then checks whether Confluence credentials are available (`node scripts/confluence-client.mjs --check-creds`).
-
-**If credentials are available**: spawn one nested **Confluence page Doc Context Sub-agent** per extracted URL in parallel. Each page sub-agent invokes `node scripts/confluence-client.mjs <url>` to fetch the storage-format page body, then produces a diff-aware plain-text summary of the page content (reading through Confluence storage-format XML directly, no dedicated parser). It returns a structured block:
-
-```
-### Confluence — {Page Title} ({URL})
-{diff-aware summary}
-```
-
-These blocks are appended after the work item summary section in the Doc Context block for that work item.
-
-**If credentials are absent and at least one Confluence URL was found**: emit a console warning (never post to the PR):
-
-```
-⚠ Confluence pages not fetched — set CONFLUENCE_URL, CONFLUENCE_USER,
-  CONFLUENCE_TOKEN (or configure ~/.unic-confluence.json) to enable
-  doc-aware review.
-```
-
-**If credentials are absent and no Confluence URLs were found**: skip silently.
-
-**If a page fetch fails** (network error, 401, 403, or any non-2xx): skip that page and emit a per-page console warning:
-
-```
-⚠ Could not fetch Confluence page <url> — <reason>
-```
-
-Continue with whatever other context was gathered. A single failing page does not block the review.
-
-## Acceptance criteria
-
-- [ ] A PR whose work item description links to a Confluence page, with valid credentials: review agents receive the work item summary plus a Confluence page summary with provenance label (title + URL)
-- [ ] Multiple Confluence links in one work item: all pages are fetched in parallel and included
-- [ ] Credentials absent + Confluence URLs found: console warning shown, review continues with work item summaries only
-- [ ] Credentials absent + no Confluence URLs: no warning, no change in output
-- [ ] Confluence page returns 403: that page skipped with per-page console warning; other pages and work items still included
-- [ ] Network error on a page fetch: same skip-and-warn behaviour as 403
-- [ ] Warnings are emitted to the console only — nothing posted to the PR
-
-## Blocked by
-
-- `docs/issues/pr-review-doc-context-enrichment/01-confluence-page-client.md`
-- `docs/issues/pr-review-doc-context-enrichment/02-work-item-doc-context-enrichment.md`
-
-## Comments
-
-> _This was generated by AI during triage._
-
-## Agent Brief
-
-**Category:** enhancement
-**Summary:** Extend each work item Doc Context Sub-agent to extract Confluence URLs from the work item description, fetch those pages using the Confluence client script, and append diff-aware page summaries to the Doc Context block — with graceful degradation when credentials are absent or a fetch fails.
-
-**Current behavior:**
-Work item Doc Context Sub-agents (from issue 02) summarise work item descriptions only. They do not follow any Confluence links found in those descriptions.
-
-**Desired behavior:**
-After summarising the work item description, each work item Doc Context Sub-agent extracts all Confluence page URLs from the description text. It then runs `node scripts/confluence-client.mjs --check-creds` to determine whether credentials are available.
-
-If credentials are present: spawn one nested Confluence page Doc Context Sub-agent per extracted URL in a single parallel message. Each page sub-agent runs `node scripts/confluence-client.mjs <url>` to fetch the storage-format page body, reads through the XML directly (no dedicated parser), and returns a diff-aware plain-text summary as:
-
-```
-### Confluence — {Page Title} ({URL})
-{diff-aware summary}
-```
-
-These sections are appended after the work item summary in the work item's Doc Context block.
-
-If credentials are absent and at least one Confluence URL was found: emit a console warning (never post to the PR):
-
-```
-⚠ Confluence pages not fetched — set CONFLUENCE_URL, CONFLUENCE_USER,
-  CONFLUENCE_TOKEN (or configure ~/.unic-confluence.json) to enable
-  doc-aware review.
-```
-
-If credentials are absent and no Confluence URLs were found: skip silently.
-
-If any individual page fetch fails (any error): skip that page, emit a per-page console warning (`⚠ Could not fetch Confluence page <url> — <reason>`), and continue with whatever other context was gathered.
-
-**Key interfaces:**
-
-- `node scripts/confluence-client.mjs --check-creds` — exit 0 = credentials present
-- `node scripts/confluence-client.mjs <url>` — stdout = storage-format XML; non-zero exit = error with message on stderr
-- Confluence page section format: `### Confluence — {Page Title} ({URL})` followed by the summary
-- Console warning format: `⚠ Confluence pages not fetched — …` (credentials absent) and `⚠ Could not fetch Confluence page <url> — <reason>` (per-page error)
-
-**Acceptance criteria:**
-
-- [ ] PR with a work item linking to a Confluence page, valid credentials: review agents receive work item summary + Confluence page summary with provenance label
-- [ ] Multiple Confluence links in one work item: all fetched in parallel and included
-- [ ] Credentials absent + Confluence URLs found: single console warning; review continues with work item summaries only
-- [ ] Credentials absent + no Confluence URLs: no warning, no change in output
-- [ ] Page fetch returns 403: that page skipped with per-page console warning; other pages and work items still included
-- [ ] Network error on a fetch: same skip-and-warn behaviour
-- [ ] All warnings go to the console only — nothing posted to the PR
-
-**Out of scope:**
-
-- Parsing Confluence storage XML with a dedicated parser — the sub-agent reads through it directly
-- Caching page content across review runs
-- Fetching Confluence pages linked from the PR description (not via a work item)
-- Any changes to the confluence-client script itself (that is issue 01)
diff --git a/docs/issues/pr-review-doc-context-enrichment/04-version-bump-and-docs.md b/docs/issues/pr-review-doc-context-enrichment/04-version-bump-and-docs.md
deleted file mode 100644
index 88d3a182..00000000
--- a/docs/issues/pr-review-doc-context-enrichment/04-version-bump-and-docs.md
+++ /dev/null
@@ -1,48 +0,0 @@
-# Version bump + CHANGELOG + docs
-
-**Status:** closed
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-doc-context-enrichment/PRD.md`
-
-## What to build
-
-Finalise metadata, documentation, and release artefacts for the doc context enrichment feature.
-
-**Before running the bump script**, check what version the re-review feature shipped at (`docs/issues/pr-review-rereview/09-version-bump-and-release.md`). The bump for this feature must target the next available minor version after re-review — do not hardcode a version number here.
-
-**Steps (in order):**
-
-1. **Add CHANGELOG entry.** Under `## [Unreleased]` in `CHANGELOG.md`, add under `### Added`:
-
-   > Doc Context enrichment: before review agents run, fetch linked ADO work items and any Confluence pages referenced in their descriptions; inject structured, diff-aware summaries as business context into every review agent's prompt. Requires Confluence credentials (`CONFLUENCE_URL`, `CONFLUENCE_USER`, `CONFLUENCE_TOKEN` or `~/.unic-confluence.json`) for Confluence page fetching; degrades gracefully when absent or unreachable.
-
-2. **Bump the version.** Run `pnpm --filter pr-review bump minor`. This updates both `.claude-plugin/plugin.json` and `.claude-plugin/marketplace.json` atomically. Never hand-edit `marketplace.json`.
-
-3. **Update `README.md`.** Add a section describing the Doc Context enrichment capability: what it does, credential prerequisites (`CONFLUENCE_URL`, `CONFLUENCE_USER`, `CONFLUENCE_TOKEN` or `~/.unic-confluence.json`), and graceful degradation behaviour.
-
-4. **Update `CLAUDE.md`.** Remove "PR description generation from diff" and any doc-context-enrichment roadmap lines if present. Verify that new step 4a is reflected in the repository layout description.
-
-5. **Verify.** Run `pnpm --filter pr-review verify:changelog` locally before opening the PR.
-
-## Acceptance criteria
-
-- [ ] `pnpm --filter pr-review verify:changelog` passes locally
-- [ ] Versions match across `plugin.json` and `marketplace.json`
-- [ ] `CHANGELOG.md` has a dated entry for the new version under `### Added`
-- [ ] `README.md` documents the enrichment capability and credential prerequisites
-- [ ] CI passes on the release PR
-
-## Blocked by
-
-- `docs/issues/pr-review-doc-context-enrichment/02-work-item-doc-context-enrichment.md`
-- `docs/issues/pr-review-doc-context-enrichment/03-confluence-page-doc-context-enrichment.md`
-
-## Comments
-
-> _This was generated by AI during triage._
-
-**Why this is ready-for-human, not ready-for-agent:**
-The version number cannot be determined without first knowing what version the re-review feature shipped at. That requires checking a real-time state (`docs/issues/pr-review-rereview/09-version-bump-and-release.md`) that may not be resolved when this issue is picked up. Additionally, the CHANGELOG entry and README section involve prose decisions that benefit from human authorship. The mechanical bump step (`pnpm --filter pr-review bump minor`) is straightforward once the version is confirmed.
diff --git a/docs/issues/pr-review-doc-context-spawn-reliability/01-adr-and-synthesizer-agent.md b/docs/issues/pr-review-doc-context-spawn-reliability/01-adr-and-synthesizer-agent.md
deleted file mode 100644
index 307c0a7e..00000000
--- a/docs/issues/pr-review-doc-context-spawn-reliability/01-adr-and-synthesizer-agent.md
+++ /dev/null
@@ -1,70 +0,0 @@
-# ADR-0012 + Doc Context Synthesizer agent
-
-**Status:** closed
-**Category:** enhancement
-**Type:** AFK
-
-## Parent
-
-`docs/issues/pr-review-doc-context-spawn-reliability/PRD.md`
-
-## What to build
-
-Write ADR-0012 recording the decision to have the Doc Context Orchestrator agent return plain markdown rather than a JSON wrapper. Then create the Doc Context Synthesizer agent — a new plugin agent that receives all Work Item Summarizer outputs and Confluence Fetcher outputs, and produces a single `## Business context for this PR` flat narrative with no per-work-item headings, no redundant content, focused on business intent. Returns an empty string if no meaningful context was gathered.
-
-## Acceptance criteria
-
-- [ ] `docs/adr/0012-plain-text-doc-context-agent-return.md` exists and records: the decision (plain markdown), the rejected alternative (JSON wrapper), and the reasons (LLM agents produce malformed JSON under long-context conditions; warnings are already emitted as side effects; `jq` extraction from LLM output is unreliable).
-- [ ] `.agents/doc-context-synthesizer.md` exists as a plugin agent file.
-- [ ] The synthesizer agent prompt instructs it to produce a flat `## Business context for this PR` narrative — no per-work-item headings, no structural grouping by ticket.
-- [ ] The prompt instructs the agent to merge overlapping content (same feature described by multiple work items or Confluence pages) rather than repeating it.
-- [ ] The prompt instructs the agent to focus on business intent (what the PR should accomplish and why) and omit implementation details already visible in the diff.
-- [ ] The prompt instructs the agent to return an empty string if no meaningful context was gathered.
-
-## Blocked by
-
-None — can start immediately.
-
----
-
-## Agent Brief
-
-> _This was generated by AI during triage._
-
-**Category:** enhancement
-**Summary:** Create ADR-0012 (plain-markdown return format) and the Doc Context Synthesizer agent
-
-**Current behavior:**
-
-Neither file exists. The ADR decision and the synthesizer agent are prerequisites for the orchestrator agent (issue 02) and the step 4a wire-up (issue 03).
-
-**Desired behavior:**
-
-`docs/adr/0012-plain-text-doc-context-agent-return.md` exists and records the decision to return plain markdown from the Doc Context Orchestrator agent (not a JSON wrapper), the rejected JSON alternative, and the three reasons it was rejected (malformed JSON under long context; warnings already emitted as side effects; `jq` extraction from LLM output is unreliable).
-
-`.agents/doc-context-synthesizer.md` exists as a plugin agent. Its prompt instructs the agent to:
-
-- Receive all Work Item Summarizer outputs and Confluence Fetcher outputs
-- Produce a single flat `## Business context for this PR` narrative — no per-work-item headings, no redundant repetition
-- Merge overlapping content where multiple sources describe the same feature
-- Focus on business intent (what the PR should accomplish and why), not implementation details visible in the diff
-- Return an empty string if no meaningful context was gathered
-
-**Key interfaces:**
-
-- `doc-context-synthesizer` plugin agent — input: work item summaries + Confluence page summaries (plain text); output: `## Business context for this PR` markdown block or empty string
-- ADR-0012 — follows the existing ADR format in `docs/adr/`
-
-**Acceptance criteria:**
-
-- [ ] `docs/adr/0012-plain-text-doc-context-agent-return.md` exists with decision, rejected alternative, and reasons
-- [ ] `.agents/doc-context-synthesizer.md` exists as a plugin agent file
-- [ ] Synthesizer prompt produces a flat narrative — no per-work-item headings
-- [ ] Synthesizer prompt merges overlapping content rather than repeating it
-- [ ] Synthesizer prompt returns an empty string when no meaningful context is available
-
-**Out of scope:**
-
-- The orchestrator agent (issue 02)
-- Changes to `commands/review-pr.md` (issue 03)
-- Any changes to `confluence-client.mjs` or its tests
diff --git a/docs/issues/pr-review-doc-context-spawn-reliability/02-orchestrator-agent.md b/docs/issues/pr-review-doc-context-spawn-reliability/02-orchestrator-agent.md
deleted file mode 100644
index c267efa6..00000000
--- a/docs/issues/pr-review-doc-context-spawn-reliability/02-orchestrator-agent.md
+++ /dev/null
@@ -1,86 +0,0 @@
-# Doc Context Orchestrator agent
-
-**Status:** closed
-**Category:** enhancement
-**Type:** AFK
-
-## Parent
-
-`docs/issues/pr-review-doc-context-spawn-reliability/PRD.md`
-
-## What to build
-
-Create the Doc Context Orchestrator agent — a new self-contained plugin agent that handles the entire Doc Context gathering phase in its own context window. It receives: `ORG_URL`, `PR_ID`, a list of work item IDs, the full diff, the changed files list, and the absolute path to `confluence-client.mjs` (pre-expanded by the caller — agents do not inherit env vars).
-
-The agent must:
-
-1. Fetch each work item via `az boards work-item show`. Skip and warn on failure.
-2. Branch on work item type: `Bug` → extract `Microsoft.VSTS.TCM.ReproSteps` + `Microsoft.VSTS.TCM.SystemInfo`; all other types (including unrecognised) → `System.Description`.
-3. Spawn one Work Item Summarizer agent per work item in parallel. Each summarizer receives the work item content, changed files, and diff; returns a plain-text summary and a list of Confluence URLs found in the description.
-4. Run `node {CONFLUENCE_CLIENT_PATH} --check-creds` exactly once. If creds are absent and Confluence URLs were found: emit the standard warning to console and skip all Confluence fetching.
-5. If creds are available: collect all unique Confluence URLs across all summarizer outputs and spawn one Confluence Fetcher agent per unique URL in parallel. Each fetcher runs `node {CONFLUENCE_CLIENT_PATH} {URL}` and returns a plain-text summary or emits a warning and returns nothing on failure.
-6. Pass all summarizer and fetcher outputs to the Doc Context Synthesizer agent.
-7. Return the synthesizer's output verbatim as plain markdown — no JSON wrapper.
-
-## Acceptance criteria
-
-- [ ] `.agents/doc-context-orchestrator.md` exists as a plugin agent file.
-- [ ] The agent prompt references `pr-review:doc-context-synthesizer` for the final synthesis step.
-- [ ] Bug work items: prompt instructs extraction of `ReproSteps` + `SystemInfo`, not `System.Description`.
-- [ ] Unrecognised work item types: prompt instructs fallback to `System.Description`.
-- [ ] Credential check runs exactly once (not per fetcher, not per work item).
-- [ ] All paths to `confluence-client.mjs` in the prompt are described as absolute literal strings passed in by the caller — no relative paths, no env-var references inside the agent.
-- [ ] Console warnings (failed work item fetch, missing creds, failed Confluence fetch) are printed as side effects; none appear in the agent's return value.
-- [ ] The agent returns plain markdown (the synthesizer's output verbatim), consistent with ADR-0012.
-
-## Blocked by
-
-`01-adr-and-synthesizer-agent.md` — synthesizer interface must be settled before this agent references it.
-
----
-
-## Agent Brief
-
-> _This was generated by AI during triage._
-
-**Category:** enhancement
-**Summary:** Create the Doc Context Orchestrator agent — self-contained gathering phase in its own context window
-
-**Current behavior:**
-
-The agent file does not exist. Step 4a currently satisfies doc-context intent inline with no `Agent()` spawn call, so no orchestrator runs and no business context is gathered.
-
-**Desired behavior:**
-
-`.agents/doc-context-orchestrator.md` exists as a plugin agent. When invoked, it receives `ORG_URL`, `PR_ID`, a list of work item IDs, the full diff, the changed files list, and `CONFLUENCE_CLIENT_PATH` (absolute path, pre-expanded by the caller). It orchestrates the entire Doc Context gathering phase and returns the Doc Context Synthesizer's output verbatim as plain markdown.
-
-Key behavioural contracts:
-
-- Per work item: fetch via `az boards work-item show`; branch on type (`Bug` → `ReproSteps` + `SystemInfo`, all others → `System.Description`); skip + warn on failure
-- Spawn one Work Item Summarizer agent per work item in parallel
-- Run `--check-creds` exactly once; if creds absent and Confluence URLs found, emit the standard warning and skip all fetching
-- If creds present: deduplicate Confluence URLs across all summarizer outputs, spawn one Confluence Fetcher agent per unique URL in parallel
-- Delegate all outputs to `pr-review:doc-context-synthesizer`; return its output verbatim
-- All `confluence-client.mjs` paths are the absolute literal string received as input — no env-var references inside the agent prompt
-
-**Key interfaces:**
-
-- `doc-context-orchestrator` plugin agent — inputs: `ORG_URL`, `PR_ID`, work item ID list, diff, changed files, `CONFLUENCE_CLIENT_PATH` (absolute string); output: plain markdown block or empty string
-- `pr-review:doc-context-synthesizer` — called as the final step; must be implemented first (issue 01)
-- Work Item Summarizer agents — spawned inline (anonymous), not as named plugin agent types
-
-**Acceptance criteria:**
-
-- [ ] `.agents/doc-context-orchestrator.md` exists as a plugin agent file
-- [ ] Prompt references `pr-review:doc-context-synthesizer` for the final step
-- [ ] Bug work items use `ReproSteps` + `SystemInfo`; all other types use `System.Description`
-- [ ] Credential check runs exactly once — not per fetcher, not per work item
-- [ ] All `confluence-client.mjs` paths in the prompt are absolute literal strings from the caller — no relative paths
-- [ ] Warnings are printed as side effects; none appear in the return value
-- [ ] Agent returns plain markdown consistent with ADR-0012
-
-**Out of scope:**
-
-- Changes to `commands/review-pr.md` (issue 03)
-- Changes to `confluence-client.mjs` or its tests
-- GitHub PR support, caching, or Jira sources
diff --git a/docs/issues/pr-review-doc-context-spawn-reliability/03-wire-up-and-housekeeping.md b/docs/issues/pr-review-doc-context-spawn-reliability/03-wire-up-and-housekeeping.md
deleted file mode 100644
index 0d5f78dc..00000000
--- a/docs/issues/pr-review-doc-context-spawn-reliability/03-wire-up-and-housekeeping.md
+++ /dev/null
@@ -1,93 +0,0 @@
-# Wire-up: step 4a rewrite + README + CHANGELOG
-
-**Status:** closed
-**Category:** bug
-**Type:** AFK
-
-## Parent
-
-`docs/issues/pr-review-doc-context-spawn-reliability/PRD.md`
-
-## What to build
-
-Rewrite step 4a in `commands/review-pr.md` to fix the three defects that cause the Doc Context phase to be silently skipped:
-
-1. Initialise `DOC_CONTEXT=''` unconditionally as the first statement in step 4a.
-2. Pre-fetch linked work item IDs in bash. If the `value` array is empty or the command fails, leave `DOC_CONTEXT=''` and proceed.
-3. If work items are found: wait for the diff from step 5, resolve `CONFLUENCE_CLIENT_PATH` from `${CLAUDE_PLUGIN_ROOT}` in bash, then spawn the Doc Context Orchestrator agent via an explicit `Agent(subagent_type: "pr-review:doc-context-orchestrator", ...)` call with all required context. Store the agent's plain-text output as `DOC_CONTEXT`.
-4. Update the parallelism note: step 4a pre-fetch runs in parallel with steps 5–7; the orchestrator agent spawn waits for the diff from step 5; step 8 waits for the orchestrator to complete.
-
-Also update `docs/plans/README.md` to add spec 11 to the status table, and add a `Fixed` entry under `[Unreleased]` in `CHANGELOG.md` describing the three defects and the orchestrator extraction.
-
-## Acceptance criteria
-
-- [ ] `DOC_CONTEXT=''` is the first statement in step 4a.
-- [ ] Step 4a contains an explicit `Agent(subagent_type: "pr-review:doc-context-orchestrator", ...)` call — the intent is not satisfied inline.
-- [ ] The `Agent()` call passes `ORG_URL`, `PR_ID`, work item ID list, `CONFLUENCE_CLIENT_PATH` (as an absolute string expanded from `${CLAUDE_PLUGIN_ROOT}`), changed files list, and diff.
-- [ ] No `node scripts/confluence-client.mjs` relative-path call remains in step 4a.
-- [ ] The parallelism note correctly reflects that the orchestrator spawn waits for the diff before starting.
-- [ ] `docs/plans/README.md` includes spec 11 in the status table.
-- [ ] `CHANGELOG.md` has a `Fixed` entry under `[Unreleased]` covering the three defects.
-- [ ] PR with ≥1 linked non-Bug work item and no Confluence links: `DOC_CONTEXT` is non-empty after a review run; review agents receive the `## Business context` preamble.
-- [ ] PR with no linked work items: step 4a exits silently; `DOC_CONTEXT=''`; step 8 prompts unchanged.
-- [ ] No new PR threads or comments produced by the Doc Context phase.
-
-## Blocked by
-
-- `01-adr-and-synthesizer-agent.md`
-- `02-orchestrator-agent.md`
-
----
-
-## Agent Brief
-
-> _This was generated by AI during triage._
-
-**Category:** bug
-**Summary:** Fix step 4a in `review-pr` command — initialise `DOC_CONTEXT`, add explicit Agent() spawn, resolve absolute path
-
-**Current behavior:**
-
-Step 4a has three defects that cause the Doc Context phase to be silently skipped on every run:
-
-1. No `DOC_CONTEXT=''` initialisation — the variable is undefined when the phase is skipped
-2. No explicit `Agent()` spawn call — the orchestrator satisfies intent inline and proceeds without spawning anything
-3. `confluence-client.mjs` is referenced with a relative path that fails when the working directory is the reviewed project's root
-
-**Desired behavior:**
-
-Step 4a in the `review-pr` command is rewritten so that:
-
-- `DOC_CONTEXT=''` is the first statement
-- If no work items are linked, the step exits silently with `DOC_CONTEXT=''`
-- If work items are found, an explicit `Agent(subagent_type: "pr-review:doc-context-orchestrator", ...)` call is made — never satisfied inline — with all required context: `ORG_URL`, `PR_ID`, work item IDs, `CONFLUENCE_CLIENT_PATH` (absolute string from `${CLAUDE_PLUGIN_ROOT}`), changed files list, and diff
-- The agent's plain-text output is stored as `DOC_CONTEXT`
-- The parallelism note reflects the correct sequencing: pre-fetch runs in parallel with steps 5–7; orchestrator spawn waits for the diff; step 8 waits for the orchestrator
-
-`docs/plans/README.md` includes spec 11 in the status table. `CHANGELOG.md` has a `Fixed` entry under `[Unreleased]` covering the three defects.
-
-**Key interfaces:**
-
-- `review-pr` command step 4a — the `Agent()` call must name `subagent_type: "pr-review:doc-context-orchestrator"` explicitly
-- `CONFLUENCE_CLIENT_PATH` — resolved as `"${CLAUDE_PLUGIN_ROOT}/scripts/confluence-client.mjs"` in bash before the spawn
-- `DOC_CONTEXT` — string; empty when no work items are found, plain markdown block otherwise
-
-**Acceptance criteria:**
-
-- [ ] `DOC_CONTEXT=''` is the first statement in step 4a
-- [ ] Explicit `Agent(subagent_type: "pr-review:doc-context-orchestrator", ...)` call present — intent not satisfied inline
-- [ ] `Agent()` call passes all required context variables
-- [ ] No `node scripts/confluence-client.mjs` relative-path call remains in step 4a
-- [ ] Parallelism note correctly reflects orchestrator spawn waiting for the diff
-- [ ] `docs/plans/README.md` includes spec 11
-- [ ] `CHANGELOG.md` has a `Fixed` entry under `[Unreleased]`
-- [ ] PR with ≥1 linked non-Bug work item: `DOC_CONTEXT` non-empty after a review run
-- [ ] PR with no linked work items: step 4a silent; `DOC_CONTEXT=''`
-- [ ] No new PR threads or comments from the Doc Context phase
-
-**Out of scope:**
-
-- The agent file implementations (issues 01 and 02)
-- Changes to `confluence-client.mjs` or its tests
-- Thread posting, re-review logic, or Bot Signature
-- GitHub PR support, caching, or Jira sources
diff --git a/docs/issues/pr-review-orchestrator-split/01-create-ado-fetcher-agent.md b/docs/issues/pr-review-orchestrator-split/01-create-ado-fetcher-agent.md
deleted file mode 100644
index 6e6e2cbf..00000000
--- a/docs/issues/pr-review-orchestrator-split/01-create-ado-fetcher-agent.md
+++ /dev/null
@@ -1,65 +0,0 @@
-# Create ADO Fetcher agent
-
-**Status:** resolved
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-orchestrator-split/PRD.md`
-
-## What to build
-
-Create a new plugin agent (`pr-review:ado-fetcher`) that encapsulates all Azure DevOps read operations required for a PR review. The agent receives a PR URL (org, project, PR ID) and an optional prior iteration ID (passed in for re-review, empty string for first-review), and returns a structured context block containing: PR metadata (title, description, source/target branches, repo ID), latest iteration ID and its commit SHA, changed files list, raw diff, and work-item IDs linked to the PR.
-
-This agent replaces the inline ADO shell commands currently scattered across Steps 2–5 of the `review-pr` command. It is invoked by first-review and re-review modes; pre-PR mode never calls it.
-
-The ADO Fetcher is a prerequisite for the Doc Context Orchestrator — the Fetcher must complete first because its output (work-item IDs) is the input the Doc Context Orchestrator needs. Once the Fetcher returns, the Doc Context Orchestrator and review aspect agents launch concurrently with each other.
-
-## Acceptance criteria
-
-- [x] The agent accepts PR URL components (org URL, project, PR ID) and returns a structured context block
-- [x] The context block includes PR metadata, latest iteration ID, latest commit SHA, changed files list, and raw diff
-- [x] The context block includes the work-item IDs linked to the PR (empty list if none)
-- [x] The agent handles the case where no iterations are returned (defaults gracefully)
-- [x] The agent handles PRs that are already merged (continues without error)
-- [x] The agent contains no write operations — it is purely a read agent
-
-## Blocked by
-
-None — can start immediately.
-
----
-
-## Agent Brief
-
-> _This was generated by AI during triage._
-
-**Category:** enhancement
-**Summary:** Create the `pr-review:ado-fetcher` agent that encapsulates all ADO read operations for a PR review.
-
-**Current behavior:**
-ADO read operations (PR metadata, iterations, changed files, diff, work-item IDs) are scattered as inline `az devops invoke` shell commands across multiple steps of the `review-pr` command. There is no dedicated agent for this concern.
-
-**Desired behavior:**
-A new plugin agent (`pr-review:ado-fetcher`) accepts PR URL components and returns a single structured context block. All other agents and the orchestrator consume this block rather than making their own ADO calls. The agent is purely read-only — it performs no write operations.
-
-**Key interfaces:**
-
-- Input: org URL, project, PR ID, optional prior iteration ID (passed in for re-review)
-- Output: structured context block — PR metadata, latest iteration ID, latest commit SHA, changed files list, raw diff, work-item IDs list
-- The agent must handle zero-iteration PRs and already-merged PRs gracefully
-
-**Acceptance criteria:**
-
-- [x] The agent accepts PR URL components and returns a structured context block
-- [x] The context block includes PR metadata, latest iteration ID, latest commit SHA, changed files list, and raw diff
-- [x] The context block includes the work-item IDs linked to the PR (empty list if none)
-- [x] The agent handles the case where no iterations are returned (defaults gracefully)
-- [x] The agent handles PRs that are already merged (continues without error)
-- [x] The agent contains no write operations — it is purely a read agent
-
-**Out of scope:**
-
-- Any ADO write operations
-- Doc Context fetching (that stays with the Doc Context Orchestrator)
-- GitHub or GitLab platform support
diff --git a/docs/issues/pr-review-orchestrator-split/02-create-ado-writer-agent.md b/docs/issues/pr-review-orchestrator-split/02-create-ado-writer-agent.md
deleted file mode 100644
index ac545a3a..00000000
--- a/docs/issues/pr-review-orchestrator-split/02-create-ado-writer-agent.md
+++ /dev/null
@@ -1,72 +0,0 @@
-# Create ADO Writer agent
-
-**Status:** resolved
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-orchestrator-split/PRD.md`
-
-## What to build
-
-Create a new plugin agent (`pr-review:ado-writer`) that encapsulates all Azure DevOps write-back operations for a PR review. The agent receives: PR context (org URL, project, repo ID, PR ID, latest iteration ID, summary thread ID), a list of compact findings, and a mode flag (first-review or re-review).
-
-For each finding it posts a new Inline Comment thread to ADO. After all findings are posted it posts the Review Summary on first-review, or a delta reply to the existing summary thread on re-review. As its final action it posts the completion marker reply to the summary thread.
-
-The compact finding schema the agent accepts: `{ severity, filePath, startLine, endLine, title, body }`. Every comment posted must end with the canonical Bot Signature trailer `---\n🤖 *Reviewed by Claude Code* — Iteration N`.
-
-This agent is used by both first-review and re-review modes. It is not invoked in pre-PR mode.
-
-## Acceptance criteria
-
-- [x] The agent posts one Inline Comment thread per finding at the correct file path and line range
-- [x] Each posted comment ends with the canonical Bot Signature including the iteration number
-- [x] On first-review, the agent posts a full Review Summary as a new general thread
-- [x] On re-review with at least one new finding, the agent posts a delta reply to the existing summary thread
-- [x] On re-review with zero new findings, the agent skips the summary reply
-- [x] The agent posts a completion marker reply (`✅ Review complete — Iteration N`) to the summary thread as its final action
-- [x] If `threadContext` is rejected by ADO (file not in diff), the agent retries without `threadContext` (general comment fallback)
-- [x] The agent returns the final `SUMMARY_THREAD_ID` and `FINDINGS_POSTED` count to the caller
-
-## Blocked by
-
-None — can start immediately.
-
----
-
-## Agent Brief
-
-> _This was generated by AI during triage._
-
-**Category:** enhancement
-**Summary:** Create the `pr-review:ado-writer` agent that encapsulates all ADO write-back operations for a PR review.
-
-**Current behavior:**
-ADO write operations (posting Inline Comment threads, patching thread status, posting the Review Summary, posting the completion marker) are inline shell commands in `review-pr.md`. There is no dedicated agent for write-back.
-
-**Desired behavior:**
-A new plugin agent (`pr-review:ado-writer`) receives a PR context block, a compact findings list, a mode flag, and an optional existing summary thread ID. It posts all Inline Comments, the Review Summary (or delta reply on re-review), and the completion marker. Every comment ends with the canonical Bot Signature.
-
-**Key interfaces:**
-
-- Input: PR context (org URL, project, repo ID, PR ID, latest iteration ID, summary thread ID), findings list as `{ severity, filePath, startLine, endLine, title, body }[]`, mode (`first-review` | `re-review`)
-- Output: `{ summaryThreadId, findingsPosted }` returned to the caller
-- Bot Signature constant: `🤖 *Reviewed by Claude Code*` prefix — must not change
-- On `threadContext` rejection by ADO, retries without `threadContext` (general comment fallback)
-
-**Acceptance criteria:**
-
-- [ ] The agent posts one Inline Comment thread per finding at the correct file path and line range
-- [ ] Each posted comment ends with the canonical Bot Signature including the iteration number
-- [ ] On first-review, the agent posts a full Review Summary as a new general thread
-- [ ] On re-review with at least one new finding, the agent posts a delta reply to the existing summary thread
-- [ ] On re-review with zero new findings, the agent skips the summary reply
-- [ ] The agent posts a completion marker reply to the summary thread as its final action
-- [ ] If `threadContext` is rejected by ADO, the agent retries without `threadContext`
-- [ ] The agent returns the final `summaryThreadId` and `findingsPosted` count
-
-**Out of scope:**
-
-- Reply posting for re-review classified threads (that is the Re-review Coordinator's responsibility)
-- Pre-PR mode (no ADO calls in that mode)
-- Reading or fetching any ADO data
diff --git a/docs/issues/pr-review-orchestrator-split/03-create-re-review-coordinator-agent.md b/docs/issues/pr-review-orchestrator-split/03-create-re-review-coordinator-agent.md
deleted file mode 100644
index 35ed7520..00000000
--- a/docs/issues/pr-review-orchestrator-split/03-create-re-review-coordinator-agent.md
+++ /dev/null
@@ -1,81 +0,0 @@
-# Create Re-review Coordinator agent
-
-**Status:** resolved
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-orchestrator-split/PRD.md`
-
-## What to build
-
-Create a new plugin agent (`pr-review:re-review-coordinator`) that owns the full re-review state machine. The agent receives the ADO Fetcher context block (which includes the raw diff) and the raw full PR threads JSON (the unfiltered ADO thread list). It parses the raw diff into diff hunks internally before calling `classify-thread` — the hunks file is a temp artefact managed inside the agent, not an input from the orchestrator.
-
-It performs in order:
-
-1. Calls the `detect-prior-review` Node.js module to identify prior bot threads and locate the summary thread.
-2. Runs the partial-run check (looks for the completion marker for the prior iteration in the summary thread). Falls back to first-review mode if the marker is absent.
-3. If no new revisions exist since the prior review (prior iteration ID equals latest iteration ID), prints outstanding pending threads to the console and exits early — no ADO writes.
-4. Calls `classify-thread` on each prior thread against the diff hunks.
-5. For each new finding passed in, calls `match-finding` to look for a matching prior thread.
-6. Based on classification, posts replies to prior threads: acknowledges disputes, confirms resolutions (and PATCHes thread status to fixed), adds new evidence to pending threads with new information, skips pending threads with no new evidence, ignores obsolete threads.
-7. Returns the classification counts (addressed, disputed, pending, obsolete), the fresh findings list (`freshFindings[]` — only unmatched findings; matched findings are consumed and not returned), and an `earlyExit` flag. `earlyExit` is `true` only on the no-new-revisions path (step 3); it is `false` on all other paths including normal completion with zero fresh findings.
-
-The four Node.js modules (`detect-prior-review`, `classify-thread`, `match-finding`, `parse-signature`) remain in `scripts/re-review/` unchanged. This agent calls them via `node --input-type=module` inline scripts, exactly as the current `review-pr.md` does.
-
-## Acceptance criteria
-
-- [ ] The agent correctly detects prior bot threads using the `detect-prior-review` module
-- [ ] The agent falls back to first-review mode when no completion marker is found for the prior iteration
-- [ ] The agent exits early (console output only, no ADO writes) when prior and latest iteration IDs are identical, and returns `earlyExit: true`
-- [ ] The agent classifies all prior threads using the `classify-thread` module
-- [ ] The agent matches new findings to prior threads using the `match-finding` module with ±3-line drift tolerance
-- [ ] The agent posts a dispute acknowledgement reply to disputed threads including the ADO nudge
-- [ ] The agent posts a resolution confirmation reply and PATCHes status to fixed for addressed threads
-- [ ] The agent posts a new-evidence reply to pending threads that have new analysis; skips pending threads with no new evidence
-- [ ] The agent returns classification counts (addressed, disputed, pending, obsolete), the unmatched (fresh) findings list, and the `earlyExit` flag
-- [ ] The existing re-review module unit tests (`detect-prior-review`, `classify-thread`, `match-finding`, `parse-signature`) pass unchanged
-
-## Blocked by
-
-None — can start immediately.
-
----
-
-## Agent Brief
-
-> _This was generated by AI during triage._
-
-**Category:** enhancement
-**Summary:** Create the `pr-review:re-review-coordinator` agent that owns the full re-review state machine.
-
-**Current behavior:**
-The re-review state machine (prior thread detection, partial-run check, Thread Classification, finding matching, reply/resolution posting) lives inline in `review-pr.md` across Steps 3.5–10-Path-B. It is loaded on every invocation regardless of mode.
-
-**Desired behavior:**
-A new plugin agent (`pr-review:re-review-coordinator`) receives the ADO Fetcher context block (which includes the raw diff) and the raw full PR threads JSON (unfiltered). It calls `detect-prior-review` internally to identify bot threads, parses the raw diff into diff hunks internally, then runs the full re-review state machine, posts classified replies directly to ADO, and returns classification counts plus the list of unmatched (fresh) findings for the ADO Writer to post as new threads. The four existing Node.js modules (`detect-prior-review`, `classify-thread`, `match-finding`, `parse-signature`) are called from this agent unchanged.
-
-**Key interfaces:**
-
-- Input: ADO Fetcher context block (includes raw diff), raw full PR threads JSON (captured by the orchestrator during mode detection via `az repos pr thread list` — not re-fetched; `detect-prior-review` filters this list inside the Coordinator), new findings list, Bot Signature prefix constant
-- The Coordinator parses the raw diff into diff hunks internally; this is not an orchestrator concern
-- Output: `{ addressed, disputed, pending, obsolete, freshFindings[], earlyExit }` — fresh findings are those with no matching prior thread; `earlyExit: true` signals the no-new-revisions path to the orchestrator
-- The agent calls the four Node.js modules via `node --input-type=module` inline scripts (same pattern as current `review-pr.md`)
-- Early-exit path: when prior iteration ID equals latest iteration ID, prints pending threads to console and returns `{ earlyExit: true, freshFindings: [], addressed: 0, disputed: 0, pending: N, obsolete: 0 }` — all count fields are always present; the orchestrator must skip the ADO Writer entirely when `earlyExit: true`
-
-**Acceptance criteria:**
-
-- [ ] The agent correctly detects prior bot threads using the `detect-prior-review` module
-- [ ] The agent falls back to first-review mode when no completion marker is found for the prior iteration
-- [ ] The agent exits early when prior and latest iteration IDs are identical (console output only, no ADO writes), returning `earlyExit: true`
-- [ ] The agent classifies all prior threads using the `classify-thread` module
-- [ ] The agent matches new findings to prior threads using `match-finding` with ±3-line drift tolerance
-- [ ] The agent posts dispute acknowledgement, resolution confirmation, and new-evidence replies appropriately
-- [ ] The agent returns classification counts (addressed, disputed, pending, obsolete), the unmatched fresh findings list, and the `earlyExit` flag
-- [ ] The four re-review module unit tests pass unchanged after this issue is implemented
-
-**Out of scope:**
-
-- Posting new Inline Comment threads for fresh findings (ADO Writer's responsibility)
-- Posting the Review Summary or completion marker (ADO Writer's responsibility)
-- First-review or pre-PR mode logic
diff --git a/docs/issues/pr-review-orchestrator-split/04-refactor-orchestrator.md b/docs/issues/pr-review-orchestrator-split/04-refactor-orchestrator.md
deleted file mode 100644
index 3a150b76..00000000
--- a/docs/issues/pr-review-orchestrator-split/04-refactor-orchestrator.md
+++ /dev/null
@@ -1,83 +0,0 @@
-# Refactor review-pr.md to thin orchestrator
-
-**Status:** resolved
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-orchestrator-split/PRD.md`
-
-## What to build
-
-Refactor `review-pr.md` into a thin orchestrator of approximately 200 lines. The orchestrator:
-
-1. Validates prerequisites in a mode-aware way: always checks `git` availability and `pr-review-toolkit`; checks Azure CLI and `azure-devops` extension only when a PR URL is present (Pre-PR mode requires no ADO credentials).
-2. Parses `$ARGUMENTS` for a PR URL. If absent, sets mode to Pre-PR; if present, proceeds to detection.
-3. For PR URL cases: makes a lightweight ADO thread-list call directly (not via the ADO Fetcher) using `az repos pr thread list` (not `az devops invoke`) to check for a prior Bot Signature, determining mode and extracting the prior iteration ID if found. The full thread list from this call is captured and passed forward to the Re-review Coordinator in step 6 — no second ADO thread-list call is made.
-4. Logs the detected mode clearly before delegating.
-5. For First-review: invokes the ADO Fetcher agent (passing org URL, project, PR ID), then runs Doc Context Orchestrator + review aspect agents in parallel, collects compact findings, delegates write-back to the ADO Writer agent.
-6. For Re-review: invokes the ADO Fetcher agent (passing org URL, project, PR ID, and prior iteration ID), then runs Doc Context Orchestrator + review aspect agents in parallel. Once all review aspect agents return their findings, passes the complete findings list and prior-thread data to the Re-review Coordinator agent (which handles replies). If the Coordinator returns `earlyExit: true` (no new revisions), the orchestrator stops — ADO Writer is not called. Otherwise passes fresh findings to the ADO Writer agent.
-7. Pre-PR mode is a stub at this slice — it detects the mode and prints a "Pre-PR mode not yet implemented" message. Full Pre-PR behaviour is delivered in issue 05.
-
-The `review-pr.md` file must contain no `az devops invoke` shell commands after this refactor — the three focused agents own all data-fetch and write-back ADO operations. The one allowed inline ADO call is the mode-detection `az repos pr thread list` in step 3, which is an orchestration concern, not a data-fetch or write-back operation. The Bot Signature constants and detection prefix are unchanged. All existing re-review module unit tests must pass.
-
-## Acceptance criteria
-
-- [ ] `review-pr.md` is ≤ 200 lines and contains no `az devops invoke` calls
-- [ ] Prerequisite checks are mode-aware: Azure CLI and `azure-devops` extension are not required in Pre-PR mode (no PR URL provided)
-- [ ] The orchestrator logs the detected mode (Pre-PR / First-review / Re-review) before delegating
-- [ ] First-review mode produces the same ADO comment output as the pre-refactor command (full Review Summary + Inline Comments + completion marker)
-- [ ] Re-review mode produces the same ADO comment output as the pre-refactor command (classified replies + fresh findings + delta summary + completion marker)
-- [ ] Pre-PR mode prints a clear "not yet implemented" message and exits cleanly
-- [ ] The ADO Fetcher and Doc Context Orchestrator still run in the correct order (Fetcher first, then both Doc Context and review agents can overlap)
-- [ ] The Bot Signature format and detection prefix are unchanged
-- [ ] `pnpm test` passes (all re-review module unit tests green)
-- [ ] `pnpm format` produces no diff
-
-## Blocked by
-
-- `docs/issues/pr-review-orchestrator-split/01-create-ado-fetcher-agent.md`
-- `docs/issues/pr-review-orchestrator-split/02-create-ado-writer-agent.md`
-- `docs/issues/pr-review-orchestrator-split/03-create-re-review-coordinator-agent.md`
-
----
-
-## Agent Brief
-
-> _This was generated by AI during triage._
-
-**Category:** enhancement
-**Summary:** Refactor `review-pr.md` into a thin orchestrator (~200 lines) that delegates to the three focused agents.
-
-**Current behavior:**
-`review-pr.md` is ~1000 lines, mixing orchestration logic, ADO shell commands, re-review state machine, and write-back in a single file. Every invocation loads the entire file into context.
-
-**Desired behavior:**
-`review-pr.md` shrinks to ~200 lines containing: mode-aware prerequisite validation (ADO tooling skipped for Pre-PR), argument parsing, mode detection (Pre-PR / First-review / Re-review), and delegation calls to the ADO Fetcher, Re-review Coordinator, and ADO Writer agents. The file contains no `az devops invoke` calls. Pre-PR mode is a stub that prints "not yet implemented" — full implementation is in issue 05.
-
-**Key interfaces:**
-
-- Mode detection sequence: no URL → Pre-PR; URL → orchestrator calls `az repos pr thread list` (not `az devops invoke`) → no Bot Signature → First-review; Bot Signature found → extract prior iteration ID → Re-review
-- Bot Signature detection prefix: `🤖 *Reviewed by Claude Code*` — must not change
-- ADO Fetcher agent invocation: passes org URL, project, PR ID (plus prior iteration ID in re-review)
-- Re-review Coordinator agent invocation (re-review only): called after all review aspect agents complete; passes ADO Fetcher context + full PR threads JSON (captured from mode detection in step 3, not re-fetched) + new findings list; returns `{ earlyExit, freshFindings[], addressed, disputed, pending, obsolete }`
-- If Coordinator returns `earlyExit: true`, orchestrator stops — ADO Writer is not called
-- ADO Writer agent invocation: passes PR context + fresh findings list + mode (`"first-review"` | `"re-review"`); the mode determines whether ADO Writer posts a new Review Summary (first-review) or a delta reply (re-review)
-
-**Acceptance criteria:**
-
-- [ ] `review-pr.md` is ≤ 200 lines and contains no `az devops invoke` calls
-- [ ] Prerequisite checks are mode-aware: Azure CLI and `azure-devops` extension are not required in Pre-PR mode
-- [ ] The orchestrator logs the detected mode before delegating
-- [ ] First-review produces the same ADO output as pre-refactor
-- [ ] Re-review produces the same ADO output as pre-refactor
-- [ ] Pre-PR mode prints "not yet implemented" and exits cleanly
-- [ ] ADO Fetcher runs before Doc Context Orchestrator (Fetcher provides work-item IDs); Doc Context and review agents may overlap
-- [ ] Bot Signature format and detection prefix unchanged
-- [ ] `pnpm test` passes; `pnpm format` produces no diff
-
-**Out of scope:**
-
-- Full Pre-PR mode implementation (issue 05)
-- Compact sub-agent output guidance (issue 06)
-- Version bump (issue 07)
diff --git a/docs/issues/pr-review-orchestrator-split/05-add-pre-pr-mode.md b/docs/issues/pr-review-orchestrator-split/05-add-pre-pr-mode.md
deleted file mode 100644
index 318c3451..00000000
--- a/docs/issues/pr-review-orchestrator-split/05-add-pre-pr-mode.md
+++ /dev/null
@@ -1,73 +0,0 @@
-# Add Pre-PR mode
-
-**Status:** resolved
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-orchestrator-split/PRD.md`
-
-## What to build
-
-Implement the Pre-PR operating mode in the orchestrator. When `/pr-review:review-pr` is invoked without a PR URL, the command:
-
-1. Diffs the current local branch against its upstream target (e.g. `git diff origin/<default-branch>...HEAD`).
-2. Reads key changed files (same skip-list as today: generated files, serialization YAMLs, etc.).
-3. Launches the same `pr-review-toolkit` review aspect agents as the ADO modes, passing the local diff and file contents. Doc Context is skipped (no work items or Confluence pages to fetch without a PR).
-4. Aggregates findings and presents them in the Claude interface as a structured list (severity, file, line, title, body) — no ADO calls are made.
-5. Prints a clear completion message when done.
-
-No ADO credentials are required and no ADO calls are made in this mode. The pre-PR Review uses the same review aspect agent selection logic as ADO modes (aspect filter from `$ARGUMENTS` applies).
-
-## Acceptance criteria
-
-- [x] Running the command without a URL enters Pre-PR mode with a console message confirming the mode
-- [x] The diff used is the local branch diff against its upstream target
-- [x] Review aspect agents receive the local diff and changed file contents
-- [x] Findings are presented in the Claude interface with severity, file path, line range, title, and body
-- [x] No ADO API calls are made in this mode
-- [x] The aspect filter argument (e.g. `code`, `errors`, `all`) is respected in pre-PR mode
-- [x] `pnpm test` passes; `pnpm format` produces no diff
-
-## Blocked by
-
-- `docs/issues/pr-review-orchestrator-split/04-refactor-orchestrator.md`
-
----
-
-## Agent Brief
-
-> _This was generated by AI during triage._
-
-**Category:** enhancement
-**Summary:** Implement Pre-PR mode — review local branch diff without a PR URL, no ADO write-back.
-
-**Current behavior:**
-The orchestrator stub from issue 04 prints "Pre-PR mode not yet implemented" and exits. No review occurs without a PR URL.
-
-**Desired behavior:**
-When the command is invoked without a URL, it diffs the local branch against its upstream target, runs the same `pr-review-toolkit` review aspect agents as ADO modes, and presents compact structured findings in the Claude interface. No ADO credentials are required or used. Doc Context gathering is skipped (no work items to fetch). The aspect filter argument applies.
-
-**Key interfaces:**
-
-- Diff source: `git diff origin/<default-branch>...HEAD`
-- File skip-list: same as current (generated files, serialization YAMLs, `*.g.cs`, `swagger.md`)
-- Review aspect agents: same selection logic as ADO modes; aspect filter from `$ARGUMENTS` applies
-- Finding presentation: compact structured list — severity, file path, line range, title, body — in the Claude interface
-- No ADO Fetcher, Re-review Coordinator, or ADO Writer agents are invoked
-
-**Acceptance criteria:**
-
-- [ ] Running the command without a URL enters Pre-PR mode with a console message confirming the mode
-- [ ] The diff used is the local branch diff against its upstream target
-- [ ] Review aspect agents receive the local diff and changed file contents
-- [ ] Findings are presented in the Claude interface with severity, file path, line range, title, and body
-- [ ] No ADO API calls are made in this mode
-- [ ] The aspect filter argument is respected
-- [ ] `pnpm test` passes; `pnpm format` produces no diff
-
-**Out of scope:**
-
-- Posting findings to ADO
-- Doc Context gathering in pre-PR mode
-- Any change to first-review or re-review behaviour
diff --git a/docs/issues/pr-review-orchestrator-split/06-compact-subagent-output.md b/docs/issues/pr-review-orchestrator-split/06-compact-subagent-output.md
deleted file mode 100644
index 7eb42921..00000000
--- a/docs/issues/pr-review-orchestrator-split/06-compact-subagent-output.md
+++ /dev/null
@@ -1,65 +0,0 @@
-# Add compact sub-agent output guidance to the review-agent launch step
-
-**Status:** resolved
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-orchestrator-split/PRD.md`
-
-## What to build
-
-Update the step in the thin orchestrator that launches `pr-review-toolkit` review aspect agents to instruct them to return compact structured findings rather than prose with embedded code quotes. (The thin orchestrator produced by issue 04 will have a different step numbering than the pre-refactor monolith — find the step by its purpose: the one that spawns the parallel review agents.)
-
-The prompt addition instructs each agent to return a JSON array where each element has: `severity` (critical / important / minor), `filePath` (leading `/`, forward slashes), `startLine` (integer), `endLine` (integer), `title` (one line, ≤ 80 chars), `body` (one paragraph — the exact text to post as the ADO or local-interface comment, no code quotes, no repeated context). The reasoning and supporting analysis should stay inside the agent's own context, not appear in the return value.
-
-No changes are made to `pr-review-toolkit` agent definitions — this guidance lives only in the orchestrator's prompt to the agents.
-
-## Acceptance criteria
-
-- [ ] The review-agent launch step explicitly requests structured JSON findings with the six required fields
-- [ ] The prompt instructs agents to omit code quotes and prose reasoning from the return value
-- [ ] The ADO Writer agent correctly receives and processes the structured finding schema
-- [ ] Pre-PR mode findings are also presented using the same structured schema
-- [ ] No `pr-review-toolkit` agent definition files are modified
-- [ ] `pnpm format` produces no diff
-
-## Blocked by
-
-- `docs/issues/pr-review-orchestrator-split/04-refactor-orchestrator.md`
-
----
-
-## Agent Brief
-
-> _This was generated by AI during triage._
-
-**Category:** enhancement
-**Summary:** Update the review-agent launch step in the thin orchestrator to request compact structured findings from review aspect agents.
-
-**Current behavior:**
-Review aspect agents return prose findings with embedded code quotes and explanatory text. The full prose flows back into the parent context as tool call results, contributing to token budget pressure.
-
-**Desired behavior:**
-The step in the thin orchestrator that spawns parallel review agents explicitly instructs each `pr-review-toolkit` review aspect agent to return a JSON array of findings. Locate this step by purpose — it is the one that launches the review agents in parallel — not by number (step numbering changed after the issue 04 refactor). Each element: `severity` (critical / important / minor), `filePath`, `startLine`, `endLine`, `title` (≤ 80 chars), `body` (one paragraph, no code quotes). Reasoning stays inside the agent's own context. No `pr-review-toolkit` agent definition files are modified.
-
-**Key interfaces:**
-
-- The structured finding schema: `{ severity, filePath, startLine, endLine, title, body }`
-- Guidance location: the review-agent launch step in the orchestrator only — not in toolkit agent definitions
-- Both ADO modes and Pre-PR mode use this schema
-
-**Acceptance criteria:**
-
-- [ ] The review-agent launch step requests structured JSON findings with all six required fields
-- [ ] The prompt instructs agents to omit code quotes and prose reasoning from the return value
-- [ ] The ADO Writer agent correctly receives and processes the structured schema
-- [ ] Pre-PR mode findings are presented using the same schema
-- [ ] No `pr-review-toolkit` agent definition files are modified
-- [ ] `pnpm format` produces no diff
-
-**Out of scope:**
-
-- Enforcing schema validation on agent output
-- Changing the toolkit agent definitions
-- Token-budget monitoring or benchmarking
diff --git a/docs/issues/pr-review-orchestrator-split/07-version-bump-and-release.md b/docs/issues/pr-review-orchestrator-split/07-version-bump-and-release.md
deleted file mode 100644
index 44a5b8b5..00000000
--- a/docs/issues/pr-review-orchestrator-split/07-version-bump-and-release.md
+++ /dev/null
@@ -1,63 +0,0 @@
-# Version bump and CHANGELOG
-
-**Status:** resolved
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-orchestrator-split/PRD.md`
-
-## What to build
-
-Bump the `pr-review` plugin version (minor bump — new features added) and add a dated CHANGELOG entry covering the orchestrator split, the three new agents, pre-PR mode, and compact sub-agent output.
-
-Run `pnpm --filter pr-review bump minor` to update both `plugin.json` and `marketplace.json`. Add a `[Unreleased]` → versioned entry to `CHANGELOG.md` following the existing format. Run `pnpm --filter pr-review verify:changelog` to confirm the entry passes validation.
-
-Update `CLAUDE.md` to reflect the new architecture: remove the claim that "the entire behaviour of the plugin lives in `commands/review-pr.md`", add the `.agents/` directory to the repository layout, and update the command conventions section to note that ADO calls now live in the three focused agents (ADO Fetcher, Re-review Coordinator, ADO Writer) rather than inline in the command.
-
-## Acceptance criteria
-
-- [ ] `plugin.json` and `marketplace.json` both reflect the new minor version
-- [ ] `CHANGELOG.md` has a dated entry for the new version describing the orchestrator split, three new agents, pre-PR mode, and compact output guidance
-- [ ] `pnpm --filter pr-review verify:changelog` passes
-- [ ] `CLAUDE.md` updated: "entire behaviour lives in `commands/review-pr.md`" claim removed, `.agents/` directory added to layout, command conventions updated to reflect ADO calls live in the focused agents
-- [ ] `pnpm format` produces no diff
-
-## Blocked by
-
-- `docs/issues/pr-review-orchestrator-split/05-add-pre-pr-mode.md`
-- `docs/issues/pr-review-orchestrator-split/06-compact-subagent-output.md`
-
----
-
-## Agent Brief
-
-> _This was generated by AI during triage._
-
-**Category:** enhancement
-**Summary:** Bump `pr-review` to the next minor version and add a CHANGELOG entry for the orchestrator split.
-
-**Current behavior:**
-`plugin.json` and `marketplace.json` carry the current version. `CHANGELOG.md` has no entry for this feature set.
-
-**Desired behavior:**
-Run `pnpm --filter pr-review bump minor` to update both version files atomically. Add a dated entry to `CHANGELOG.md` under the new version number describing: orchestrator split, three new focused agents (ADO Fetcher, Re-review Coordinator, ADO Writer), pre-PR mode, and compact sub-agent output guidance. Verify with `pnpm --filter pr-review verify:changelog`.
-
-**Key interfaces:**
-
-- `pnpm --filter pr-review bump minor` — updates `plugin.json` and `marketplace.json`
-- `CHANGELOG.md` entry format — must match the existing dated em-dash format enforced by `verify:changelog`
-- `pnpm --filter pr-review verify:changelog` — must pass
-
-**Acceptance criteria:**
-
-- [ ] `plugin.json` and `marketplace.json` both reflect the new minor version
-- [ ] `CHANGELOG.md` has a dated entry for the new version covering all four feature areas
-- [ ] `pnpm --filter pr-review verify:changelog` passes
-- [ ] `CLAUDE.md` updated to reflect the three-agent architecture
-- [ ] `pnpm format` produces no diff
-
-**Out of scope:**
-
-- Publishing to the marketplace (manual step)
-- Creating the git tag (handled by the release workflow on `main`)
diff --git a/docs/issues/pr-review-pre-pr-default-branch-override/01-env-var-override.md b/docs/issues/pr-review-pre-pr-default-branch-override/01-env-var-override.md
deleted file mode 100644
index aec87183..00000000
--- a/docs/issues/pr-review-pre-pr-default-branch-override/01-env-var-override.md
+++ /dev/null
@@ -1,25 +0,0 @@
----
-title: pre-pr: env var override for default-branch fallback chain
-created: 2026-05-14
----
-
-**Status:** needs-triage
-**Category:** enhancement
-**Plugin:** `apps/claude-code/pr-review`
-**Depends on:** orchestrator-split PR merged
-
-## Problem Statement
-
-`detect-default-branch.mjs` uses a Gitflow-aware fallback chain (`remote-show → develop → main → master → none`) when the remote is unreachable. Repos that intentionally use `main` as their integration branch but have a stale `origin/develop` ref will silently diff against `develop` with no escape hatch — the user has no way to override the choice.
-
-## Solution
-
-Add an environment variable override (e.g. `PR_REVIEW_DEFAULT_BRANCH`) that, when set, short-circuits the fallback chain and uses the specified branch directly. A Notice should still be emitted if the override bypasses a `remote-show` that would have returned a different result.
-
-## Acceptance criteria
-
-- `PR_REVIEW_DEFAULT_BRANCH=main` forces the pre-PR diff to use `main` regardless of what `remote-show` or the fallback chain would have returned
-- When the env var is set, the fallback chain is not consulted
-- An info-level Notice is emitted when the env var is used, so the invoker knows the override is active
-- Existing behavior (no env var set) is unchanged
-- Unit tests cover the override path
diff --git a/docs/issues/pr-review-rereview/01-normalize-bot-signature.md b/docs/issues/pr-review-rereview/01-normalize-bot-signature.md
deleted file mode 100644
index 1d473a3a..00000000
--- a/docs/issues/pr-review-rereview/01-normalize-bot-signature.md
+++ /dev/null
@@ -1,30 +0,0 @@
-# Normalize bot signature
-
-**Status:** closed
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-rereview/PRD.md`
-
-## What to build
-
-Update every signature occurrence in `commands/review-pr.md` to a single canonical form: `🤖 *Reviewed by Claude Code* — Iteration N` (asterisk italics, iteration suffix). The detection substring remains `🤖 *Reviewed by Claude Code*` (prefix match) so it stays backwards-compatible with any legacy comments already on live PRs.
-
-Two types of locations must be updated end-to-end:
-
-- **Runtime-emitted** locations (inside JSON payloads written to `/tmp/` in Steps 10 and 11) must emit the full `— Iteration {LATEST_ITERATION_ID}` suffix.
-- **Documentation/example** locations (the summary structure markdown block) must switch from underscore to asterisk italics, using a `{N}` placeholder.
-
-A note is added to the Notes section documenting the prefix/full-form distinction so future editors do not break detection.
-
-## Acceptance criteria
-
-- [ ] `grep -nF '🤖 _Reviewed by Claude Code_' commands/review-pr.md` → 0 matches
-- [ ] `grep -nF '🤖 *Reviewed by Claude Code*' commands/review-pr.md` → matches at every signature location
-- [ ] All runtime-emitted signatures include `— Iteration` followed by a variable reference
-- [ ] Notes section documents the invariant prefix and the full emitted form
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/pr-review-rereview/02-detect-prior-review.md b/docs/issues/pr-review-rereview/02-detect-prior-review.md
deleted file mode 100644
index 85aa13f2..00000000
--- a/docs/issues/pr-review-rereview/02-detect-prior-review.md
+++ /dev/null
@@ -1,37 +0,0 @@
-# Detect prior review + extract parse-signature and detect-prior-review modules
-
-**Status:** closed
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-rereview/PRD.md`
-
-## What to build
-
-Add a new Step 3.5 to `commands/review-pr.md` that fetches all PR comment threads from ADO, identifies bot-authored threads by signature prefix (no `createdBy` identity check), tags the summary thread, and parses the prior iteration ID from the newest bot comment's signature.
-
-Two `scripts/re-review/` modules are extracted alongside the command changes so the logic is unit-testable in isolation:
-
-**`parse-signature.mjs`** — pure function. Given a comment body string, returns `{ iterationId: number }` if the canonical signature suffix `— Iteration N` is present, or `null` for legacy/non-bot comments.
-
-**`detect-prior-review.mjs`** — given the full array of already-fetched ADO thread objects, returns `{ isRereview, priorThreads, summaryThread, priorIterationId }`. Identifies bot threads by prefix match. Tags the summary thread (`isSummaryThread = true`) when a general thread's first comment contains the summary heading. Stores full `{start: {line, offset}, end: {line, offset}}` ranges per thread, not flat line numbers.
-
-Pagination in the command: loop on `continuationToken` until all threads are fetched before calling `detect-prior-review.mjs`.
-
-The command logs: `Detected N prior Claude Code threads — re-review mode ON` (or `OFF`).
-
-Each module ships with a `node:test` test file covering its core scenarios using JSON fixtures.
-
-## Acceptance criteria
-
-- [ ] Step 3.5 runs unconditionally and logs the detection summary line
-- [ ] All threads fetched before detection (pagination loop on `continuationToken`)
-- [ ] `isSummaryThread` set on the correct general thread (or no thread if none exists)
-- [ ] Thread ranges stored as `{start: {line, offset}, end: {line, offset}}`
-- [ ] `parse-signature.mjs` returns `{ iterationId }` for current-format comments and `null` for legacy/non-bot
-- [ ] `pnpm --filter pr-review test` passes
-
-## Blocked by
-
-`docs/issues/pr-review-rereview/01-normalize-bot-signature.md`
diff --git a/docs/issues/pr-review-rereview/03-target-latest-iteration.md b/docs/issues/pr-review-rereview/03-target-latest-iteration.md
deleted file mode 100644
index c07aa736..00000000
--- a/docs/issues/pr-review-rereview/03-target-latest-iteration.md
+++ /dev/null
@@ -1,33 +0,0 @@
-# Target latest PR iteration
-
-**Status:** closed
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-rereview/PRD.md`
-
-## What to build
-
-Replace all hardcoded `iterationId=1` uses in `commands/review-pr.md` with a dynamic lookup of the latest iteration ID. There are two locations: the file-list call (Step 4, `pullRequestIterationChanges`) and the diff step (Step 5). Both must use `LATEST_ITERATION_ID`.
-
-The implementation also resolves `PRIOR_ITERATION_ID` and both commit IDs for use by the incremental diff step:
-
-- Fetch all PR iterations via `pullRequestIterations`; select the highest `id` as `LATEST_ITERATION_ID` using `jq 'max_by(.id)'`.
-- Capture `sourceRefCommit.commitId` for the latest iteration as `LATEST_COMMIT_ID`.
-- When `IS_REREVIEW=true`: use `PRIOR_ITERATION_ID` (parsed from the prior comment's signature by `parse-signature.mjs`; timestamp fallback for legacy comments) to look up that iteration's `sourceRefCommit.commitId` as `PRIOR_COMMIT_ID`.
-- Export `LATEST_COMMIT_ID` and `PRIOR_COMMIT_ID` for the next step.
-
-Update the `CLAUDE.md` rule: remove the `iterationId=1` guidance; replace with a description of the latest-iteration approach and the `PRIOR_ITERATION_ID` fallback.
-
-## Acceptance criteria
-
-- [ ] `grep -n 'iterationId=1' commands/review-pr.md` → 0 matches
-- [ ] Step 4 file list and Step 5 diff both reference `LATEST_ITERATION_ID`
-- [ ] CLI logs the resolved iteration ID on every run
-- [ ] `PRIOR_COMMIT_ID` resolved from signature suffix when available; timestamp fallback for legacy
-- [ ] `CLAUDE.md` rule updated
-
-## Blocked by
-
-`docs/issues/pr-review-rereview/02-detect-prior-review.md`
diff --git a/docs/issues/pr-review-rereview/04-incremental-diff-baseline.md b/docs/issues/pr-review-rereview/04-incremental-diff-baseline.md
deleted file mode 100644
index d4504c2c..00000000
--- a/docs/issues/pr-review-rereview/04-incremental-diff-baseline.md
+++ /dev/null
@@ -1,35 +0,0 @@
-# Incremental diff baseline
-
-**Status:** closed
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-rereview/PRD.md`
-
-## What to build
-
-Branch Step 5 of `commands/review-pr.md` on `IS_REREVIEW` so re-reviews only diff the new commits, not the entire branch.
-
-When `IS_REREVIEW=true`:
-
-- Diff between `PRIOR_COMMIT_ID` and `LATEST_COMMIT_ID` (from the previous issue).
-- If both commit IDs are identical (no new pushes since last review): print `No new commits since last review.`, list all `pending` threads from `PRIOR_THREADS` (file path and line range per thread) to the console, then exit cleanly without proceeding to Steps 6–11.
-- Attempt `git fetch origin {PRIOR_COMMIT_ID}` before diffing; if the fetch fails (force-push / garbage collection), fall back to full diff with a warning that includes both commit IDs: `Warning: prior commit {PRIOR_COMMIT_ID} unreachable; latest commit {LATEST_COMMIT_ID} — falling back to full diff.`
-- If `PRIOR_COMMIT_ID` is null (legacy comment), fall back to full diff with a simpler warning.
-
-Export diff hunk boundaries (file path, start line, end line per hunk) as structured JSON to `$TMPDIR` for consumption by the classify-thread module.
-
-When `IS_REREVIEW=false`: behaviour is unchanged from before this issue.
-
-## Acceptance criteria
-
-- [ ] First-time review diff behaviour unchanged
-- [ ] Re-review diffs only between `PRIOR_COMMIT_ID` and `LATEST_COMMIT_ID`
-- [ ] Early exit on identical commits: pending threads listed to console; no ADO writes
-- [ ] Force-push fallback warning includes both commit IDs
-- [ ] Diff hunk boundaries exported as structured JSON to `$TMPDIR`
-
-## Blocked by
-
-`docs/issues/pr-review-rereview/03-target-latest-iteration.md`
diff --git a/docs/issues/pr-review-rereview/05-classify-existing-threads.md b/docs/issues/pr-review-rereview/05-classify-existing-threads.md
deleted file mode 100644
index 95b2a671..00000000
--- a/docs/issues/pr-review-rereview/05-classify-existing-threads.md
+++ /dev/null
@@ -1,39 +0,0 @@
-# Classify existing threads + extract classify-thread module
-
-**Status:** closed
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-rereview/PRD.md`
-
-## What to build
-
-Add thread classification to the re-review flow and extract the logic into a standalone `scripts/re-review/classify-thread.mjs` module that the command calls via Bash.
-
-**`classify-thread.mjs`** — pure function. Given a prior thread object and the diff hunk JSON from the previous step, returns one of four states:
-
-- `addressed` — ADO thread status is one of `fixed` (2), `wontFix` (3), `closed` (4), or `byDesign` (5), **or** status is `active` (1) or `pending` (6) and the thread's line range (`[start.line, end.line]`) intersects a changed hunk (`max(thread.start, hunk.start) ≤ min(thread.end, hunk.end)`). ADO `pending` (6) is treated like `active` — diff intersection is required. Line numbers only — offsets not used in intersection logic.
-- `disputed` — status is `active` and at least one comment in the thread does not contain the signature prefix `🤖 *Reviewed by Claude Code*`. No `createdBy` identity check.
-- `pending` — status is `active` and no comment lacks the signature prefix.
-- `obsolete` — the thread's `filePath` is not present in the diff at all.
-
-The summary thread (`isSummaryThread = true`) is skipped — classification does not apply to it.
-
-After classifying all threads, the command prints a one-line summary: `Threads: N addressed, N disputed, N pending, N obsolete`.
-
-The module ships with a `node:test` test file covering all four states, multi-line range intersection, general threads (filePath null), and ADO status codes.
-
-## Acceptance criteria
-
-- [ ] Every non-summary prior thread receives exactly one classification
-- [ ] Summary thread skipped
-- [ ] Classification summary line printed before Step 6
-- [ ] `classify-thread.mjs` returns correct state for all four scenarios
-- [ ] ADO status codes 2–5 (fixed, wontFix, closed, byDesign) auto-map to `addressed`; status 6 (ADO pending) requires diff intersection like `active`
-- [ ] Human reply detection uses signature prefix only — no identity check
-- [ ] `pnpm --filter pr-review test` passes
-
-## Blocked by
-
-`docs/issues/pr-review-rereview/02-detect-prior-review.md`
diff --git a/docs/issues/pr-review-rereview/06-reply-to-threads.md b/docs/issues/pr-review-rereview/06-reply-to-threads.md
deleted file mode 100644
index 71efb80d..00000000
--- a/docs/issues/pr-review-rereview/06-reply-to-threads.md
+++ /dev/null
@@ -1,49 +0,0 @@
-# Reply to threads + extract match-finding module + completion marker
-
-**Status:** closed
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-rereview/PRD.md`
-
-## What to build
-
-Branch Step 10 of `commands/review-pr.md` on `IS_REREVIEW` so re-reviews reply to existing threads instead of creating duplicates. Extract the matching logic into `scripts/re-review/match-finding.mjs`.
-
-**`match-finding.mjs`** — pure function. Given a finding (file path + line range) and the list of prior threads, returns the best-matching prior thread or null. Matching requires file path equality and line-range overlap (`max(finding.start, thread.start) ≤ min(finding.end, thread.end)`) with a ±3 line drift tolerance applied to both endpoints before the overlap test.
-
-**Reply actions per classification:**
-
-| Classification           | Action                                                                                                                         |
-| ------------------------ | ------------------------------------------------------------------------------------------------------------------------------ |
-| `pending` (unchanged)    | Skip — do not post                                                                                                             |
-| `pending` (new evidence) | Reply via `pullRequestThreadComments` with new evidence only                                                                   |
-| `disputed`               | Reply acknowledging the author's point; include nudge to mark thread fixed in ADO                                              |
-| `addressed`              | Reply `Resolved as of Iteration N — thanks!`; PATCH thread status to `fixed` via `pullRequestThreads` (body `{ "status": 2 }`) |
-| `obsolete`               | No action                                                                                                                      |
-
-New findings (no prior match): create a fresh thread as before; signature suffix carries the iteration number — no separate "Iteration N" header line needed.
-
-**`pending` general threads** (no filePath, not the summary thread): skip — same rule as inline pending.
-
-**Completion marker**: after all posts, post one final reply to the summary thread: `✅ Review complete — Iteration {LATEST_ITERATION_ID} ({N} findings posted)`. This is the last action of every successful run. Absence of this marker for the current iteration signals a partial prior run; on the next run, treat the current iteration as first-review mode.
-
-**PATCH concurrency**: if status PATCH returns 409 (author resolved mid-run), log and continue.
-
-## Acceptance criteria
-
-- [ ] No duplicate thread created when a matching prior thread exists
-- [ ] All replies carry the canonical signature on their last line
-- [ ] `addressed` threads PATCH status via `pullRequestThreads` (not `pullRequestThreadComments`)
-- [ ] `disputed` replies include ADO nudge text
-- [ ] `pending` threads skipped uniformly (inline and general)
-- [ ] Completion marker is the final comment on every successful run
-- [ ] Partial prior run detected on next run (no completion marker → first-review mode)
-- [ ] `match-finding.mjs` returns correct match/null for overlap, drift, and no-match scenarios
-- [ ] `pnpm --filter pr-review test` passes
-
-## Blocked by
-
-- `docs/issues/pr-review-rereview/04-incremental-diff-baseline.md`
-- `docs/issues/pr-review-rereview/05-classify-existing-threads.md`
diff --git a/docs/issues/pr-review-rereview/07-summary-comment-policy.md b/docs/issues/pr-review-rereview/07-summary-comment-policy.md
deleted file mode 100644
index 4a99b889..00000000
--- a/docs/issues/pr-review-rereview/07-summary-comment-policy.md
+++ /dev/null
@@ -1,36 +0,0 @@
-# Summary comment policy on re-review
-
-**Status:** closed
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-rereview/PRD.md`
-
-## What to build
-
-Branch Step 11 of `commands/review-pr.md` so re-reviews never post a second full-length summary. Instead, when there is something to report, a compact delta reply is posted to the existing summary thread.
-
-**Behaviour by case:**
-
-- `IS_REREVIEW=false`: full summary posted in a new thread — unchanged from before.
-- `IS_REREVIEW=true` and at least one of {new threads created, addressed replies, disputed replies}: post a reply to the existing summary thread (identified by `SUMMARY_THREAD_ID` from the detection step) with:
-  - Counts only: `X new findings, Y resolved, Z disputed, W pending`
-  - Bullet list of new finding titles linked to their threads
-  - No prose, no section headings beyond the title line
-- `IS_REREVIEW=true` and nothing changed: post nothing at the summary level.
-- Prior summary thread not found (deleted by human): fall back to first-review mode and post a new full summary.
-
-Track the action counters (new threads, addressed replies, disputed replies) during Step 10 and feed them into Step 11. The delta reply uses `pullRequestThreadComments` against `SUMMARY_THREAD_ID`.
-
-## Acceptance criteria
-
-- [ ] First-time review posts full summary in a new thread (regression check)
-- [ ] Re-review with activity posts a delta reply to the existing summary thread — not a new thread
-- [ ] Re-review with no activity posts nothing
-- [ ] Delta reply contains counts and bullet list only — no prose paragraphs
-- [ ] Fallback to full summary when prior summary thread is absent
-
-## Blocked by
-
-`docs/issues/pr-review-rereview/06-reply-to-threads.md`
diff --git a/docs/issues/pr-review-rereview/08-test-fixture-suite.md b/docs/issues/pr-review-rereview/08-test-fixture-suite.md
deleted file mode 100644
index cff6e5e8..00000000
--- a/docs/issues/pr-review-rereview/08-test-fixture-suite.md
+++ /dev/null
@@ -1,52 +0,0 @@
-# Complete test fixture suite
-
-**Status:** closed
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-rereview/PRD.md`
-
-## What to build
-
-Author the complete set of JSON fixture files and finalize the `node:test` suite so all four extracted modules are covered across all eleven fixture scenarios. Earlier issues include inline unit tests for each module's happy path; this issue completes the edge-case coverage and ensures all scenarios are represented as named fixtures.
-
-**Fixture files to create under `tests/fixtures/`:**
-
-| File                            | Scenario                                                        |
-| ------------------------------- | --------------------------------------------------------------- |
-| `threads-fresh-pr.json`         | No prior bot threads                                            |
-| `threads-pending.json`          | Bot threads, active status, no human replies                    |
-| `threads-disputed.json`         | Bot threads with human replies                                  |
-| `threads-addressed-status.json` | Bot threads with ADO status `fixed`                             |
-| `threads-addressed-diff.json`   | Bot threads, active status, line range in diff hunk             |
-| `threads-obsolete.json`         | Bot threads on file not present in diff                         |
-| `threads-partial-run.json`      | Bot threads present, no completion marker for current iteration |
-| `threads-paginated-p1.json`     | First page (100 threads + `continuationToken`)                  |
-| `threads-paginated-p2.json`     | Second page (remaining threads, no token)                       |
-| `diff-hunks-no-change.json`     | Empty hunk set (identical commits)                              |
-| `diff-hunks-with-changes.json`  | Hunks covering a known line range                               |
-
-Fixture JSON shapes must match the ADO `pullRequestThreads` API response format (including `threadContext`, `comments`, `status`, `id` fields).
-
-**Test coverage to complete or add:**
-
-- `detect-prior-review.test.mjs`: paginated scenario (p1 + p2 combined), partial-run detection, summary thread tagging.
-- `classify-thread.test.mjs`: multi-line range intersection, general thread edge cases, all ADO non-active status codes.
-- `match-finding.test.mjs`: outside drift tolerance, multi-line finding vs single-line thread, no-match cases.
-- `parse-signature.test.mjs`: iteration number boundary values, malformed suffix.
-
-No test may import `az` or make any network calls.
-
-## Acceptance criteria
-
-- [ ] All 11 fixture files present under `tests/fixtures/`
-- [ ] `pnpm --filter pr-review test` passes with zero failures
-- [ ] Every fixture scenario exercised by at least one test assertion
-- [ ] Deleting any single fixture file causes at least one test to fail with a clear message
-
-## Blocked by
-
-- `docs/issues/pr-review-rereview/02-detect-prior-review.md`
-- `docs/issues/pr-review-rereview/05-classify-existing-threads.md`
-- `docs/issues/pr-review-rereview/06-reply-to-threads.md`
diff --git a/docs/issues/pr-review-rereview/09-version-bump-and-release.md b/docs/issues/pr-review-rereview/09-version-bump-and-release.md
deleted file mode 100644
index b178d894..00000000
--- a/docs/issues/pr-review-rereview/09-version-bump-and-release.md
+++ /dev/null
@@ -1,43 +0,0 @@
-# Version bump, README, CLAUDE.md, ADR 0009
-
-**Status:** closed
-**Category:** enhancement
-
-## Parent
-
-`docs/issues/pr-review-rereview/PRD.md`
-
-## What to build
-
-Finalise metadata, documentation, and release artefacts for the re-review feature. This is a HITL issue: the bump script and release PR require human review before merging.
-
-**Steps (in order):**
-
-1. **Add CHANGELOG entries.** Under `## [Unreleased]` in `CHANGELOG.md`, add bullets under `### Added` describing the re-review feature. Do this before running the bump script — the bump will fail verification if the versioned section only has `(none)` placeholders.
-
-2. **Check `.prettierignore`.** At the monorepo root, verify `**/CHANGELOG.md` is present (added during spec 00 preflight). If missing, re-run spec 00 before continuing. Prettier must not reformat CHANGELOG files.
-
-3. **Bump the version.** Run `pnpm --filter pr-review bump minor`. This updates both `.claude-plugin/plugin.json` and `marketplace.json` atomically. Never hand-edit `marketplace.json`.
-
-4. **Update `CLAUDE.md`.** Remove the roadmap line about re-review. Verify that the signature-prefix and latest-iteration rules added by earlier issues are present.
-
-5. **Add Re-review section to `README.md`** covering: trigger condition, what changes (detection, thread reuse, delta summary reply, completion marker), new signature format, and known limitations (force-push fallback, partial-run recovery).
-
-6. **Verify `apps/claude-code/pr-review/docs/adr/0009-summary-delta-as-reply.md` exists** (created in spec 00 preflight). If missing, re-run spec 00 before continuing. Do not create a duplicate.
-
-7. **Verify.** Run `pnpm --filter pr-review verify:changelog` locally before opening the PR.
-
-## Acceptance criteria
-
-- [ ] `pnpm --filter pr-review verify:changelog` passes locally
-- [ ] Versions match across `plugin.json` and `marketplace.json`
-- [ ] `README.md` includes a Re-review section
-- [ ] Roadmap line removed from `CLAUDE.md`; signature and iteration rules present
-- [ ] `apps/claude-code/pr-review/docs/adr/0009-summary-delta-as-reply.md` exists, references ADR 0007, status Accepted
-- [ ] Monorepo `.prettierignore` contains `**/CHANGELOG.md`
-- [ ] CI passes on the release PR
-
-## Blocked by
-
-- `docs/issues/pr-review-rereview/07-summary-comment-policy.md`
-- `docs/issues/pr-review-rereview/08-test-fixture-suite.md`
diff --git a/docs/issues/unic-archon-dlc-dogfood-banner/PRD.md b/docs/issues/unic-archon-dlc-dogfood-banner/PRD.md
new file mode 100644
index 00000000..02c2a58e
--- /dev/null
+++ b/docs/issues/unic-archon-dlc-dogfood-banner/PRD.md
@@ -0,0 +1,163 @@
+---
+title: unic-archon-dlc — banner auto-generated dogfood output
+created: 2026-05-23
+---
+
+**Status:** ready-for-agent
+**Plugin:** `apps/claude-code/unic-archon-dlc`
+**Category:** enhancement
+
+## Problem Statement
+
+This monorepo dogfoods the `unic-archon-dlc` Plugin against itself: someone has already run its setup on this repo, so `docs/agents/issue-tracker.md`, `docs/agents/labels.md`, `docs/agents/branching.md`, `docs/agents/domain.md`, `docs/agents/workflow.md`, and a marker-delimited block inside the root `AGENTS.md` are auto-generated outputs of `apps/claude-code/unic-archon-dlc/lib/agent-docs-writer.mjs`.
+
+Nothing in those files signals that:
+
+1. They are auto-generated and will be overwritten on the next setup run — a contributor who edits them by hand loses their work.
+2. They describe the **target** `unic-archon-dlc` workflow (seven `unic-dlc-*` phases, `docs/workflow/<slug>/` artifacts), not the current practice in this repo (Matt Pocock's skills + manual `/tdd`).
+3. The HTML-comment markers in `AGENTS.md` (`<!-- unic-archon-dlc:begin -->` / `<!-- unic-archon-dlc:end -->`) wrap content that is owned by the harness and refreshed on every install.
+
+The result is a Strange Loop: this repo's own consumer-template documentation reads as canonical project guidance, and there is no visible cue separating "what the harness installs into a Consumer" from "what this repo actually does today."
+
+## Solution
+
+The Plugin gains a small `dogfood-banner` module that owns two banner strings. `agent-docs-writer.mjs` consumes those strings so every file it generates carries a visible, regeneration-safe banner. The banner content names the file as auto-generated, points the reader at the canonical regeneration command, and clarifies that the file describes the unic-archon-dlc target workflow rather than current local practice.
+
+The banner is applied in two places:
+
+- **`docs/agents/*.md` (five files):** banner prepended to the file body, so it appears at the very top of every generated doc.
+- **`AGENTS.md` auto-managed block:** banner inserted as the first line _inside_ the marker region, so it lives alongside the auto-managed link list and is refreshed in lockstep.
+
+Because both write paths overwrite their target on every run, the banner is automatically refreshed on every setup — no idempotence trickery required beyond what already exists.
+
+After the Plugin change ships, the existing five `docs/agents/*.md` files in this repo and the AGENTS.md block are updated so the dogfood state matches the new behaviour.
+
+## User Stories
+
+1. As a contributor opening `docs/agents/workflow.md` for the first time, I want to immediately see that the file is auto-generated so that I do not waste time editing it by hand.
+2. As a contributor about to edit one of the `docs/agents/*.md` files, I want the banner to tell me which command regenerates the file so that I know where to make the change instead.
+3. As a contributor reading `docs/agents/workflow.md`, I want the banner to tell me that the seven `unic-dlc-*` phases describe the target lifecycle and not the current driver, so that I do not assume `/unic-dlc-build` is available in this repo today.
+4. As a contributor reading `AGENTS.md`, I want the auto-managed block to announce itself in plain text — not just via easy-to-miss HTML comments — so that I do not hand-edit content that will be overwritten.
+5. As a maintainer running the unic-archon-dlc setup against this repo, I want the banner to be re-applied on every run so that the warning never drifts away as the generated content evolves.
+6. As a maintainer changing the banner wording, I want the banner constants to live in a single module so that I do not have to hunt across template builders.
+7. As a maintainer adding a new generated `docs/agents/*.md` file in the future, I want the banner prefix mechanism to be applied uniformly so that I do not have to remember to copy-paste banner text into a new template.
+8. As a Consumer installing `unic-archon-dlc` for the first time, I want the banner to appear on my generated files too, so that my own team is warned not to hand-edit them.
+9. As a Consumer who has unic-archon-dlc set up but is still using interim tools, I want the banner to make explicit that the generated docs describe the unic-archon-dlc workflow so that my contributors do not assume the workflow is already operational locally.
+10. As a maintainer running `pnpm --filter unic-archon-dlc test`, I want a dedicated test for the banner module so that wording changes are deliberate and reviewed.
+11. As a maintainer running `pnpm --filter unic-archon-dlc test`, I want the existing `docs/agents` and `AGENTS.md` install tests to assert banner presence so that a regression in template wiring fails CI rather than slipping through.
+12. As a maintainer verifying the change locally on this repo, I want to either re-run setup or apply the banner manually to bring the dogfood state into sync with the new behaviour, so that there is no in-tree inconsistency after the Plugin Release ships.
+
+## Implementation Decisions
+
+### New deep module: `lib/dogfood-banner.mjs`
+
+A small, pure module that owns the dogfood banner strings and a formatting helper. Lives next to `agent-docs-writer.mjs` so the relationship is obvious.
+
+Exports:
+
+- `AGENT_DOC_BANNER` — the HTML-comment banner prepended to each `docs/agents/*.md` file. Content names the file as auto-generated, the source module, the regeneration command, and the workflow-vs-current-practice clarification.
+- `SKILLS_BLOCK_BANNER` — a single-line banner inserted as the first line inside the `<!-- unic-archon-dlc:begin -->` / `<!-- unic-archon-dlc:end -->` block of `AGENTS.md`. Distinct from `AGENT_DOC_BANNER` because the markdown surrounding it differs (a Markdown link list vs. a full file body).
+- `prependBanner(banner, body)` — pure helper that joins a banner to a body with consistent spacing (banner + blank line + body), normalising trailing newlines. Used by each template builder.
+
+Module size target: under 60 lines. No I/O, no `node:fs`, no `node:path`. Pure string manipulation only.
+
+### Modified module: `lib/agent-docs-writer.mjs`
+
+- Imports `AGENT_DOC_BANNER`, `SKILLS_BLOCK_BANNER`, and `prependBanner` from `dogfood-banner.mjs`.
+- Each of the five `build*Doc()` template builders returns `prependBanner(AGENT_DOC_BANNER, <existing body>)` instead of the raw body.
+- `updateAgentSkillsBlock()` is updated so the rendered block becomes:
+
+  ```
+  <!-- unic-archon-dlc:begin -->
+  {SKILLS_BLOCK_BANNER}
+
+  {AGENT_SKILLS_LINKS}
+  <!-- unic-archon-dlc:end -->
+  ```
+
+  The marker-delimited replacement logic is unchanged — only the inner body is augmented.
+
+No other lib modules are touched. `install-runner.mjs`, `tracker-adapter.mjs`, `labels-config.mjs`, and the rest of the Plugin remain as-is.
+
+### Banner content (canonical wording — owned by this PRD)
+
+`AGENT_DOC_BANNER` (full-file prefix, HTML comment so it stays out of rendered prose):
+
+```
+<!--
+  AUTO-GENERATED by unic-archon-dlc — DO NOT EDIT BY HAND.
+  Source: apps/claude-code/unic-archon-dlc/lib/agent-docs-writer.mjs
+  Regenerate: re-run `/unic-archon-dlc:setup` (or call `runInstall()` from `lib/install-runner.mjs`).
+  This file describes the unic-archon-dlc target workflow; the current driver
+  in any given Consumer repo may differ.
+-->
+```
+
+`SKILLS_BLOCK_BANNER` (single-line, lives inside the marker block):
+
+```
+<!-- AUTO-GENERATED by unic-archon-dlc — do not edit; re-run `/unic-archon-dlc:setup` to regenerate. -->
+```
+
+The wording is intentionally short on the AGENTS.md block (which is read inline alongside surrounding prose) and longer on the `docs/agents/*.md` prefix (where it stands alone at the top of a fresh file). Both reference the regenerate command by name so a reader can act on the warning without hunting.
+
+### Plugin Release
+
+- Version impact: **patch** — the change affects on-disk output of generated files only; no CLI flag, exit code, or programmatic interface changes.
+- Run `pnpm --filter unic-archon-dlc bump patch` and add a dated CHANGELOG entry under the new version naming the banner addition.
+- After the Plugin Release ships, the existing five `docs/agents/*.md` files and the AGENTS.md auto-managed block in this monorepo are updated so the dogfood state matches the new behaviour. Either re-run setup against the repo, or hand-apply the banner to the current files — both routes leave the repo in the same end state. Implementation can choose whichever is simpler.
+
+### Out-of-tree side-effects
+
+None. No changes to commands, hooks, agents, workflows, or `.archon/` artifacts. No new runtime dependencies.
+
+## Testing Decisions
+
+A good test for this change asserts the **observable output** of generated files and the AGENTS.md block — not the internal call graph of the template builders. We care that contributors and Consumers see the banner; we do not care how each template builder concatenates strings.
+
+Three test files are in scope:
+
+### New: `test/dogfood-banner.test.mjs`
+
+Tests the new module in isolation:
+
+- `AGENT_DOC_BANNER` contains the required signal phrases: "AUTO-GENERATED", the source module reference, and the regenerate hint.
+- `SKILLS_BLOCK_BANNER` contains "AUTO-GENERATED" and the regenerate hint.
+- `prependBanner(banner, body)` produces banner + blank line + body, regardless of whether `body` already ends with a newline.
+- `prependBanner(banner, '')` returns just the banner (with no trailing junk).
+
+These are pure-function assertions — no I/O, no temp dirs.
+
+### Modified: `test/install-agent-docs.test.mjs`
+
+The existing tests verify file presence and key content fragments. Extend each test so that, in addition to its current assertions:
+
+- Each of the five generated files begins with the `AGENT_DOC_BANNER` string.
+- The banner appears exactly once per file (no duplication on a second `writeAgentDocs` call against the same project dir).
+
+### Modified: `test/install-claude-md.test.mjs`
+
+The existing tests verify the marker block is written, refreshed idempotently, and does not destroy surrounding content. Extend so that:
+
+- After `updateAgentSkillsBlock()`, the content between the begin and end markers contains `SKILLS_BLOCK_BANNER`.
+- After running `updateAgentSkillsBlock()` three times, the banner appears exactly once inside the block (matches the existing idempotence test pattern).
+- Surrounding non-block content is still preserved.
+
+### Prior art
+
+The existing test files under `apps/claude-code/unic-archon-dlc/test/` use `node:test`, `assert/strict`, and temp directories created under `os.tmpdir()`. The new banner test follows the same patterns but does not need a temp dir because the module is pure.
+
+## Out of Scope
+
+- Relocating any `docs/agents/*.md` file out of `docs/agents/` — the dogfooding is intentional; this PRD adds visibility only.
+- Changing the contents of `docs/agents/*.md` beyond the prepended banner.
+- Anything inside `.archon/` (config, workflows, commands).
+- ADR reshelving from `docs/adr/` into `apps/claude-code/unic-archon-dlc/docs/adr/` — tracked separately.
+- Refreshing per-plugin `CLAUDE.md` files for monorepo context — tracked separately.
+- The chore PR sweeping retired `docs/plans/`, `.claude/skills/` symlinks, and root orphans.
+
+## Further Notes
+
+- The `unic-archon-dlc` plugin is not yet operational as a Feature Runner in this repo (per ADR-0030 and ADR-0031, manual `/tdd` remains the current path). The banner makes that gap visible at the doc level, which has educational value even before the runner ships.
+- Future ADRs that record the dogfood-banner decision belong under `apps/claude-code/unic-archon-dlc/docs/adr/`, not the root `docs/adr/` directory.
+- If the banner module turns out to need additional banner kinds in the future (e.g. for generated workflow YAMLs in a Consumer), it is the right home — keep them all in `dogfood-banner.mjs`.
diff --git a/docs/issues/unic-archon-dlc/01-plugin-scaffold-and-tracer-install.md b/docs/issues/unic-archon-dlc/01-plugin-scaffold-and-tracer-install.md
deleted file mode 100644
index 2c2a3b41..00000000
--- a/docs/issues/unic-archon-dlc/01-plugin-scaffold-and-tracer-install.md
+++ /dev/null
@@ -1,45 +0,0 @@
-# Plugin scaffold and tracer install hook
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-Scaffold the `unic-archon-dlc` plugin at `apps/claude-code/unic-archon-dlc/` with a working install hook that runs end-to-end from `claude /plugin install` and proves the wiring. This is the tracer slice: minimal but complete path through manifest → install hook → on-disk artifacts.
-
-In scope:
-
-- Plugin directory under `apps/claude-code/unic-archon-dlc/` following the same shape as `auto-format`, `pr-review`, `unic-confluence`.
-- Manifests: `.claude-plugin/plugin.json` and `.claude-plugin/marketplace.json`. Plugin name is `unic-archon-dlc`. Marketplace entry follows the monorepo convention (no hand edits; subsequent slices use `pnpm --filter ... bump`).
-- `.archon/workflows/` and `.archon/commands/` directories shipped inside the plugin (empty placeholders accepted — populated by later slices).
-- Install hook as an idempotent Node.js script (`.mjs`, ESM, zero external runtime deps) triggered by the Claude Code plugin install mechanism.
-- **Setup explorer** module: reads `git remote`, `CLAUDE.md`, `CONTEXT.md`, `CONTEXT-MAP.md`, `docs/adr/`, existing `.archon/` and returns a structured snapshot of what is already configured. Missing files are reported as absent, not errored.
-- **Archon binary check**: install hook verifies `archon` is on `PATH`. If missing, exits with a clear error referencing the README dependency note. Also reads `archon --version` and warns if the version is on the known-incompatible list (start with an empty list shipped as a JSON constant — populated as drift is observed).
-- **Mandatory question tier only** in this slice: issue tracker (auto-detected from `git remote`; falls back to a prompt), PR strategy (deduced from tracker), branching strategy (Gitflow default, GitHub Flow alternative).
-- Already-configured values are shown back to the user and skippable — re-running the hook never overwrites without consent.
-- **Outputs:**
-  - `.archon/unic-dlc.config.json` (machine-readable) populated with the mandatory tier only.
-  - `docs/agents/issue-tracker.md` (human-readable) describing the chosen tracker, its CLI, and the create/update conventions.
-- Config loader/validator module: `loadConfig(path)` returns a typed config object or a structured error. Mandatory fields validated; unknown keys ignored.
-- Tests (`node:test`, `.mjs`) for config loader (valid parse, missing mandatory fields, unknown keys) and setup explorer (mock project directory snapshot, absent-vs-errored handling).
-
-Out of scope for this slice: label tiers (state/type/priority), e2e command prompt, multi-context detection, remaining `docs/agents/*` files, `CLAUDE.md` block, any workflow YAML or command file contents.
-
-## Acceptance criteria
-
-- [ ] `apps/claude-code/unic-archon-dlc/` exists with valid `.claude-plugin/plugin.json` and `.claude-plugin/marketplace.json` (`pnpm check` and `pnpm typecheck` pass at repo root).
-- [ ] Running the install hook against a fresh project writes `.archon/unic-dlc.config.json` with at minimum `{ tracker, pr_strategy, branching }` populated.
-- [ ] Running the install hook against a fresh project writes `docs/agents/issue-tracker.md` describing the chosen tracker.
-- [ ] Re-running the install hook is idempotent: no overwrites, already-configured values shown and skipped.
-- [ ] Missing `archon` binary produces a clear, actionable error and a non-zero exit.
-- [ ] Setup explorer returns a structured snapshot for a mock project directory; missing files are reported as absent rather than throwing.
-- [ ] `node:test` suite covers config loader (3 cases) and setup explorer (1 case) and passes on macOS, Windows, and Linux paths (use `node:path`).
-- [ ] No external runtime deps added to the plugin (`auto-format`-style bar).
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/unic-archon-dlc/02-full-install-hook-and-agent-docs.md b/docs/issues/unic-archon-dlc/02-full-install-hook-and-agent-docs.md
deleted file mode 100644
index dbdb4dd2..00000000
--- a/docs/issues/unic-archon-dlc/02-full-install-hook-and-agent-docs.md
+++ /dev/null
@@ -1,47 +0,0 @@
-# Full install hook with all config tiers and agent docs
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-Expand the install hook from the tracer slice to cover every configuration tier and write the full `docs/agents/` set. The hook remains idempotent and additive: re-running it fills in only what is missing.
-
-In scope:
-
-- **Skippable tier:** e2e test command. If the user defers, the config records `e2e_command: null` and the hook can fill it in on a later run.
-- **Defaulted tier (no prompt unless `--reconfigure`):** model profile (`balanced`), TDD mode (`on`), Nyquist validation (`on`), slopsquatting gate (`on`).
-- **Label configuration tier (canonical → tracker string mapping):**
-  - **State labels:** `needs-triage` → `needs-info` → `needs-specs` → `ready-for-agent` / `ready-for-human` → `resolved` → `closed` (or `rejected`). Canonical role names match `docs/agents/triage-labels.md` from this repo.
-  - **Type labels:** `feature`, `bug`, `spike`, `tech-debt`, `docs`.
-  - **Priority labels:** `p0`, `p1`, `p2`, `p3`.
-  - Each canonical name is mapped to the actual string used in the project's tracker.
-- **Multi-context detection:** if `CONTEXT-MAP.md` exists at repo root, the hook records `repo_layout: "multi-context"` and discovers per-context `CONTEXT.md` paths; otherwise `"single-context"` with the root `CONTEXT.md` path.
-- **`docs/agents/` outputs** (write or update — never clobber unrelated content):
-  - `issue-tracker.md` — already written in slice 1; expand with backend-specific create/update conventions.
-  - `labels.md` — three-tier taxonomy with canonical-to-string mappings for the chosen tracker.
-  - `branching.md` — chosen strategy, default branch names, PR target conventions.
-  - `domain.md` — single-context vs multi-context layout, CONTEXT.md and ADR locations.
-  - `workflow.md` — maps each of the six workflow phases to its artifact outputs and `docs/workflow/` paths.
-- **`CLAUDE.md` update:** append (or update in place) an `## Agent skills` block listing each `docs/agents/*.md` file with a one-line hook. If the block already exists, refresh its contents idempotently.
-- Reasonable defaults applied automatically when the tracker is local markdown (canonical strings == tracker strings).
-
-Out of scope: workflow YAML or command file contents (later slices).
-
-## Acceptance criteria
-
-- [ ] Fresh install writes all five `docs/agents/*.md` files plus the `## Agent skills` block in `CLAUDE.md`.
-- [ ] `.archon/unic-dlc.config.json` includes all label tiers, e2e command (or `null`), defaulted fields, and `repo_layout`.
-- [ ] Multi-context repo (with `CONTEXT-MAP.md`) is detected and recorded; single-context repo records the root `CONTEXT.md` path.
-- [ ] Re-running the hook after the tracer install adds only missing fields; existing fields are preserved unless `--reconfigure` is passed.
-- [ ] If `CLAUDE.md` already has an `## Agent skills` block, it is refreshed without duplicating sections or destroying neighbouring content.
-- [ ] Each `docs/agents/*.md` matches the canonical-to-tracker mapping in `.archon/unic-dlc.config.json` (no drift between machine and human readable).
-- [ ] Tests cover the canonical-to-tracker mapping for at least the local-markdown and GitHub backends.
-
-## Blocked by
-
-- `docs/issues/unic-archon-dlc/01-plugin-scaffold-and-tracer-install.md`
diff --git a/docs/issues/unic-archon-dlc/03-triage-workflow-and-tracker-adapter.md b/docs/issues/unic-archon-dlc/03-triage-workflow-and-tracker-adapter.md
deleted file mode 100644
index 25fa363f..00000000
--- a/docs/issues/unic-archon-dlc/03-triage-workflow-and-tracker-adapter.md
+++ /dev/null
@@ -1,40 +0,0 @@
-# Triage workflow and tracker adapter
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-The simplest of the six workflows — `triage` — paired with the tracker adapter module that every later workflow will depend on. Builds the foundation for issue I/O across all backends.
-
-In scope:
-
-- **Tracker adapter module:** translates canonical label names to tracker-specific strings and produces create/update CLI command strings for each configured backend. Backends covered in v1: GitHub Issues, Azure DevOps, Jira, local markdown (per the `docs/agents/issue-tracker.md` convention in this repo: file under `docs/issues/<slug>/`, `Status:` line records triage state, comments append to a `## Comments` heading). Deep module — all tracker-specific knowledge encapsulated.
-- **`/unic-dlc-triage` command** (note the `unic-dlc-` prefix — avoids collision with existing `/triage` skills) plus `.archon/workflows/triage.yaml`.
-- **Workflow behaviour:** reads current issue states from the configured tracker, reconciles them against `docs/workflow/ROADMAP.md` (creates it if absent), and produces `HANDOFF.md` capturing:
-  - Current phase (which workflow last ran, when, on which feature slug)
-  - Open issues grouped by triage state
-  - Blockers (issues whose `blocked_by` references are still open)
-  - Recent decisions (latest ADR file IDs)
-- **Standalone invocation:** runnable at any point in the lifecycle without prerequisites.
-- **Reused as final node of `cleanup`:** the same workflow file is referenced from the cleanup DAG in slice 13. Both invocations produce the same `HANDOFF.md`.
-- Tests covering the tracker adapter: for each backend, canonical label input produces the expected CLI command string. Tests also assert that absent state files are treated as "no data" rather than errors.
-
-Out of scope: any auto-promotion of issues across states (PRD explicitly excludes this).
-
-## Acceptance criteria
-
-- [ ] `/unic-dlc-triage` runs end-to-end against a project with the local-markdown tracker and produces a valid `HANDOFF.md`.
-- [ ] Tracker adapter handles all four backends. Each emits a syntactically valid CLI command (or the local-markdown file write) for a canonical-label input.
-- [ ] `docs/workflow/ROADMAP.md` is created on first run and updated on subsequent runs; updates do not clobber human-edited sections (use marker-delimited regions for the auto-generated parts).
-- [ ] `HANDOFF.md` includes all four sections: phase, open issues by state, blockers, recent decisions.
-- [ ] `node:test` covers the tracker adapter (one case per backend) and the HANDOFF generator (golden-file-style assertion on a mock project state).
-- [ ] Command file uses the `unic-dlc-` prefix consistently.
-
-## Blocked by
-
-- `docs/issues/unic-archon-dlc/02-full-install-hook-and-agent-docs.md`
diff --git a/docs/issues/unic-archon-dlc/04-explore-parallel-research-and-synthesize.md b/docs/issues/unic-archon-dlc/04-explore-parallel-research-and-synthesize.md
deleted file mode 100644
index ecff2cca..00000000
--- a/docs/issues/unic-archon-dlc/04-explore-parallel-research-and-synthesize.md
+++ /dev/null
@@ -1,39 +0,0 @@
-# Explore workflow — parallel research and synthesize
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-First half of the `explore` workflow: four parallel research agents converging on a `synthesize` node that produces a coherent brief, written to `docs/workflow/<slug>/findings.md`.
-
-In scope:
-
-- `.archon/workflows/explore.yaml` (will be extended in slice 05 with the prototype phase).
-- `/unic-dlc-explore` command file.
-- **Four parallel research nodes** running simultaneously, each scoped to a single concern:
-  - `research-stack` — current and emerging stack relevant to the feature
-  - `research-features` — comparable existing features in the codebase and prior art
-  - `research-architecture` — how the change fits the existing architecture and ADRs
-  - `research-pitfalls` — known traps, deprecations, security/privacy considerations
-- **`synthesize` node** combining the four research outputs with the user's stated intent into a coherent brief. Pulls in CONTEXT.md and (if multi-context) CONTEXT-MAP.md so the brief uses the project glossary.
-- **`docs/workflow/<slug>/findings.md`** is committed to the repo. The slug is taken from the user-supplied feature name; the hook creates the directory if absent.
-- **Optional input from `explore` is honoured by `plan`:** `plan` works without findings, but consumes them when present (the contract is "findings.md may or may not exist" — slice 06's `specs` node handles both cases).
-
-Out of scope (covered in slice 05): prototype node, VALIDATED/INVALIDATED/PARTIAL verdicts, spike-branch gate, spike ticket creation.
-
-## Acceptance criteria
-
-- [ ] `/unic-dlc-explore <slug>` runs all four research nodes in parallel (verifiable from the Archon execution log or YAML structure).
-- [ ] `synthesize` produces `docs/workflow/<slug>/findings.md` containing sections for each research dimension plus the integrated brief.
-- [ ] `findings.md` references CONTEXT.md vocabulary and at least one existing ADR by ID when one is relevant.
-- [ ] Running `explore` without an existing `docs/workflow/<slug>/` directory creates it; running with an existing directory does not destroy prior contents.
-- [ ] Command file uses the `unic-dlc-` prefix.
-
-## Blocked by
-
-- `docs/issues/unic-archon-dlc/02-full-install-hook-and-agent-docs.md`
diff --git a/docs/issues/unic-archon-dlc/05-explore-prototype-and-spike-gate.md b/docs/issues/unic-archon-dlc/05-explore-prototype-and-spike-gate.md
deleted file mode 100644
index 38a61239..00000000
--- a/docs/issues/unic-archon-dlc/05-explore-prototype-and-spike-gate.md
+++ /dev/null
@@ -1,36 +0,0 @@
-# Explore workflow — prototype, spike verdicts, and spike-branch gate
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-Second half of the `explore` workflow: a prototype node that runs structured experiments, an optional human gate for preserving valuable spike code, and a spike ticket in the configured tracker.
-
-In scope:
-
-- **`prototype` node** appended to `.archon/workflows/explore.yaml`. Reads `findings.md` and produces one or more experiments, each emitting a structured verdict — exactly one of:
-  - `VALIDATED` — approach works as expected, evidence captured.
-  - `INVALIDATED` — approach failed, reason captured.
-  - `PARTIAL` — partially works, gaps documented.
-  - Verdicts and per-experiment notes are written into a `## Spike verdicts` section of `findings.md` (single source of truth — no separate spike file).
-- **Optional interactive gate** after `prototype`: asks the user whether the spike produced code worth preserving on a branch. If yes, the workflow creates a `spike/<slug>` branch with the prototype code committed; if no, the worktree is left clean.
-- **Spike ticket creation** via the tracker adapter (slice 03): a ticket of type `spike`, labelled with the configured priority default, linking to `docs/workflow/<slug>/findings.md`. The ticket records VALIDATED/INVALIDATED/PARTIAL summary in its body so the tracker view is informative.
-
-Out of scope: feeding spike code into `plan` automatically — `plan` reads `findings.md` only, not the branch.
-
-## Acceptance criteria
-
-- [ ] `prototype` node runs after `synthesize` and writes a `## Spike verdicts` section to `findings.md` with at least one experiment carrying exactly one verdict from {VALIDATED, INVALIDATED, PARTIAL}.
-- [ ] Spike-branch gate is interactive: answering "yes" creates a `spike/<slug>` branch with a commit; answering "no" leaves no branch behind.
-- [ ] A spike ticket is created in the configured tracker via the tracker adapter, with type label `spike`, body linking to `findings.md`, and a one-line verdict summary.
-- [ ] Re-running `/unic-dlc-explore <slug>` on an existing slug does not duplicate the spike ticket; the existing ticket is updated.
-
-## Blocked by
-
-- `docs/issues/unic-archon-dlc/03-triage-workflow-and-tracker-adapter.md`
-- `docs/issues/unic-archon-dlc/04-explore-parallel-research-and-synthesize.md`
diff --git a/docs/issues/unic-archon-dlc/06-plan-specs-and-to-prd.md b/docs/issues/unic-archon-dlc/06-plan-specs-and-to-prd.md
deleted file mode 100644
index 4e072d6a..00000000
--- a/docs/issues/unic-archon-dlc/06-plan-specs-and-to-prd.md
+++ /dev/null
@@ -1,40 +0,0 @@
-# Plan workflow — specs, to-prd, and first PR gate
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-First half of the `plan` workflow: a grill-with-docs interview, live ADR writes for non-obvious decisions, a structured PRD output, and the first human approval gate.
-
-In scope:
-
-- `.archon/workflows/plan.yaml` (extended by slice 07).
-- `/unic-dlc-plan` command file.
-- **`specs` node — grill-with-docs style:**
-  - Loads `CONTEXT.md`, `CONTEXT-MAP.md` (if present), all existing ADRs from `docs/adr/`, and `docs/workflow/<slug>/findings.md` if present.
-  - Conducts an adversarial interview, **one question at a time**, challenging terminology against the domain glossary.
-  - Proposes ADR entries for non-obvious decisions as they crystallise — not for every answer. Written live into `docs/adr/NNNN-<slug>.md` at the moment the decision is made.
-  - Configurable assumptions-first mode via `workflow.discuss_mode: assumptions` in `.archon/unic-dlc.config.json` (defaults to `interview`). Assumptions mode is for experienced developers in established codebases — surfaces a set of assumptions for the user to correct rather than asking from scratch.
-- **`to-prd` node:** synthesises the interview transcript and live-written ADRs into `docs/workflow/<slug>/PRD.md` with the agreed structure: Problem Statement, Solution, User Stories, Implementation Decisions, Testing Decisions, Out of Scope, Further Notes.
-- **First human PR gate (`interactive: true`):** opens a PR against the working branch with `PRD.md` and any new ADRs staged for review. Workflow pauses until the gate is approved.
-
-Out of scope (slice 07 and 08): `to-issues`, `nyquist-map`, `plan-checker`, `yaml-gen`, second PR gate.
-
-## Acceptance criteria
-
-- [ ] `specs` loads CONTEXT.md and existing ADRs before the first interview question.
-- [ ] Interview asks one question at a time (no multi-question batches) and rephrases when the user pushes back on terminology.
-- [ ] At least one ADR is written into `docs/adr/` during a representative session that contains a non-obvious decision; sessions with only obvious answers produce zero ADRs.
-- [ ] `to-prd` writes `docs/workflow/<slug>/PRD.md` with all seven structural sections present.
-- [ ] `workflow.discuss_mode: assumptions` flips the interview into assumptions mode; the default mode runs full interviews.
-- [ ] First human PR gate blocks the workflow until approved; rejected approvals return control to `specs` for another iteration.
-- [ ] Command file uses the `unic-dlc-` prefix.
-
-## Blocked by
-
-- `docs/issues/unic-archon-dlc/02-full-install-hook-and-agent-docs.md`
diff --git a/docs/issues/unic-archon-dlc/07-plan-to-issues-and-nyquist.md b/docs/issues/unic-archon-dlc/07-plan-to-issues-and-nyquist.md
deleted file mode 100644
index 5ce98e1d..00000000
--- a/docs/issues/unic-archon-dlc/07-plan-to-issues-and-nyquist.md
+++ /dev/null
@@ -1,38 +0,0 @@
-# Plan workflow — to-issues, nyquist-map, and validation gate
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-Decomposes the approved PRD into vertically-sliced issues with explicit dependencies and a test command for each issue, then asks the user to validate the breakdown before publication.
-
-In scope:
-
-- **`to-issues` node** extending `plan.yaml`:
-  - Reads `docs/workflow/<slug>/PRD.md` and emits `docs/workflow/<slug>/issues.json` with a list of issues; each issue carries `title`, `type` (canonical type label), `priority` (canonical priority label), `blocked_by: []` (array of issue IDs in this same file), `acceptance_criteria: []`, and `summary`.
-  - Slices are vertical tracer bullets — each slice cuts through every integration layer end-to-end, not horizontally by layer.
-  - Before writing the issues to the tracker, presents the breakdown to the user (numbered list with type/priority/blocked-by/summary) and asks the user to validate, split, merge, or relabel. Iterates until approved.
-  - After approval, publishes each issue to the configured tracker via the tracker adapter, in dependency order so blocker IDs are real.
-- **`nyquist-map` node:** for each issue, attaches a `test_command` field (the exact shell command that proves the issue's acceptance criteria). For issues whose test command does not yet exist, attaches a `test_command_planned` flag so `plan-checker` (slice 08) can flag them.
-- This slice publishes the issues to the tracker so the user can see them — but does **not** generate the build YAML (slice 08) and does **not** invoke the `plan-checker` loop (slice 08). It is acceptable for a user to stop here.
-
-Out of scope: `plan-checker` loop, `yaml-gen`, second PR gate (all in slice 08).
-
-## Acceptance criteria
-
-- [ ] `to-issues` reads PRD.md and writes `issues.json` whose structure matches the field set above.
-- [ ] Issues are vertical slices — each entry's acceptance criteria can be demonstrated independently (verified by a representative example in tests).
-- [ ] Validation gate is interactive and supports split/merge/relabel before publication.
-- [ ] After approval, each issue exists in the configured tracker with correct type and priority labels (verified for the local-markdown backend at minimum).
-- [ ] `nyquist-map` attaches a `test_command` (or `test_command_planned: true`) to every issue.
-- [ ] Tracker publication is dependency-ordered: every `blocked_by` reference points to an issue that already exists in the tracker.
-
-## Blocked by
-
-- `docs/issues/unic-archon-dlc/03-triage-workflow-and-tracker-adapter.md`
-- `docs/issues/unic-archon-dlc/06-plan-specs-and-to-prd.md`
diff --git a/docs/issues/unic-archon-dlc/08-plan-checker-and-yaml-gen.md b/docs/issues/unic-archon-dlc/08-plan-checker-and-yaml-gen.md
deleted file mode 100644
index 24bc3309..00000000
--- a/docs/issues/unic-archon-dlc/08-plan-checker-and-yaml-gen.md
+++ /dev/null
@@ -1,41 +0,0 @@
-# Plan workflow — plan-checker loop, yaml-gen, and second PR gate
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-Closes out the `plan` workflow with a validating loop that hardens the issue plan, a runtime YAML generator that turns the dependency tree into a build DAG, and the second human PR gate.
-
-In scope:
-
-- **`plan-checker` loop node:**
-  - Max 3 iterations before escalation to a human gate.
-  - **Stall detection:** if the issue count does not decrease between consecutive iterations, escalate immediately to the human gate rather than burning the remaining retries.
-  - Validates four dimensions:
-    - **Consistency** — no orphaned `blocked_by` references, all PRD acceptance criteria are covered by at least one issue.
-    - **Completeness** — every mandatory field is present on every issue.
-    - **Decision coverage** — every trackable decision in `CONTEXT.md` (and any decisions captured during `specs`) appears in at least one issue.
-    - **Nyquist compliance** — every issue maps to a concrete test command (no lingering `test_command_planned: true` unless explicitly accepted at the gate).
-- **`yaml-gen` bash node:** reads `issues.json`, builds the dependency DAG, detects parallel groups (issues with no shared `blocked_by`), and writes `.archon/workflows/build-<slug>.yaml`. Both `code-red` and `code-green` for independent issues are emitted as parallel-capable nodes. Linear chains produce serial nodes; circular dependencies are detected, reported, and abort the generation. Output YAML must be valid Archon syntax.
-- **Dependency tree builder + YAML generator module:** pure data transformation, separately unit-tested. Test cases: linear chain → serial; independents → parallel; circular → reported; output round-trips through Archon's YAML parser.
-- **Second human PR gate (`interactive: true`):** opens a PR containing `issues.json`, the generated `build-<slug>.yaml`, and the `plan-checker` report. Workflow pauses until approved. Rejected approvals return control to `plan-checker` for another validation pass (not a full re-decomposition).
-
-Out of scope: executing the generated build YAML (slice 09).
-
-## Acceptance criteria
-
-- [ ] `plan-checker` runs at most 3 iterations and escalates on stall (issue count stays equal between consecutive iterations).
-- [ ] `plan-checker` reports failures across all four validation dimensions in a single structured report.
-- [ ] `yaml-gen` produces a syntactically valid Archon YAML; circular dependencies abort with a clear error.
-- [ ] Independent issues appear as parallel nodes; chained issues appear as serial nodes; both `code-red` and `code-green` carry parallel grouping when applicable.
-- [ ] Dependency tree builder unit tests cover linear, parallel-fan, diamond, and circular cases.
-- [ ] Second human PR gate stages `issues.json` + `build-<slug>.yaml` + `plan-checker` report in a PR and blocks until approved.
-
-## Blocked by
-
-- `docs/issues/unic-archon-dlc/07-plan-to-issues-and-nyquist.md`
diff --git a/docs/issues/unic-archon-dlc/09-build-tdd-core-and-slopcheck.md b/docs/issues/unic-archon-dlc/09-build-tdd-core-and-slopcheck.md
deleted file mode 100644
index 77abe16e..00000000
--- a/docs/issues/unic-archon-dlc/09-build-tdd-core-and-slopcheck.md
+++ /dev/null
@@ -1,43 +0,0 @@
-# Build workflow — TDD core and slopcheck
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-The TDD execution core of `build`: the generated `build-<slug>.yaml` is executed by Archon, enforcing `code-red` before `code-green` per issue, with both phases parallelisable across independent issues. A `slopcheck` gate sits in front of any package install to guard against AI-hallucinated dependencies.
-
-In scope:
-
-- **`/unic-dlc-build <slug>` command** that executes the runtime-generated `.archon/workflows/build-<slug>.yaml` produced in slice 08.
-- **Per-issue TDD discipline enforced by the DAG:**
-  - `code-red` writes failing acceptance tests for the issue.
-  - `code-green` writes the minimum implementation to pass those tests.
-  - `depends_on` edges in the generated YAML force red → green within an issue.
-- **Cross-issue parallelism:** both `code-red` and `code-green` may run in parallel across issues whose `blocked_by` is empty (or whose blockers are already complete). The `yaml-gen` output from slice 08 must already encode this; this slice verifies the runtime honours it.
-- **`slopcheck` bash node** placed before any `package install` step:
-  - Detects newly introduced package names by diffing manifest files (`package.json`, `pyproject.toml`, `Cargo.toml`, etc. depending on the project — start with the JS/TS ecosystem and document how to extend).
-  - For each new package, checks registry existence (npm registry HEAD request for JS; equivalent for the other ecosystems already in scope).
-  - Packages that fail the registry check are flagged `[ASSUMED]`.
-  - For each `[ASSUMED]` package, creates a `checkpoint:human-verify` task inline so the user explicitly approves before installation continues.
-  - If the optional `slopcheck` Python tool (GSD's slopsquatting gate) is on `PATH`, defer to it for the registry check; otherwise use the fallback registry HEAD checks.
-  - If neither path is available, default to marking every new package `[ASSUMED]` — stricter, not silently permissive.
-
-Out of scope: verification, goals-check, report, `/unic-dlc-review` (slices 10 and 11).
-
-## Acceptance criteria
-
-- [ ] Executing `/unic-dlc-build <slug>` runs the runtime-generated YAML and produces commits in the order: failing test → passing implementation, per issue.
-- [ ] Independent issues are observed to run `code-red` (and `code-green`) in parallel for at least one representative `issues.json` test case.
-- [ ] `slopcheck` runs before any package install and creates a `checkpoint:human-verify` task for every `[ASSUMED]` package.
-- [ ] Approving the checkpoint allows the install to proceed; rejecting it aborts the install for that package and surfaces a clear message.
-- [ ] When neither the Python `slopcheck` tool nor the fallback registry check is available, every new package is treated as `[ASSUMED]`.
-- [ ] Command file uses the `unic-dlc-` prefix.
-
-## Blocked by
-
-- `docs/issues/unic-archon-dlc/08-plan-checker-and-yaml-gen.md`
diff --git a/docs/issues/unic-archon-dlc/10-build-verification-goals-check-and-report.md b/docs/issues/unic-archon-dlc/10-build-verification-goals-check-and-report.md
deleted file mode 100644
index d2f82038..00000000
--- a/docs/issues/unic-archon-dlc/10-build-verification-goals-check-and-report.md
+++ /dev/null
@@ -1,42 +0,0 @@
-# Build workflow — verification, goals-check, report, and PR gate
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-Quality and consolidation phase of `build`: a verification node that catches low-quality implementations, a goals-check coverage matrix that ties code back to the PRD, a consolidated report, and the human PR gate that follows.
-
-In scope:
-
-- **`verification` node:**
-  - Runs the full test suite for the slug's touched packages.
-  - Checks coverage thresholds (from `.archon/unic-dlc.config.json`).
-  - Detects stub patterns: `TODO` / `FIXME` / empty function bodies that only `return` or `pass` / hardcoded sentinel values returned where logic was expected.
-  - Audits wiring across layers: component → API → DB; form → handler. Mismatches are reported with file:line references so the report is actionable.
-- **`goals-check` node:** reads `docs/workflow/<slug>/PRD.md` and produces a **coverage matrix** mapping each acceptance criterion (from every user story / issue) to implementation evidence (file paths, test names, or "MISSING"). The matrix is the artifact reviewers look at before merging.
-- **`report` node:** writes `docs/workflow/<slug>/report.md` consolidating:
-  - What was built (high-level summary of merged issues)
-  - Goals-check matrix (verbatim from the previous node)
-  - Test outcomes (pass/fail counts, coverage numbers)
-  - Decisions made during implementation (links to any new ADRs)
-  - Tech debt flagged for cleanup (consumed by slice 13)
-- **Human PR gate (`interactive: true`)** after `report`: opens a PR with the implementation plus `report.md`. Workflow pauses until approved. Rejected approvals return control to `verification` for another pass.
-
-Out of scope: the `/unic-dlc-review` command itself (slice 11) — this slice's PR gate is the human review; slice 11 adds the automated AFK review.
-
-## Acceptance criteria
-
-- [ ] `verification` fails when coverage is below threshold, when stub patterns are present, or when a wiring mismatch is detected.
-- [ ] `goals-check` produces a coverage matrix where each row maps an acceptance criterion to evidence or "MISSING"; matrices with any "MISSING" rows fail the node.
-- [ ] `report.md` contains all five sections and is committed to `docs/workflow/<slug>/`.
-- [ ] Human PR gate blocks until approved; rejection returns control to `verification`.
-- [ ] Tests cover the stub-pattern detector with at least three positive and three negative cases.
-
-## Blocked by
-
-- `docs/issues/unic-archon-dlc/09-build-tdd-core-and-slopcheck.md`
diff --git a/docs/issues/unic-archon-dlc/11-build-self-contained-review-command.md b/docs/issues/unic-archon-dlc/11-build-self-contained-review-command.md
deleted file mode 100644
index 3688660a..00000000
--- a/docs/issues/unic-archon-dlc/11-build-self-contained-review-command.md
+++ /dev/null
@@ -1,42 +0,0 @@
-# Build workflow — self-contained /unic-dlc-review command
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-A self-contained `/unic-dlc-review` command that performs a multi-aspect PR review as an AFK Archon node. It is independent of any external review plugin so users can run the full DLC without installing `pr-review-toolkit`.
-
-In scope:
-
-- **`/unic-dlc-review` command + workflow node** invoked after the human PR gate from slice 10 (the human gate approves the PR; the review runs automatically on the diff).
-- **Four review aspects, each emitting a structured finding list:**
-  1. **Code quality** — readability, naming, function length, project-convention adherence (reads `CLAUDE.md`/`AGENTS.md`).
-  2. **Test coverage adequacy** — exercises through public interfaces, asserts on outputs not internals, no mocks of internal collaborators.
-  3. **Silent failure patterns** — swallowed exceptions, empty `catch` blocks, fallbacks that hide bugs.
-  4. **Type design quality** — encapsulation, invariants expressed in types, illegal-state-unrepresentable patterns.
-- **Inspiration references** documented in the command file:
-  - This monorepo's `apps/claude-code/pr-review/` (patterns, not runtime dependency).
-  - Matt Pocock's `review` skill (https://github.com/mattpocock/skills/blob/main/skills/in-progress/review/SKILL.md).
-  - Note in the command file that this is a base review intended to evolve; later slices may deepen each aspect.
-- **Output:** a structured review comment posted via the tracker adapter against the PR opened by slice 10. Findings are grouped by aspect with file:line references.
-- **No runtime dependency** on `pr-review-toolkit` or any other plugin — verified by running the review in a project where `pr-review-toolkit` is not installed.
-
-Out of scope: integrating the review with the `qa` workflow's gates (slice 12 reads but does not depend on the review output).
-
-## Acceptance criteria
-
-- [ ] `/unic-dlc-review` runs to completion on a project that does not have `pr-review-toolkit` installed.
-- [ ] The review posts a single structured comment grouped by the four aspects, each with at least one finding or an explicit "no findings" line.
-- [ ] Findings include file:line references where applicable.
-- [ ] Command file uses the `unic-dlc-` prefix and includes the inspiration references in its prologue.
-- [ ] Re-running the review on the same PR updates the prior comment rather than appending a duplicate.
-
-## Blocked by
-
-- `docs/issues/unic-archon-dlc/03-triage-workflow-and-tracker-adapter.md`
-- `docs/issues/unic-archon-dlc/10-build-verification-goals-check-and-report.md`
diff --git a/docs/issues/unic-archon-dlc/12-qa-workflow.md b/docs/issues/unic-archon-dlc/12-qa-workflow.md
deleted file mode 100644
index 2cb506ef..00000000
--- a/docs/issues/unic-archon-dlc/12-qa-workflow.md
+++ /dev/null
@@ -1,37 +0,0 @@
-# QA workflow — e2e, coverage gate, UAT, merge
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-A `qa` workflow that runs the e2e suite first, fails fast on insufficient coverage, presents an informed UAT checklist, and merges the PR via the tracker's CLI after sign-off.
-
-In scope:
-
-- `/unic-dlc-qa <slug>` command and `.archon/workflows/qa.yaml`.
-- **E2e suite runs first.** Uses the `e2e_command` configured during install (slice 02). If unset, the workflow halts with a clear message asking the user to run install with `--reconfigure` and set the command.
-- **`coverage-gate` bash node** enforces the coverage thresholds configured in `.archon/unic-dlc.config.json`. Fails fast — the human UAT gate never opens unless automated coverage already passes.
-- **`uat-gate` interactive node:** presents e2e results alongside the UAT checklist extracted from the PRD's acceptance criteria. Approval continues to merge; rejection returns control to whoever owns the failing checklist item.
-- **Merge step via tracker adapter:** after `uat-gate` approves, the workflow merges the PR using the configured tracker's CLI (no manual terminal commands required). Branch deletion follows the configured branching strategy (Gitflow vs GitHub Flow).
-- All workflow YAML uses canonical label names; the adapter translates at write time.
-
-Out of scope: post-merge cleanup (slice 13).
-
-## Acceptance criteria
-
-- [ ] `/unic-dlc-qa <slug>` runs the configured `e2e_command` before any human gate opens.
-- [ ] If `e2e_command` is unset, the workflow halts with a clear, actionable message and a non-zero exit.
-- [ ] `coverage-gate` blocks `uat-gate` when coverage is below threshold.
-- [ ] `uat-gate` shows the PRD's acceptance criteria as a checklist alongside the e2e summary.
-- [ ] After `uat-gate` approves, the PR is merged via the tracker adapter; branch handling matches the configured branching strategy.
-- [ ] Command file uses the `unic-dlc-` prefix.
-
-## Blocked by
-
-- `docs/issues/unic-archon-dlc/03-triage-workflow-and-tracker-adapter.md`
-- `docs/issues/unic-archon-dlc/10-build-verification-goals-check-and-report.md`
diff --git a/docs/issues/unic-archon-dlc/13-cleanup-workflow.md b/docs/issues/unic-archon-dlc/13-cleanup-workflow.md
deleted file mode 100644
index 3f375a50..00000000
--- a/docs/issues/unic-archon-dlc/13-cleanup-workflow.md
+++ /dev/null
@@ -1,40 +0,0 @@
-# Cleanup workflow — arch-review, ADR consolidation, triage
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-Post-merge cleanup that catches drift, locks in architectural decisions, and refreshes project state. Ends by reusing the triage workflow from slice 03.
-
-In scope:
-
-- `/unic-dlc-cleanup <slug>` command and `.archon/workflows/cleanup.yaml`.
-- **`arch-review` node:**
-  - Reads both `docs/workflow/<slug>/PRD.md` (intent) and `docs/workflow/<slug>/report.md` (technical outcome) alongside the changed code.
-  - Catches **technical drift** (modules that became too shallow, tight coupling, leaky abstractions) and **intent drift** (delivered behaviour diverges from the PRD).
-  - Proposes **deepening opportunities** for modules that grew thin during implementation.
-  - Writes `docs/workflow/<slug>/arch-review.md`.
-- **`adr-consolidation` interactive node:**
-  - Reviews decisions logged during `build` (the "Decisions made during implementation" section of `report.md`) plus any ADR drafts proposed during `arch-review`.
-  - **Per-ADR approval gate**: each proposed ADR is presented individually for accept / reject / edit. Only accepted ADRs are written to `docs/adr/`.
-- **Final node reuses slice 03's `triage` workflow** — the same `.archon/workflows/triage.yaml` is referenced via Archon's workflow-include mechanism. Both invocations (standalone vs as cleanup's final node) produce the same `HANDOFF.md` and the same `ROADMAP.md` update.
-
-Out of scope: changes to `triage` itself (any new behaviour belongs in slice 03's file).
-
-## Acceptance criteria
-
-- [ ] `/unic-dlc-cleanup <slug>` writes `docs/workflow/<slug>/arch-review.md` covering technical drift, intent drift, and at least one deepening opportunity (or "none identified").
-- [ ] `adr-consolidation` presents each proposed ADR individually; reject/accept choices are recorded; only accepted ADRs land in `docs/adr/`.
-- [ ] Cleanup's final node reuses `.archon/workflows/triage.yaml` (no duplicate triage logic); `HANDOFF.md` is produced exactly once per cleanup run.
-- [ ] The reused triage node updates `docs/workflow/ROADMAP.md` when invoked as cleanup's final node (no duplicate ROADMAP logic in cleanup itself).
-- [ ] Command file uses the `unic-dlc-` prefix.
-
-## Blocked by
-
-- `docs/issues/unic-archon-dlc/12-qa-workflow.md`
-- `docs/issues/unic-archon-dlc/03-triage-workflow-and-tracker-adapter.md`
diff --git a/docs/issues/unic-archon-dlc/14-readme-and-documentation.md b/docs/issues/unic-archon-dlc/14-readme-and-documentation.md
deleted file mode 100644
index 32d4eb7f..00000000
--- a/docs/issues/unic-archon-dlc/14-readme-and-documentation.md
+++ /dev/null
@@ -1,46 +0,0 @@
-# README and complete documentation
-
-**Status:** ready-for-agent
-**Category:** docs
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-The plugin README — the only user-facing entry point for adopters. Final slice because it documents the shipped behaviour, not the intended behaviour.
-
-In scope (`apps/claude-code/unic-archon-dlc/README.md`):
-
-1. **Mermaid flowchart** showing all six workflows (`explore`, `plan`, `build`, `qa`, `cleanup`, `triage`), every node, and human gate markers (✓ on interactive nodes).
-2. **Node reference table** — one row per node:
-   - Workflow · Node · Type (`prompt` / `loop` / `bash` / `interactive`) · Human gate (✓ / —)
-3. **Quick-start** — install + first run in three steps. Example commands use the `unic-dlc-` prefix.
-4. **Configuration reference** — every key in `.archon/unic-dlc.config.json` with default and valid values: `tracker`, `pr_strategy`, `branching`, `e2e_command`, `model_profile`, `tdd_mode`, `nyquist_validation`, `slopsquatting_gate`, label tier mappings, `repo_layout`, `workflow.discuss_mode`, coverage thresholds.
-5. **`docs/workflow/` layout diagram** — visual of what gets created where and who owns each artifact. State the three-layer separation explicitly: transient (`$ARTIFACTS_DIR`) vs persistent (`docs/workflow/<slug>/`) vs tracker.
-6. **Dependency map:** Archon version requirement, no required peer plugins, optional `slopcheck` Python tool.
-7. **Distribution note:** Archon has no marketplace; this plugin fills the gap by riding the Claude Code plugin marketplace and scaffolding workflows into the target project.
-
-Also in scope:
-
-- The plugin's `CHANGELOG.md` carries a dated `0.1.0` entry summarising the lifecycle (CI's `verify:changelog` will fail without it).
-- The marketplace listing description matches the README's one-line summary.
-
-Out of scope: per-workflow deep-dive docs (those live in `.archon/commands/*.md` shipped by each workflow slice).
-
-## Acceptance criteria
-
-- [ ] README renders correctly (Mermaid block parses; tables render).
-- [ ] Mermaid flowchart shows all 6 workflows, every node from slices 03–13, and human gates marked with ✓.
-- [ ] Node reference table includes every node in the shipped workflows.
-- [ ] Quick-start gets a developer from zero to a successful `/unic-dlc-triage` run in three steps.
-- [ ] Every key written by the install hook (slices 01 and 02) appears in the configuration reference with default and valid values.
-- [ ] `docs/workflow/` layout diagram names the three persistence layers explicitly.
-- [ ] `CHANGELOG.md` has a dated `0.1.0` entry and `pnpm --filter unic-archon-dlc verify:changelog` passes.
-
-## Blocked by
-
-- `docs/issues/unic-archon-dlc/05-explore-prototype-and-spike-gate.md`
-- `docs/issues/unic-archon-dlc/11-build-self-contained-review-command.md`
-- `docs/issues/unic-archon-dlc/13-cleanup-workflow.md`
diff --git a/docs/issues/unic-archon-dlc/15-fix-interactive-loop-fresh-context.md b/docs/issues/unic-archon-dlc/15-fix-interactive-loop-fresh-context.md
deleted file mode 100644
index ec426fdd..00000000
--- a/docs/issues/unic-archon-dlc/15-fix-interactive-loop-fresh-context.md
+++ /dev/null
@@ -1,31 +0,0 @@
-# Fix interactive loop nodes: add fresh_context to prevent session expiry crashes
-
-**Status:** ready-for-agent
-**Category:** bug
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-All interactive loop nodes in the six workflows (`explore.yaml`, `plan.yaml`, `cleanup.yaml`) crash on iteration 2+ when a human responds to a gate. The root cause is a known Archon bug (coleam00/Archon#1208): the DAG executor tries to resume an expired Claude SDK session from the previous iteration instead of starting a fresh one. The fix is to add `fresh_context: true` to every node that is both `interactive: true` and uses a `loop:` block, so each iteration gets a clean session.
-
-Affected nodes across the six workflows:
-
-- `explore.yaml` → `code-preserve-gate`
-- `plan.yaml` → `specs` (grill loop), `prd-gate`, `plan-gate`, `stall-gate`
-- `cleanup.yaml` → `adr-consolidation`
-- `qa.yaml` → `uat-gate`
-
-No lib modules, tests, or install hook are touched — this is a YAML-only patch.
-
-## Acceptance criteria
-
-- [ ] Every `interactive: true` loop node in all six workflow YAMLs has `fresh_context: true` set.
-- [ ] No non-interactive loop node (e.g. `plan-checker`) receives `fresh_context: true` — only interactive gates get it.
-- [ ] `pnpm check` passes at repo root after the change.
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/issues/unic-archon-dlc/16-qa-verify-pr-base.md b/docs/issues/unic-archon-dlc/16-qa-verify-pr-base.md
deleted file mode 100644
index 5903439d..00000000
--- a/docs/issues/unic-archon-dlc/16-qa-verify-pr-base.md
+++ /dev/null
@@ -1,35 +0,0 @@
-# QA workflow: add verify-pr-base guard after merge-pr
-
-**Status:** ready-for-agent
-**Category:** feature
-
-## Parent
-
-`docs/issues/unic-archon-dlc/PRD.md`
-
-## What to build
-
-`qa.yaml` currently calls `merge-pr` (which runs the configured tracker's merge CLI command) but does not verify the PR targets the correct base branch first. If the PR was opened against the wrong target — e.g. `main` instead of `develop` on a Gitflow project — the merge silently ships to the wrong branch.
-
-Add a `verify-pr-base` bash node that runs before `merge-pr`:
-
-1. Reads the configured branching strategy from `.archon/unic-dlc.config.json`.
-2. Derives the expected base branch (`develop` for Gitflow, `main` for GitHub Flow).
-3. For GitHub tracker: calls `gh pr view --json baseRefName` and compares.
-4. For ADO tracker: calls `az repos pr show` and compares `targetRefName`.
-5. For Jira / local: logs a warning and continues (no CLI to query PR base).
-6. If the base is wrong: logs a clear error with the expected vs actual branch names and exits non-zero, preventing `merge-pr` from running.
-
-This is a YAML-only addition of one `bash` node with an inline shell snippet — no lib modules or tests are added.
-
-## Acceptance criteria
-
-- [ ] `qa.yaml` has a `verify-pr-base` bash node with `depends_on: [uat-gate]`.
-- [ ] `merge-pr` has `depends_on: [verify-pr-base]`.
-- [ ] For GitHub and ADO trackers, the node exits non-zero with a clear message if the PR base does not match the expected branch.
-- [ ] For Jira and local trackers, the node logs a warning and exits zero (no-op).
-- [ ] `pnpm check` passes at repo root after the change.
-
-## Blocked by
-
-None — can start immediately.
diff --git a/docs/plans/00-bootstrap-validation.md b/docs/plans/00-bootstrap-validation.md
deleted file mode 100644
index 3c2543c5..00000000
--- a/docs/plans/00-bootstrap-validation.md
+++ /dev/null
@@ -1,84 +0,0 @@
-# 00. Bootstrap Validation
-
-**Status: done — 2026-04-29**
-
-**Priority:** P0
-**Effort:** XS
-**Version impact:** none
-**Depends on:** —
-**Touches:** `CHANGELOG.md`, `docs/plans/README.md`
-
-## Context
-
-The monorepo seed was created by hand. Before any substantive work begins, verify that `pnpm install` resolves the workspace cleanly, Biome and Prettier can run without errors, and tsc accepts the base config. This spec produces the first commit confirming the seed is valid.
-
-## Current behaviour
-
-- `apps/claude-code/` contains only `.gitkeep`
-- `packages/biome-config/`, `packages/tsconfig/`, `packages/release-tools/` each contain only `package.json` stubs
-- `CHANGELOG.md` exists with a single `- Initial monorepo seed` bullet under `[Unreleased]`
-- No git history yet (seed was never committed)
-
-## Target behaviour
-
-- `pnpm install` completes without errors
-- `pnpm check` passes (Biome + Prettier — no JS/TS files to lint yet, so Biome is effectively a no-op; Prettier checks all `.md` files)
-- `pnpm typecheck` exits cleanly (no `.mjs` files to check yet)
-- The seed is committed as `chore(spec-00): validate monorepo seed`
-
-## Affected files
-
-| File                   | Change                                                         |
-| ---------------------- | -------------------------------------------------------------- |
-| `CHANGELOG.md`         | Modify — add dated entry under a new `## [YYYY-MM-DD]` section |
-| `docs/plans/README.md` | No change required                                             |
-
-## Implementation steps
-
-1. Run `pnpm install`. Confirm it exits 0. If it fails due to a catalog resolution error, inspect `pnpm-workspace.yaml` and fix the offending entry.
-
-2. Run `pnpm check`. Confirm it exits 0. Fix any Prettier formatting issues in the seed `.md` files if it reports them (run `pnpm format` to auto-fix).
-
-3. Run `pnpm typecheck`. Confirm it exits 0 (no packages have `.mjs` files yet, so this is a no-op).
-
-4. In `CHANGELOG.md`, move the `- Initial monorepo seed` bullet from `## [Unreleased]` into a new dated section:
-
-   ```markdown
-   ## [2025-XX-XX]
-
-   ### Added
-
-   - Initial monorepo seed: workspace tooling, package stubs, and spec roadmap
-   ```
-
-   Replace `2025-XX-XX` with today's date. Leave `## [Unreleased]` above it with `### Added\n\n- (none)`.
-
-5. Stage all files and commit:
-
-   ```sh
-   git add -A
-   git commit -m "chore(spec-00): validate monorepo seed"
-   ```
-
-## Verification
-
-```sh
-pnpm install    # exits 0, no errors
-pnpm check      # exits 0
-pnpm typecheck  # exits 0
-git log --oneline -1  # shows the chore(spec-00) commit
-```
-
-## Acceptance criteria
-
-- [ ] `pnpm install` succeeds
-- [ ] `pnpm check` passes
-- [ ] `pnpm typecheck` passes
-- [ ] `CHANGELOG.md` has a dated section for today
-- [ ] One commit exists: `chore(spec-00): validate monorepo seed`
-
-## Out of scope
-
-- Installing any npm packages beyond what the seed already declares
-- Creating any new source files
-- Touching any spec file other than this one
diff --git a/docs/plans/01-tsconfig-package.md b/docs/plans/01-tsconfig-package.md
deleted file mode 100644
index 46f77575..00000000
--- a/docs/plans/01-tsconfig-package.md
+++ /dev/null
@@ -1,107 +0,0 @@
-# 01. tsconfig Workspace Package
-
-**Status: done — 2026-04-29**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** none
-**Depends on:** 00
-**Touches:** `packages/tsconfig/`, `tsconfig.base.json` (root)
-
-## Context
-
-Plugin packages will need a shared TypeScript base config for `tsc --checkJs --noEmit`. Rather than each package copying the same options, `@unic/tsconfig` exports a `tsconfig.base.json` that any plugin extends. The root `tsconfig.base.json` is a placeholder seed; this spec moves the content into the package and removes the root file.
-
-## Current behaviour
-
-`packages/tsconfig/package.json` is a stub (no `tsconfig.base.json` inside). Root `tsconfig.base.json` contains:
-
-```json
-{
-  "$schema": "https://json.schemastore.org/tsconfig",
-  "compilerOptions": {
-    "target": "ES2022",
-    "module": "NodeNext",
-    "moduleResolution": "NodeNext",
-    "strict": true,
-    "allowJs": true,
-    "checkJs": true,
-    "noEmit": true,
-    "skipLibCheck": true
-  }
-}
-```
-
-## Target behaviour
-
-- `packages/tsconfig/tsconfig.base.json` contains the compiler options above
-- `packages/tsconfig/package.json` exports the file correctly
-- Root `tsconfig.base.json` is deleted
-- `@unic/tsconfig` is added as `workspace:*` devDependency in the root `package.json`
-- `pnpm install` succeeds; `pnpm typecheck` still passes
-
-## Affected files
-
-| File                                   | Change                                            |
-| -------------------------------------- | ------------------------------------------------- |
-| `packages/tsconfig/tsconfig.base.json` | Create                                            |
-| `packages/tsconfig/package.json`       | Modify — add `exports`                            |
-| `tsconfig.base.json` (root)            | Delete                                            |
-| `package.json` (root)                  | Modify — add `@unic/tsconfig: workspace:*` devDep |
-
-## Implementation steps
-
-1. Create `packages/tsconfig/tsconfig.base.json` with the content from the current root `tsconfig.base.json`.
-
-2. Update `packages/tsconfig/package.json`:
-
-   ```json
-   {
-     "name": "@unic/tsconfig",
-     "version": "0.0.0",
-     "private": true,
-     "license": "LGPL-3.0-or-later",
-     "files": ["tsconfig.base.json"],
-     "exports": {
-       "./tsconfig.base.json": "./tsconfig.base.json"
-     }
-   }
-   ```
-
-3. Delete root `tsconfig.base.json`.
-
-4. Add `"@unic/tsconfig": "workspace:*"` to `devDependencies` in root `package.json`.
-
-5. Run `pnpm install` to link the new workspace package.
-
-## Verification
-
-```sh
-pnpm install            # exits 0
-pnpm typecheck          # exits 0
-node -e "import('@unic/tsconfig/tsconfig.base.json', { with: { type: 'json' } }).then(m => console.log(m.default.compilerOptions.target))"
-# prints: ES2022
-```
-
-## Acceptance criteria
-
-- [ ] `packages/tsconfig/tsconfig.base.json` exists with `"target": "ES2022"`
-- [ ] `packages/tsconfig/package.json` has an `exports` field
-- [ ] Root `tsconfig.base.json` is deleted
-- [ ] `pnpm install` succeeds
-- [ ] `pnpm typecheck` passes
-
-## Out of scope
-
-- Updating individual plugin packages to extend from `@unic/tsconfig` — that happens in specs 05–07
-- Adding any linting rules or additional config files
-
-## Deviations
-
-The spec's verification command uses the `assert` import attribute syntax:
-
-```sh
-node -e "import('@unic/tsconfig/tsconfig.base.json', { with: { type: 'json' } })..."
-```
-
-Node 24 requires `with` instead of `assert` (the `assert` form throws `ERR_IMPORT_ATTRIBUTE_MISSING`). The package works correctly; the verification was run with `{ with: { type: 'json' } }` and printed `ES2022` as expected.
diff --git a/docs/plans/02-biome-config-package.md b/docs/plans/02-biome-config-package.md
deleted file mode 100644
index 941e15bf..00000000
--- a/docs/plans/02-biome-config-package.md
+++ /dev/null
@@ -1,102 +0,0 @@
-# 02. biome-config Workspace Package
-**Status: done — 2026-04-29**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** none
-**Depends on:** 01
-**Touches:** `packages/biome-config/`, `biome.json` (root), `package.json` (root)
-
-## Context
-
-The root `biome.json` currently contains all Biome rules inline. Extracting them into `@unic/biome-config` lets plugin packages extend the same rules, and makes it easy to update rules in one place.
-
-## Current behaviour
-
-Root `biome.json` (full inline rules, see current file).
-
-`packages/biome-config/package.json` is a stub (no `biome.json` inside).
-
-## Target behaviour
-
-- `packages/biome-config/biome.json` contains all current Biome rules
-- Root `biome.json` is simplified to extend from `@unic/biome-config/biome.json` plus workspace-level file overrides
-- `@unic/biome-config` is added as `workspace:*` devDependency in the root `package.json`
-- `pnpm check` still passes after the change
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `packages/biome-config/biome.json` | Create |
-| `packages/biome-config/package.json` | Modify — add `exports` |
-| `biome.json` (root) | Modify — replace inline rules with `extends` |
-| `package.json` (root) | Modify — add `@unic/biome-config: workspace:*` devDep |
-
-## Implementation steps
-
-1. Copy the full content of the current root `biome.json` into `packages/biome-config/biome.json`. Remove the `"files"` section from the copy (that section is workspace-specific and belongs in the root).
-
-2. Update `packages/biome-config/package.json`:
-
-   ```json
-   {
-     "name": "@unic/biome-config",
-     "version": "0.0.0",
-     "private": true,
-     "license": "LGPL-3.0-or-later",
-     "files": ["biome.json"],
-     "exports": {
-       "./biome.json": "./biome.json"
-     }
-   }
-   ```
-
-3. Replace root `biome.json` with a minimal file that extends the package and keeps the workspace-level `"files"` overrides:
-
-   ```json
-   {
-     "$schema": "https://biomejs.dev/schemas/2.4.0/schema.json",
-     "extends": ["@unic/biome-config/biome.json"],
-     "files": {
-       "includes": [
-         "**",
-         "!**/node_modules",
-         "!**/.history",
-         "!**/pnpm-lock.yaml",
-         "!**/.claude",
-         "!**/.ralph",
-         "!**/*.min.js"
-       ]
-     }
-   }
-   ```
-
-4. Add `"@unic/biome-config": "workspace:*"` to `devDependencies` in root `package.json`.
-
-5. Run `pnpm install` to link the new workspace package.
-
-## Verification
-
-```sh
-pnpm install   # exits 0
-pnpm check     # exits 0 — Biome still finds and checks all files
-pnpm ci:check  # exits 0
-```
-
-## Acceptance criteria
-
-- [ ] `packages/biome-config/biome.json` exists with `"formatter"`, `"linter"`, `"javascript"`, `"json"` sections
-- [ ] `packages/biome-config/package.json` has `exports` field
-- [ ] Root `biome.json` has an `"extends"` key pointing to `@unic/biome-config/biome.json`
-- [ ] `pnpm install` succeeds
-- [ ] `pnpm check` passes
-
-## Out of scope
-
-- Updating individual plugin packages to extend from `@unic/biome-config` — happens in specs 05–07
-- Changing any linting rules
-
-## Deviations
-
-- `packages/biome-config/biome.json` required `"root": false` as the first key. Biome 2.4.0 auto-discovers all `biome.json` files in the workspace and treats any without `"root": false` as a root configuration, causing the error "Found a nested root configuration, but there's already a root configuration." Running `biome migrate --write` added `"root": false` automatically. The spec content is otherwise identical.
diff --git a/docs/plans/03-release-tools-package.md b/docs/plans/03-release-tools-package.md
deleted file mode 100644
index 3b5aca3e..00000000
--- a/docs/plans/03-release-tools-package.md
+++ /dev/null
@@ -1,137 +0,0 @@
-# 03. release-tools Workspace Package
-**Status: done — 2026-04-29**
-
-**Priority:** P0
-**Effort:** L
-**Version impact:** none
-**Depends on:** 02
-**Touches:** `packages/release-tools/`, `package.json` (root)
-
-## Context
-
-Each existing plugin repo has its own copy of `bump-version.mjs`, `sync-version.mjs`, `tag.mjs`, `verify-changelog.mjs`, `lib/platform.mjs`, and `lib/types.mjs`. The confluence versions are the most mature. This spec lifts them into `@unic/release-tools`, parameterises them to work from any plugin's cwd, and exposes them as bin commands so plugin `package.json` scripts can call them without relative paths.
-
-## Current behaviour
-
-`packages/release-tools/package.json` is a stub. The source scripts live at:
-
-- `~/Sites/UNIC/unic-claude-code-confluence/scripts/bump-version.mjs`
-- `~/Sites/UNIC/unic-claude-code-confluence/scripts/sync-version.mjs`
-- `~/Sites/UNIC/unic-claude-code-confluence/scripts/tag.mjs`
-- `~/Sites/UNIC/unic-claude-code-confluence/scripts/verify-changelog.mjs`
-- `~/Sites/UNIC/unic-claude-code-confluence/scripts/verify-changelog.test.mjs`
-- `~/Sites/UNIC/unic-claude-code-confluence/scripts/lib/platform.mjs`
-- `~/Sites/UNIC/unic-claude-code-confluence/scripts/lib/types.mjs`
-
-## Target behaviour
-
-`packages/release-tools/` contains:
-
-```
-scripts/
-  bump-version.mjs
-  sync-version.mjs
-  tag.mjs
-  verify-changelog.mjs
-  verify-changelog.test.mjs
-  lib/
-    platform.mjs
-    types.mjs
-package.json  ← updated with bin, scripts, devDeps
-```
-
-The scripts operate on the **calling package's directory** via `process.cwd()` — no `--package` flag needed. When run with `pnpm --filter pr-review bump patch`, pnpm sets cwd to `apps/claude-code/pr-review/` automatically.
-
-Bin commands exposed:
-
-| Bin | Script |
-|---|---|
-| `unic-bump` | `scripts/bump-version.mjs` |
-| `unic-sync-version` | `scripts/sync-version.mjs` |
-| `unic-tag` | `scripts/tag.mjs` |
-| `unic-verify-changelog` | `scripts/verify-changelog.mjs` |
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `packages/release-tools/scripts/bump-version.mjs` | Create (lifted + adapted) |
-| `packages/release-tools/scripts/sync-version.mjs` | Create (lifted + adapted) |
-| `packages/release-tools/scripts/tag.mjs` | Create (lifted + adapted) |
-| `packages/release-tools/scripts/verify-changelog.mjs` | Create (lifted + adapted) |
-| `packages/release-tools/scripts/verify-changelog.test.mjs` | Create (lifted + adapted) |
-| `packages/release-tools/scripts/lib/platform.mjs` | Create (lifted as-is) |
-| `packages/release-tools/scripts/lib/types.mjs` | Create (lifted as-is) |
-| `packages/release-tools/package.json` | Modify — add `bin`, `scripts`, `devDependencies` |
-| `package.json` (root) | Modify — add `@unic/release-tools: workspace:*` devDep |
-
-## Implementation steps
-
-1. Read each source script from `~/Sites/UNIC/unic-claude-code-confluence/scripts/` to understand its current implementation.
-
-2. Copy `lib/platform.mjs` and `lib/types.mjs` verbatim into `packages/release-tools/scripts/lib/`. Update any import paths that reference sibling scripts.
-
-3. Copy and adapt each main script into `packages/release-tools/scripts/`:
-   - Replace any hardcoded relative paths (e.g. `'./lib/platform.mjs'`) with paths relative to the script's new location.
-   - Ensure each script resolves the plugin root via `process.cwd()` (not `import.meta.dirname`). The plugin root is wherever the script is run from.
-   - `bump-version.mjs`: reads `.claude-plugin/plugin.json` from `process.cwd()`; updates version in `plugin.json`, `marketplace.json`, `package.json` (all relative to cwd); promotes `[Unreleased]` in `CHANGELOG.md`.
-   - `sync-version.mjs`: mirrors version from `plugin.json` → `marketplace.json` + `package.json`.
-   - `tag.mjs`: reads version from `plugin.json`; creates git tag `<plugin-name>@<version>` (tag scheme: `name@version`, not `vX.Y.Z`). Update this from confluence's `v` prefix.
-   - `verify-changelog.mjs`: checks that `CHANGELOG.md` has an `[Unreleased]` section with at least one non-`(none)` bullet; no version bump needed.
-
-4. Copy and adapt `verify-changelog.test.mjs`. Update import paths to point to the script's new location. Ensure tests still pass with `node --test`.
-
-5. Update `packages/release-tools/package.json`:
-
-   ```json
-   {
-     "name": "@unic/release-tools",
-     "version": "0.0.0",
-     "private": true,
-     "license": "LGPL-3.0-or-later",
-     "type": "module",
-     "bin": {
-       "unic-bump": "./scripts/bump-version.mjs",
-       "unic-sync-version": "./scripts/sync-version.mjs",
-       "unic-tag": "./scripts/tag.mjs",
-       "unic-verify-changelog": "./scripts/verify-changelog.mjs"
-     },
-     "scripts": {
-       "test": "node --test scripts/verify-changelog.test.mjs"
-     },
-     "devDependencies": {
-       "@types/node": "catalog:",
-       "typescript": "catalog:"
-     }
-   }
-   ```
-
-6. Add `"@unic/release-tools": "workspace:*"` to `devDependencies` in root `package.json`.
-
-7. Run `pnpm install`.
-
-8. Run `pnpm --filter @unic/release-tools test` to confirm the tests pass.
-
-## Verification
-
-```sh
-pnpm install
-pnpm --filter @unic/release-tools test    # all tests pass
-which unic-bump 2>/dev/null || node_modules/.bin/unic-bump --help  # prints usage or exits cleanly
-```
-
-## Acceptance criteria
-
-- [ ] All 4 scripts exist in `packages/release-tools/scripts/`
-- [ ] `lib/platform.mjs` and `lib/types.mjs` exist
-- [ ] `packages/release-tools/package.json` has `bin` with 4 entries
-- [ ] `pnpm --filter @unic/release-tools test` passes
-- [ ] `pnpm install` succeeds
-- [ ] Each script reads the plugin root from `process.cwd()`
-- [ ] `tag.mjs` uses `<name>@<version>` tag scheme (not `v<version>`)
-
-## Out of scope
-
-- Wiring plugin packages to use these commands — happens in specs 05–07
-- Adding tests for bump/sync/tag scripts (verify-changelog test is sufficient for now)
-- Changing the Biome auto-formatting step in `bump-version.mjs` — keep as-is if it exists
diff --git a/docs/plans/04-root-docs.md b/docs/plans/04-root-docs.md
deleted file mode 100644
index 4c83bc1d..00000000
--- a/docs/plans/04-root-docs.md
+++ /dev/null
@@ -1,115 +0,0 @@
-# 04. Root Documentation and Process Templates
-**Status: done — 2026-04-29**
-
-**Priority:** P2
-**Effort:** S
-**Version impact:** none
-**Depends on:** 03
-**Touches:** `docs/process/`, `CLAUDE.md`, `CONTRIBUTING.md`, `AGENTS.md`
-
-## Context
-
-`docs/process/` templates don't exist yet — new plugin authors have no canonical reference for the spec format or the Ralph loop pattern. This spec creates them. Root docs (`AGENTS.md`, `CONTRIBUTING.md`) are already accurate after the seed was manually consolidated; only the `docs/process/` reference in `CONTRIBUTING.md` needs a minor fix.
-
-Note: `CLAUDE.md` is a symlink to `AGENTS.md` (created manually before this spec ran). Do not edit `CLAUDE.md` directly — edit `AGENTS.md`.
-
-## Current behaviour
-
-- `docs/process/` does not exist
-- `CONTRIBUTING.md` contains a stale reference to "coming in spec 05" for `docs/process/`
-- `AGENTS.md` (and its symlink `CLAUDE.md`) is accurate and requires no changes
-
-## Target behaviour
-
-- `docs/process/spec-template.md` — canonical empty spec template with all required sections
-- `docs/process/ralph-loop-guide.md` — concise guide explaining the Ralph loop pattern
-- `CONTRIBUTING.md` reference to `docs/process/` updated (remove "coming in spec 05")
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `docs/process/spec-template.md` | Create |
-| `docs/process/ralph-loop-guide.md` | Create |
-| `CONTRIBUTING.md` | Modify — remove stale "coming in spec 05" note |
-
-## Implementation steps
-
-1. Create `docs/process/spec-template.md`:
-
-   ```markdown
-   # NN. Title
-
-   **Priority:** P0 | P1 | P2
-   **Effort:** XS | S | M | L
-   **Version impact:** none | patch (plugin: <name>) | minor (plugin: <name>) | major (plugin: <name>)
-   **Depends on:** (spec numbers, or "none")
-   **Touches:** (comma-separated list of files/dirs)
-
-   ## Context
-
-   _Why this change is needed._
-
-   ## Current behaviour
-
-   _Exact code snapshots or behaviour **before** the change. Ralph uses this to verify the starting state._
-
-   ## Target behaviour
-
-   _What the code/behaviour should look like **after**._
-
-   ## Affected files
-
-   | File | Change |
-   |---|---|
-   | `path/to/file` | Create / Modify / Delete |
-
-   ## Implementation steps
-
-   1. _Numbered steps with exact before → after diffs or code to write. No ambiguity._
-
-   ## Verification
-
-   _Shell commands to run and their expected output._
-
-   ## Acceptance criteria
-
-   - [ ] _Checkbox item_
-
-   ## Out of scope
-
-   - _Explicit list of things NOT to change in this spec._
-   ```
-
-2. Create `docs/process/ralph-loop-guide.md` — a concise guide (under 300 words) explaining:
-   - What `ralph.yml` and `PROMPT.md` do in each plugin/package
-   - How to start a loop: `pnpm ralph` from the relevant directory
-   - How to stop and resume
-   - How a paused loop (`## Questions` + `LOOP_COMPLETE`) is handled
-   - The single-spec-per-iteration discipline
-
-3. In `CONTRIBUTING.md`, find the line referencing `docs/process/` with "(coming in spec 05)" and remove that parenthetical — the directory now exists.
-
-4. Run `pnpm check` to verify all new `.md` files are Prettier-compliant.
-
-## Verification
-
-```sh
-pnpm check                  # exits 0
-ls docs/process/             # lists spec-template.md and ralph-loop-guide.md
-ls -la CLAUDE.md             # shows symlink → AGENTS.md
-```
-
-## Acceptance criteria
-
-- [ ] `docs/process/spec-template.md` exists with all required section headers
-- [ ] `docs/process/ralph-loop-guide.md` exists and is under 300 words
-- [ ] `CONTRIBUTING.md` no longer has the "coming in spec 05" note
-- [ ] `CLAUDE.md` is a symlink to `AGENTS.md` (pre-existing; verify, do not recreate)
-- [ ] `pnpm check` passes
-
-## Out of scope
-
-- Editing `CLAUDE.md` directly (it is a symlink — edit `AGENTS.md` instead)
-- Changing any functional code
-- Adding process docs for individual plugins (they inherit from this)
diff --git a/docs/plans/05-migrate-pr-review.md b/docs/plans/05-migrate-pr-review.md
deleted file mode 100644
index 3b7ef033..00000000
--- a/docs/plans/05-migrate-pr-review.md
+++ /dev/null
@@ -1,190 +0,0 @@
-# 05. Migrate pr-review Plugin
-**Status: done — 2026-04-29**
-
-**Priority:** P0
-**Effort:** M
-**Version impact:** patch (plugin: pr-review)
-**Depends on:** 03
-**Touches:** `apps/claude-code/pr-review/`
-
-## Context
-
-`unic-pr-review` is a standalone repo. This spec imports its git history into the monorepo under `apps/claude-code/pr-review/`, renames the plugin from `unic-pr-review` to `pr-review`, and adds the workspace integration files (`package.json`, `CHANGELOG.md`) that did not exist in the original repo.
-
-## Current behaviour
-
-- `apps/claude-code/pr-review/` does not exist
-- Source repo: `~/Sites/UNIC/unic-pr-review/` (main branch)
-- Plugin name in `plugin.json`: `"unic-pr-review"`, version `"0.1.0"`
-
-## Target behaviour
-
-- `apps/claude-code/pr-review/` exists with full git history
-- `plugin.json` name: `"pr-review"`, version `"0.1.1"`
-- `marketplace.json` name: `"pr-review"`
-- New `package.json` wires into workspace and release-tools
-- New `CHANGELOG.md` with initial `[Unreleased]` and a `[0.1.1]` section
-- `LICENSE` is present in the plugin dir (maintainer adds it manually — Ralph must not touch it)
-- `pnpm install` succeeds with the new workspace member
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `apps/claude-code/pr-review/` (whole tree) | Create (via filter-repo + merge) |
-| `apps/claude-code/pr-review/.claude-plugin/plugin.json` | Modify — rename + bump |
-| `apps/claude-code/pr-review/.claude-plugin/marketplace.json` | Modify — rename |
-| `apps/claude-code/pr-review/package.json` | Create |
-| `apps/claude-code/pr-review/CHANGELOG.md` | Create |
-| `apps/claude-code/pr-review/LICENSE` | Keep — maintained manually; do not delete |
-
-## Implementation steps
-
-1. Verify `git filter-repo` is installed:
-
-   ```sh
-   git filter-repo --version
-   ```
-
-   If not found: `brew install git-filter-repo` (macOS) or `pip install git-filter-repo` (cross-platform).
-
-2. Create a fresh clone of the source repo (leave the original untouched):
-
-   ```sh
-   git clone --no-local /Users/oriol.torrent/Sites/UNIC/unic-pr-review /tmp/pr-review-rewrite
-   ```
-
-3. Drop runtime artifacts from history (no-op if they were never committed):
-
-   ```sh
-   cd /tmp/pr-review-rewrite
-   git filter-repo --invert-paths --path .ralph --path .history --force
-   ```
-
-4. Move the whole tree into the monorepo subdirectory:
-
-   ```sh
-   git filter-repo --to-subdirectory-filter apps/claude-code/pr-review/ --force
-   ```
-
-5. Back in the monorepo root, import the rewritten history:
-
-   ```sh
-   cd /Users/oriol.torrent/Sites/UNIC/unic-agents-plugins
-   git remote add pr-review-import /tmp/pr-review-rewrite
-   git fetch pr-review-import
-   git merge --allow-unrelated-histories pr-review-import/main \
-     -m "feat(spec-05): import pr-review plugin history"
-   git remote remove pr-review-import
-   ```
-
-6. Clean up the temp clone:
-
-   ```sh
-   rm -rf /tmp/pr-review-rewrite
-   ```
-
-7. Update `apps/claude-code/pr-review/.claude-plugin/plugin.json`:
-   - Change `"name"` from `"unic-pr-review"` to `"pr-review"`
-   - Leave version as `"0.1.0"` for now — the bump step below will advance it
-
-8. Update `apps/claude-code/pr-review/.claude-plugin/marketplace.json`:
-   - Change `"name"` to `"pr-review"`
-   - Update `"homepage"` to `"https://github.com/unic/unic-agents-plugins"` (placeholder until repo is public)
-
-9. ⚠️ **Maintainer action required:** After the merge, verify that `apps/claude-code/pr-review/LICENSE` exists. If the import did not bring it over, copy the root `LICENSE` file there manually. Ralph/Claude must **not** create or delete LICENSE files.
-
-10. Create `apps/claude-code/pr-review/CHANGELOG.md`:
-
-    ```markdown
-    # Changelog
-
-    ## [Unreleased]
-
-    ### Added
-
-    - (none)
-
-    ## [0.1.0] — (see archived repo)
-
-    _History migrated from [unic/unic-pr-review](https://github.com/unic/unic-pr-review)._
-    ```
-
-11. Create `apps/claude-code/pr-review/package.json`:
-
-    ```json
-    {
-      "name": "pr-review",
-      "version": "0.1.0",
-      "private": true,
-      "license": "LGPL-3.0-or-later",
-      "type": "module",
-      "packageManager": "pnpm@10.33.0",
-      "engines": {
-        "node": ">=24",
-        "pnpm": ">=10"
-      },
-      "scripts": {
-        "bump": "unic-bump",
-        "sync-version": "unic-sync-version",
-        "tag": "unic-tag",
-        "verify:changelog": "unic-verify-changelog",
-        "ralph": "ralph run -c ralph.yml -H builtin:code-assist"
-      },
-      "devDependencies": {
-        "@ralph-orchestrator/ralph-cli": "catalog:",
-        "@unic/release-tools": "workspace:*"
-      }
-    }
-    ```
-
-12. Run `pnpm install` to register the new workspace member.
-
-13. Add a CHANGELOG entry for the rename and run the version bump:
-
-    In `apps/claude-code/pr-review/CHANGELOG.md`, under `## [Unreleased] → ### Added`:
-    ```
-    - Migrated to unic-agents-plugins monorepo; plugin renamed from `unic-pr-review` to `pr-review`
-    ```
-
-    Then:
-    ```sh
-    pnpm --filter pr-review bump patch
-    pnpm --filter pr-review verify:changelog
-    ```
-
-14. Stage remaining changes and commit:
-
-    ```sh
-    git add -A
-    git commit -m "feat(spec-05): finalize pr-review migration (v0.1.1)"
-    ```
-
-## Verification
-
-```sh
-pnpm install                             # exits 0
-pnpm --filter pr-review verify:changelog # exits 0
-cat apps/claude-code/pr-review/.claude-plugin/plugin.json | grep '"name"'
-# → "name": "pr-review"
-git log --oneline apps/claude-code/pr-review/ | tail -5
-# shows commits from original unic-pr-review repo
-```
-
-## Acceptance criteria
-
-- [ ] `apps/claude-code/pr-review/` exists
-- [ ] `plugin.json` has `"name": "pr-review"` and version `"0.1.1"`
-- [ ] `marketplace.json` has `"name": "pr-review"`
-- [ ] `package.json` exists with `bump`, `sync-version`, `tag`, `verify:changelog` scripts
-- [ ] `CHANGELOG.md` exists with a `[0.1.1]` dated section
-- [ ] `LICENSE` exists in the plugin dir (add manually if missing — Ralph does not manage this file)
-- [ ] `git log apps/claude-code/pr-review/` shows the original repo's commit history
-- [ ] `pnpm install` succeeds
-- [ ] `pnpm --filter pr-review verify:changelog` passes
-
-## Out of scope
-
-- Updating `apps/claude-code/pr-review/CLAUDE.md` for monorepo context — can be done in a future plugin spec
-- Adding tests (pr-review has no test suite currently)
-- Migrating the plugin's own `docs/plans/` specs — they migrate as part of the history
diff --git a/docs/plans/06-migrate-auto-format.md b/docs/plans/06-migrate-auto-format.md
deleted file mode 100644
index 32512b97..00000000
--- a/docs/plans/06-migrate-auto-format.md
+++ /dev/null
@@ -1,207 +0,0 @@
-# 06. Migrate auto-format Plugin
-**Status: done — 2026-04-29**
-
-**Priority:** P0
-**Effort:** M
-**Version impact:** patch (plugin: auto-format)
-**Depends on:** 03
-**Touches:** `apps/claude-code/auto-format/`
-
-## Context
-
-`unic-claude-code-format` is a standalone repo with a hooks-based Claude Code plugin. This spec imports its git history into the monorepo under `apps/claude-code/auto-format/`, renames the plugin to `auto-format`, removes the per-repo tooling files superseded by the monorepo root, and rewires `package.json` to use `@unic/release-tools`.
-
-## Current behaviour
-
-- `apps/claude-code/auto-format/` does not exist
-- Source repo: `~/Sites/UNIC/unic-claude-code-format/` (main branch)
-- Plugin name in `plugin.json`: `"unic-claude-code-format"`, version `"0.5.8"`
-
-## Target behaviour
-
-- `apps/claude-code/auto-format/` exists with full git history
-- `plugin.json` name: `"auto-format"`, version `"0.5.9"`
-- `marketplace.json` name: `"auto-format"`
-- `package.json` updated: release scripts use `@unic/release-tools` bin commands; per-repo tooling scripts removed
-- Per-repo tooling files removed: `.editorconfig`, `.nvmrc`, `.npmrc`, `pnpm-workspace.yaml`, `.github/`, `scripts/bump.mjs`, `scripts/sync-version.mjs`, `scripts/tag.mjs`, `scripts/backfill-tags.mjs`
-- `scripts/format-hook.mjs` and `tests/` preserved (plugin-specific)
-- `pnpm install` and `pnpm --filter auto-format test` succeed
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `apps/claude-code/auto-format/` (whole tree) | Create (via filter-repo + merge) |
-| `apps/claude-code/auto-format/.claude-plugin/plugin.json` | Modify — rename + bump |
-| `apps/claude-code/auto-format/.claude-plugin/marketplace.json` | Modify — rename |
-| `apps/claude-code/auto-format/package.json` | Modify — rewire scripts, remove tooling deps |
-| `apps/claude-code/auto-format/.editorconfig` | Delete |
-| `apps/claude-code/auto-format/.nvmrc` | Delete |
-| `apps/claude-code/auto-format/.npmrc` | Delete |
-| `apps/claude-code/auto-format/pnpm-workspace.yaml` | Delete |
-| `apps/claude-code/auto-format/.github/` | Delete |
-| `apps/claude-code/auto-format/scripts/bump.mjs` | Delete |
-| `apps/claude-code/auto-format/scripts/sync-version.mjs` | Delete |
-| `apps/claude-code/auto-format/scripts/tag.mjs` | Delete |
-| `apps/claude-code/auto-format/scripts/backfill-tags.mjs` | Delete (if present) |
-| `apps/claude-code/auto-format/LICENSE` | Keep — maintained manually; do not delete |
-
-## Implementation steps
-
-1. Verify `git filter-repo` is installed (see spec 05 step 1 if needed).
-
-2. Fresh clone:
-
-   ```sh
-   git clone --no-local /Users/oriol.torrent/Sites/UNIC/unic-claude-code-format /tmp/auto-format-rewrite
-   ```
-
-3. Drop runtime artifacts from history:
-
-   ```sh
-   cd /tmp/auto-format-rewrite
-   git filter-repo --invert-paths --path .ralph --path .history --force
-   ```
-
-4. Move to subdirectory:
-
-   ```sh
-   git filter-repo --to-subdirectory-filter apps/claude-code/auto-format/ --force
-   ```
-
-5. Import into monorepo:
-
-   ```sh
-   cd /Users/oriol.torrent/Sites/UNIC/unic-agents-plugins
-   git remote add auto-format-import /tmp/auto-format-rewrite
-   git fetch auto-format-import
-   git merge --allow-unrelated-histories auto-format-import/main \
-     -m "feat(spec-06): import auto-format plugin history"
-   git remote remove auto-format-import
-   rm -rf /tmp/auto-format-rewrite
-   ```
-
-6. Delete per-repo tooling files (superseded by monorepo root):
-
-   ```sh
-   rm -f apps/claude-code/auto-format/.editorconfig
-   rm -f apps/claude-code/auto-format/.nvmrc
-   rm -f apps/claude-code/auto-format/.npmrc
-   rm -f apps/claude-code/auto-format/pnpm-workspace.yaml
-   rm -rf apps/claude-code/auto-format/.github
-   rm -f apps/claude-code/auto-format/scripts/bump.mjs
-   rm -f apps/claude-code/auto-format/scripts/sync-version.mjs
-   rm -f apps/claude-code/auto-format/scripts/tag.mjs
-   rm -f apps/claude-code/auto-format/scripts/backfill-tags.mjs
-   ```
-
-7. ⚠️ **Maintainer action required:** After the merge, verify that `apps/claude-code/auto-format/LICENSE` exists. If the import did not bring it over, copy the root `LICENSE` file there manually. Ralph/Claude must **not** create or delete LICENSE files.
-
-8. Update `apps/claude-code/auto-format/.claude-plugin/plugin.json`:
-   - `"name"`: `"unic-claude-code-format"` → `"auto-format"`
-   - Leave version as current (`"0.5.8"`) for now
-
-9. Update `apps/claude-code/auto-format/.claude-plugin/marketplace.json`:
-   - `"name"`: → `"auto-format"`
-   - `"homepage"`: → `"https://github.com/unic/unic-agents-plugins"`
-
-10. Rewrite `apps/claude-code/auto-format/package.json`. Keep `test` and `typecheck` scripts. Replace release scripts. Remove `lint`/`format`/`check`/`ci:check` (handled at monorepo root). Remove `devDependencies` for biome/typescript (now at root catalog; add only what's plugin-specific):
-
-   ```json
-   {
-     "name": "auto-format",
-     "version": "0.5.8",
-     "private": true,
-     "license": "LGPL-3.0-or-later",
-     "type": "module",
-     "packageManager": "pnpm@10.33.0",
-     "engines": {
-       "node": ">=24",
-       "pnpm": ">=10"
-     },
-     "scripts": {
-       "test": "node --test tests/format-hook.test.mjs",
-       "typecheck": "tsc --noEmit --project tsconfig.json",
-       "bump": "unic-bump",
-       "sync-version": "unic-sync-version",
-       "tag": "unic-tag",
-       "verify:changelog": "unic-verify-changelog",
-       "ralph": "ralph run -c ralph.yml -H builtin:code-assist"
-     },
-     "devDependencies": {
-       "@ralph-orchestrator/ralph-cli": "catalog:",
-       "@types/node": "catalog:",
-       "@unic/release-tools": "workspace:*",
-       "@unic/tsconfig": "workspace:*",
-       "typescript": "catalog:"
-     }
-   }
-   ```
-
-11. Create `apps/claude-code/auto-format/tsconfig.json`:
-
-    ```json
-    {
-      "extends": "@unic/tsconfig/tsconfig.base.json",
-      "include": ["scripts/**/*.mjs", "tests/**/*.mjs"]
-    }
-    ```
-
-12. Run `pnpm install`.
-
-13. Add CHANGELOG entry and bump version:
-
-    In `apps/claude-code/auto-format/CHANGELOG.md`, under `## [Unreleased] → ### Added`:
-    ```
-    - Migrated to unic-agents-plugins monorepo; plugin renamed from `unic-claude-code-format` to `auto-format`
-    ```
-
-    ```sh
-    pnpm --filter auto-format bump patch
-    pnpm --filter auto-format verify:changelog
-    ```
-
-14. Run `pnpm --filter auto-format test` and verify it passes.
-
-15. Commit:
-
-    ```sh
-    git add -A
-    git commit -m "feat(spec-06): finalize auto-format migration (v0.5.9)"
-    ```
-
-## Verification
-
-```sh
-pnpm install
-pnpm --filter auto-format test
-pnpm --filter auto-format typecheck
-pnpm --filter auto-format verify:changelog
-cat apps/claude-code/auto-format/.claude-plugin/plugin.json | grep '"name"'
-# → "name": "auto-format"
-```
-
-## Acceptance criteria
-
-- [ ] `apps/claude-code/auto-format/` exists with original commit history
-- [ ] `plugin.json` has `"name": "auto-format"` and version `"0.5.5"` (see Deviations)
-- [ ] `marketplace.json` has `"name": "auto-format"`
-- [ ] `package.json` uses `unic-bump` etc. (no local release script paths)
-- [ ] Per-repo tooling files (`.editorconfig`, `.nvmrc`, `pnpm-workspace.yaml`, `.github/`, etc.) deleted
-- [ ] `LICENSE` exists in the plugin dir (add manually if missing — Ralph does not manage this file)
-- [ ] `scripts/format-hook.mjs` and `tests/` preserved
-- [ ] `pnpm --filter auto-format test` passes
-- [ ] `pnpm --filter auto-format verify:changelog` passes
-
-## Out of scope
-
-- Updating `CLAUDE.md` inside the plugin for monorepo context
-- Migrating the plugin's own release workflow (replaced by spec 10)
-
-## Deviations
-
-- **Version mismatch**: The spec expected the imported `plugin.json` version to be `"0.5.8"` (Current behaviour section), but the actual imported version was `"0.5.4"`. The `package.json` rewrite keeps `"0.5.4"` as the base. After the patch bump in step 13, the final version will be `"0.5.5"` rather than `"0.5.9"` as stated in the acceptance criteria. The acceptance criteria should be interpreted as `"0.5.5"`.
-- **`.nvmrc` absent**: The spec listed `.nvmrc` in the delete list, but the file was not present in the imported history — no action needed.
-- **`scripts/verify-changelog.mjs` retained**: Not in the delete list; left in place as dead code (superseded by `unic-verify-changelog`). Can be removed in a future cleanup.
-- **`marketplace.json` restructured**: The imported `marketplace.json` was a flat object; `sync-version` requires a `{ plugins: [...] }` structure. Restructured to match the monorepo convention.
-- **CHANGELOG date format**: Historical entries used `## [X.Y.Z] - YYYY-MM-DD` (hyphen) from the original repo; `unic-verify-changelog` requires `## [X.Y.Z] — YYYY-MM-DD` (em-dash). All historical entries updated.
diff --git a/docs/plans/07-migrate-confluence-publish.md b/docs/plans/07-migrate-confluence-publish.md
deleted file mode 100644
index 4aafd695..00000000
--- a/docs/plans/07-migrate-confluence-publish.md
+++ /dev/null
@@ -1,222 +0,0 @@
-# 07. Migrate confluence-publish Plugin
-**Status: done — 2026-04-29**
-
-**Priority:** P0
-**Effort:** M
-**Version impact:** patch (plugin: confluence-publish)
-**Depends on:** 03
-**Touches:** `apps/claude-code/confluence-publish/`
-
-## Context
-
-`unic-claude-code-confluence` is a standalone repo. This spec imports its git history into the monorepo under `apps/claude-code/confluence-publish/`, renames the plugin from `unic-confluence` to `confluence-publish`, removes per-repo tooling files superseded by the monorepo root, and rewires `package.json` to use `@unic/release-tools`. The substantial `scripts/push-to-confluence.mjs` and `scripts/lib/` are preserved as-is.
-
-## Current behaviour
-
-- `apps/claude-code/confluence-publish/` does not exist
-- Source repo: `~/Sites/UNIC/unic-claude-code-confluence/` (main branch)
-- Plugin name in `plugin.json`: `"unic-confluence"`, version `"2.4.1"`
-
-## Target behaviour
-
-- `apps/claude-code/confluence-publish/` exists with full git history
-- `plugin.json` name: `"confluence-publish"`, version `"2.4.2"`
-- `marketplace.json` name: `"confluence-publish"`
-- `package.json` updated: release scripts use `@unic/release-tools`; per-repo tooling stripped
-- Per-repo tooling removed: `.editorconfig`, `.nvmrc`, `.npmrc`, `pnpm-workspace.yaml`, `biome.json`, `.github/`, `.githooks/`, `scripts/bump-version.mjs`, `scripts/sync-version.mjs`, `scripts/tag.mjs`, `scripts/backfill-tags.mjs` (if present), `scripts/verify-changelog.mjs`, `scripts/verify-changelog.test.mjs`
-- `scripts/push-to-confluence.mjs`, `scripts/lib/`, and all plugin tests preserved
-- `pnpm install` and `pnpm --filter confluence-publish test` succeed
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `apps/claude-code/confluence-publish/` (whole tree) | Create (via filter-repo + merge) |
-| `apps/claude-code/confluence-publish/.claude-plugin/plugin.json` | Modify — rename + bump |
-| `apps/claude-code/confluence-publish/.claude-plugin/marketplace.json` | Modify — rename |
-| `apps/claude-code/confluence-publish/package.json` | Modify — rewire |
-| `apps/claude-code/confluence-publish/.editorconfig` | Delete |
-| `apps/claude-code/confluence-publish/.nvmrc` | Delete |
-| `apps/claude-code/confluence-publish/.npmrc` | Delete |
-| `apps/claude-code/confluence-publish/pnpm-workspace.yaml` | Delete |
-| `apps/claude-code/confluence-publish/biome.json` | Delete |
-| `apps/claude-code/confluence-publish/.github/` | Delete |
-| `apps/claude-code/confluence-publish/.githooks/` | Delete |
-| `apps/claude-code/confluence-publish/scripts/bump-version.mjs` | Delete |
-| `apps/claude-code/confluence-publish/scripts/sync-version.mjs` | Delete |
-| `apps/claude-code/confluence-publish/scripts/tag.mjs` | Delete |
-| `apps/claude-code/confluence-publish/scripts/verify-changelog.mjs` | Delete |
-| `apps/claude-code/confluence-publish/scripts/verify-changelog.test.mjs` | Delete |
-| `apps/claude-code/confluence-publish/LICENSE` | Keep — maintained manually; do not delete |
-
-## Implementation steps
-
-1. Verify `git filter-repo` is installed (see spec 05 step 1 if needed).
-
-2. Fresh clone:
-
-   ```sh
-   git clone --no-local /Users/oriol.torrent/Sites/UNIC/unic-claude-code-confluence /tmp/confluence-publish-rewrite
-   ```
-
-3. Drop runtime artifacts:
-
-   ```sh
-   cd /tmp/confluence-publish-rewrite
-   git filter-repo --invert-paths --path .ralph --path .history --force
-   ```
-
-4. Move to subdirectory:
-
-   ```sh
-   git filter-repo --to-subdirectory-filter apps/claude-code/confluence-publish/ --force
-   ```
-
-5. Import into monorepo:
-
-   ```sh
-   cd /Users/oriol.torrent/Sites/UNIC/unic-agents-plugins
-   git remote add confluence-import /tmp/confluence-publish-rewrite
-   git fetch confluence-import
-   git merge --allow-unrelated-histories confluence-import/main \
-     -m "feat(spec-07): import confluence-publish plugin history"
-   git remote remove confluence-import
-   rm -rf /tmp/confluence-publish-rewrite
-   ```
-
-6. Delete per-repo tooling (superseded by monorepo root):
-
-   ```sh
-   rm -f  apps/claude-code/confluence-publish/.editorconfig
-   rm -f  apps/claude-code/confluence-publish/.nvmrc
-   rm -f  apps/claude-code/confluence-publish/.npmrc
-   rm -f  apps/claude-code/confluence-publish/pnpm-workspace.yaml
-   rm -f  apps/claude-code/confluence-publish/biome.json
-   rm -rf apps/claude-code/confluence-publish/.github
-   rm -rf apps/claude-code/confluence-publish/.githooks
-   rm -f  apps/claude-code/confluence-publish/scripts/bump-version.mjs
-   rm -f  apps/claude-code/confluence-publish/scripts/sync-version.mjs
-   rm -f  apps/claude-code/confluence-publish/scripts/tag.mjs
-   rm -f  apps/claude-code/confluence-publish/scripts/verify-changelog.mjs
-   rm -f  apps/claude-code/confluence-publish/scripts/verify-changelog.test.mjs
-   ```
-
-7. ⚠️ **Maintainer action required:** After the merge, verify that `apps/claude-code/confluence-publish/LICENSE` exists. If the import did not bring it over, copy the root `LICENSE` file there manually. Ralph/Claude must **not** create or delete LICENSE files.
-
-8. Update `apps/claude-code/confluence-publish/.claude-plugin/plugin.json`:
-   - `"name"`: `"unic-confluence"` → `"confluence-publish"`
-   - Leave version as `"2.4.1"` for now
-
-9. Update `apps/claude-code/confluence-publish/.claude-plugin/marketplace.json`:
-   - `"name"`: → `"confluence-publish"`
-   - `"homepage"`: → `"https://github.com/unic/unic-agents-plugins"`
-
-10. Rewrite `apps/claude-code/confluence-publish/package.json`. Preserve `confluence`, `test`, `typecheck` scripts. Replace release scripts. Remove tooling scripts now at root:
-
-   ```json
-   {
-     "name": "confluence-publish",
-     "version": "2.4.1",
-     "private": true,
-     "license": "LGPL-3.0-or-later",
-     "type": "module",
-     "packageManager": "pnpm@10.33.0",
-     "engines": {
-       "node": ">=24",
-       "pnpm": ">=10"
-     },
-     "scripts": {
-       "confluence": "node scripts/push-to-confluence.mjs",
-       "test": "node --test scripts/lib/frontmatter.test.mjs scripts/lib/inject.test.mjs scripts/lib/resolve.test.mjs scripts/lib/slug.test.mjs scripts/lib/pages-file.test.mjs scripts/lib/postprocess.test.mjs scripts/lib/sanitise.test.mjs scripts/lib/url.test.mjs",
-       "typecheck": "tsc --noEmit --project tsconfig.json",
-       "bump": "unic-bump",
-       "sync-version": "unic-sync-version",
-       "tag": "unic-tag",
-       "verify:changelog": "unic-verify-changelog",
-       "ralph": "ralph run -c ralph.yml -H builtin:code-assist"
-     },
-     "dependencies": {
-       "marked": "17.0.5"
-     },
-     "devDependencies": {
-       "@ralph-orchestrator/ralph-cli": "catalog:",
-       "@types/node": "catalog:",
-       "@unic/release-tools": "workspace:*",
-       "@unic/tsconfig": "workspace:*",
-       "typescript": "catalog:"
-     }
-   }
-   ```
-
-   > Note: the test script removes `verify-changelog.test.mjs` since that's now in release-tools.
-
-11. Create `apps/claude-code/confluence-publish/tsconfig.json`:
-
-    ```json
-    {
-      "extends": "@unic/tsconfig/tsconfig.base.json",
-      "include": ["scripts/**/*.mjs"]
-    }
-    ```
-
-12. Run `pnpm install`.
-
-13. Update the command file name to match the new plugin name. In `apps/claude-code/confluence-publish/commands/`, rename `unic-confluence.md` → `confluence-publish.md` and update the command title/description inside the file to reference `confluence-publish`.
-
-14. Update `apps/claude-code/confluence-publish/.claude-plugin/plugin.json` `commands` array to reference `./commands/confluence-publish.md`.
-
-15. Add CHANGELOG entry and bump:
-
-    In `apps/claude-code/confluence-publish/CHANGELOG.md`, under `## [Unreleased] → ### Added`:
-    ```
-    - Migrated to unic-agents-plugins monorepo; plugin renamed from `unic-confluence` to `confluence-publish`
-    ```
-
-    ```sh
-    pnpm --filter confluence-publish bump patch
-    pnpm --filter confluence-publish verify:changelog
-    ```
-
-16. Run `pnpm --filter confluence-publish test`.
-
-17. Commit:
-
-    ```sh
-    git add -A
-    git commit -m "feat(spec-07): finalize confluence-publish migration (v2.4.2)"
-    ```
-
-## Verification
-
-```sh
-pnpm install
-pnpm --filter confluence-publish test
-pnpm --filter confluence-publish typecheck
-pnpm --filter confluence-publish verify:changelog
-cat apps/claude-code/confluence-publish/.claude-plugin/plugin.json | grep '"name"'
-# → "name": "confluence-publish"
-```
-
-## Acceptance criteria
-
-- [ ] `apps/claude-code/confluence-publish/` exists with original commit history
-- [ ] `plugin.json` has `"name": "confluence-publish"` and version `"2.1.6"` (see Deviations)
-- [ ] `marketplace.json` has `"name": "confluence-publish"`
-- [ ] `package.json` uses `unic-bump` etc.
-- [ ] Per-repo tooling files deleted
-- [ ] `LICENSE` exists in the plugin dir (add manually if missing — Ralph does not manage this file)
-- [ ] `scripts/push-to-confluence.mjs` and `scripts/lib/` preserved
-- [ ] `commands/confluence-publish.md` exists (renamed from `unic-confluence.md`)
-- [ ] `pnpm --filter confluence-publish test` passes
-- [ ] `pnpm --filter confluence-publish verify:changelog` passes
-
-## Out of scope
-
-- Updating `CLAUDE.md` inside the plugin for monorepo context
-- Changing any push-to-confluence business logic
-
-## Deviations
-
-- **Version mismatch**: The spec expected the imported `plugin.json` version to be `"2.4.1"` (Current behaviour section), but the actual imported version was `"2.1.5"`. The `package.json` rewrite keeps `"2.1.5"` as the base. After the patch bump in step 15, the final version is `"2.1.6"` rather than `"2.4.2"` as stated in the acceptance criteria. The acceptance criteria should be interpreted as `"2.1.6"`.
-- **Test script preserved from import**: The spec's step 10 test script includes `postprocess.test.mjs`, `sanitise.test.mjs`, and `url.test.mjs` which do not exist in the imported history. The existing test script (covering the 5 test files that do exist) was preserved instead.
-- **`scripts/verify-changelog.test.mjs` absent**: Listed in step 6 delete list but not present in the imported history — no action needed.
diff --git a/docs/plans/08-root-marketplace.md b/docs/plans/08-root-marketplace.md
deleted file mode 100644
index 40e48c92..00000000
--- a/docs/plans/08-root-marketplace.md
+++ /dev/null
@@ -1,82 +0,0 @@
-# 08. Root Marketplace Manifest
-
-**Status: done — 2026-04-29**
-
-**Priority:** P0
-**Effort:** S
-**Version impact:** none
-**Depends on:** 05, 06, 07
-**Touches:** `.claude-plugin/marketplace.json`
-
-## Context
-
-Claude Code's marketplace mechanism allows a single JSON file to list multiple plugins. Once all three plugins are in the monorepo (specs 05–07), this spec creates the root `.claude-plugin/marketplace.json` that users add as a single marketplace source to access all Unic plugins.
-
-## Current behaviour
-
-`.claude-plugin/marketplace.json` does not exist. Users would need to add three separate marketplaces.
-
-## Target behaviour
-
-`.claude-plugin/marketplace.json` lists all three Claude Code plugins. Users run:
-
-```sh
-claude marketplace add <github-url-or-local-path>
-# then:
-claude plugins install pr-review
-claude plugins install auto-format
-claude plugins install confluence-publish
-```
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `.claude-plugin/marketplace.json` | Create |
-
-## Implementation steps
-
-1. Read the Claude Code plugin documentation or an existing marketplace.json from one of the old repos to confirm the multi-plugin marketplace format. The expected shape is an array of plugin entries, each with a `source` pointing to the plugin's directory relative to the marketplace file.
-
-   > If the format turns out to be a single-plugin manifest (not an array), create individual per-plugin marketplace.json files instead, and document the per-plugin install commands. Add a `## Deviations` section and proceed.
-
-2. Assuming the multi-plugin format is an object with a `plugins` array, create `.claude-plugin/marketplace.json`:
-
-   ```json
-   {
-     "name": "Unic Agent Plugins",
-     "description": "Official Unic AI agent plugins",
-     "plugins": [
-       {
-         "source": "apps/claude-code/pr-review"
-       },
-       {
-         "source": "apps/claude-code/auto-format"
-       },
-       {
-         "source": "apps/claude-code/confluence-publish"
-       }
-     ]
-   }
-   ```
-
-3. Update `README.md` — replace the `# coming after spec 09` placeholder in the "Installing plugins" section with the real marketplace install command. Use a placeholder GitHub URL (`https://github.com/unic/unic-agents-plugins`) since the repo may not be public yet.
-
-## Verification
-
-```sh
-ls .claude-plugin/marketplace.json   # file exists
-cat .claude-plugin/marketplace.json  # valid JSON, lists 3 plugins
-```
-
-## Acceptance criteria
-
-- [ ] `.claude-plugin/marketplace.json` exists and is valid JSON
-- [ ] All three plugins (`pr-review`, `auto-format`, `confluence-publish`) are listed
-- [ ] Each entry has a `source` path pointing to the plugin directory
-- [ ] `README.md` install snippet is updated
-
-## Out of scope
-
-- Actually publishing or testing the marketplace install (that's spec 11)
-- Adding plugins for other agent targets (future work)
diff --git a/docs/plans/09-ci-workflow.md b/docs/plans/09-ci-workflow.md
deleted file mode 100644
index d7376c99..00000000
--- a/docs/plans/09-ci-workflow.md
+++ /dev/null
@@ -1,150 +0,0 @@
-# 09. CI Workflow
-**Status: done — 2026-04-29**
-
-**Priority:** P0
-**Effort:** M
-**Version impact:** none
-**Depends on:** 08
-**Touches:** `.github/workflows/ci.yml`
-
-## Context
-
-Each old repo had its own CI workflow. The monorepo needs a single workflow that runs only the checks relevant to changed packages (path filtering), avoiding unnecessary CI cost. The matrix covers Ubuntu + macOS + Windows × Node 22 + 24 to satisfy the cross-platform requirement.
-
-## Current behaviour
-
-`.github/workflows/ci.yml` does not exist in the monorepo root. Per-plugin `.github/` directories were removed during migration (specs 05–07).
-
-## Target behaviour
-
-`.github/workflows/ci.yml` triggers on push to `main` and on PRs. It uses `dorny/paths-filter` to detect which packages changed, then runs a job matrix limited to the affected packages. Each job runs on all 3 OSes × 2 Node versions (6 jobs per affected package).
-
-Each job runs:
-1. `pnpm install --frozen-lockfile`
-2. `pnpm --filter <name> test` (if package has tests)
-3. `pnpm --filter <name> typecheck` (if package has typecheck script)
-4. `pnpm --filter <name> verify:changelog` (on PRs only)
-
-Root-level checks (Biome + Prettier) run once on Ubuntu/Node 24 regardless of which package changed.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `.github/workflows/ci.yml` | Create |
-
-## Implementation steps
-
-1. Create `.github/workflows/ci.yml`:
-
-```yaml
-name: CI
-
-on:
-  push:
-    branches: [main]
-  pull_request:
-
-jobs:
-  root-checks:
-    name: Root checks (Biome + Prettier)
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-      - uses: pnpm/action-setup@v4
-      - uses: actions/setup-node@v4
-        with:
-          node-version-file: .nvmrc
-          cache: pnpm
-      - run: pnpm install --frozen-lockfile
-      - run: pnpm ci:check
-
-  changes:
-    name: Detect changed packages
-    runs-on: ubuntu-latest
-    outputs:
-      pr-review: ${{ steps.filter.outputs.pr-review }}
-      auto-format: ${{ steps.filter.outputs.auto-format }}
-      confluence-publish: ${{ steps.filter.outputs.confluence-publish }}
-      release-tools: ${{ steps.filter.outputs.release-tools }}
-    steps:
-      - uses: actions/checkout@v4
-      - uses: dorny/paths-filter@v3
-        id: filter
-        with:
-          filters: |
-            pr-review:
-              - 'apps/claude-code/pr-review/**'
-            auto-format:
-              - 'apps/claude-code/auto-format/**'
-            confluence-publish:
-              - 'apps/claude-code/confluence-publish/**'
-            release-tools:
-              - 'packages/release-tools/**'
-
-  test:
-    name: Test ${{ matrix.package }} / ${{ matrix.os }} / Node ${{ matrix.node }}
-    needs: changes
-    if: |
-      needs.changes.outputs.auto-format == 'true' ||
-      needs.changes.outputs.confluence-publish == 'true' ||
-      needs.changes.outputs.release-tools == 'true'
-    strategy:
-      fail-fast: false
-      matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
-        node: ['22', '24']
-        package:
-          - name: auto-format
-            changed: ${{ needs.changes.outputs.auto-format }}
-          - name: confluence-publish
-            changed: ${{ needs.changes.outputs.confluence-publish }}
-          - name: release-tools
-            changed: ${{ needs.changes.outputs.release-tools }}
-        exclude:
-          - package:
-              changed: 'false'
-    runs-on: ${{ matrix.os }}
-    steps:
-      - uses: actions/checkout@v4
-      - uses: pnpm/action-setup@v4
-      - uses: actions/setup-node@v4
-        with:
-          node-version: ${{ matrix.node }}
-          cache: pnpm
-      - run: pnpm install --frozen-lockfile
-      - run: pnpm --filter ${{ matrix.package.name }} --if-present test
-      - run: pnpm --filter ${{ matrix.package.name }} --if-present typecheck
-      - if: github.event_name == 'pull_request'
-        run: pnpm --filter ${{ matrix.package.name }} --if-present verify:changelog
-```
-
-   > Note: `pr-review` has no tests or typecheck, so it is intentionally excluded from the test matrix. It is still subject to root Biome checks if its `.md` files change.
-
-2. Run `pnpm check` to verify no Prettier issues in the new YAML file (YAML is excluded from Biome and Prettier in `.prettierignore`, so this is a no-op — but confirm).
-
-## Verification
-
-Commit the file and open a draft PR (or push a branch) to trigger the workflow. Alternatively, use `act` locally if available:
-
-```sh
-# Confirm the workflow file is valid YAML
-node -e "import('node:fs').then(fs => { JSON.stringify(fs.readFileSync('.github/workflows/ci.yml', 'utf8')); console.log('readable') })"
-```
-
-Full verification happens when the GitHub repo is set up (spec 11).
-
-## Acceptance criteria
-
-- [ ] `.github/workflows/ci.yml` exists and is valid YAML
-- [ ] Workflow triggers on push to `main` and on PRs
-- [ ] Root checks job runs Biome + Prettier
-- [ ] `dorny/paths-filter` detects per-package changes
-- [ ] Test matrix covers Ubuntu + macOS + Windows × Node 22 + 24
-- [ ] `verify:changelog` runs on PRs only
-
-## Out of scope
-
-- Actually running the workflow (requires GitHub remote — spec 11)
-- Release workflow (spec 10)
-- Adding new packages to the filter list (handled when new packages are added)
diff --git a/docs/plans/10-release-workflow.md b/docs/plans/10-release-workflow.md
deleted file mode 100644
index dc1770b9..00000000
--- a/docs/plans/10-release-workflow.md
+++ /dev/null
@@ -1,108 +0,0 @@
-# 10. Release Workflow
-**Status: done — 2026-04-29**
-
-**Priority:** P0
-**Effort:** M
-**Version impact:** none
-**Depends on:** 09
-**Touches:** `.github/workflows/release.yml`
-
-## Context
-
-The old `unic-claude-code-format` repo had a `release.yml` that auto-created a git tag whenever it detected a version bump in `plugin.json` on push to `main`. This spec brings that pattern to the monorepo, extended to detect which plugin's `plugin.json` changed and to emit `<name>@<version>` tags.
-
-## Current behaviour
-
-`.github/workflows/release.yml` does not exist.
-
-## Target behaviour
-
-On every push to `main`, the workflow checks if any plugin's `.claude-plugin/plugin.json` version differs from the previous commit. For each one that does, it creates and pushes a git tag `<plugin-name>@<version>` (e.g. `auto-format@0.5.9`). Supports optional GPG signing via the `UNIC_SIGN_TAGS` repository secret.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `.github/workflows/release.yml` | Create |
-
-## Implementation steps
-
-1. Read `~/Sites/UNIC/unic-claude-code-format/.github/workflows/release.yml` to understand the current version-diff detection logic. Adapt it for three plugins.
-
-2. Create `.github/workflows/release.yml`:
-
-```yaml
-name: Release
-
-on:
-  push:
-    branches: [main]
-
-jobs:
-  tag:
-    name: Tag released plugins
-    runs-on: ubuntu-latest
-    permissions:
-      contents: write
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          fetch-depth: 2  # need HEAD and HEAD~1 to diff
-
-      - name: Detect version changes and tag
-        env:
-          SIGN_TAGS: ${{ secrets.UNIC_SIGN_TAGS }}
-        run: |
-          tag_if_changed() {
-            local plugin_dir="$1"
-            local json_path="$plugin_dir/.claude-plugin/plugin.json"
-
-            if [ ! -f "$json_path" ]; then return; fi
-
-            current=$(node -p "JSON.parse(require('fs').readFileSync('$json_path','utf8')).version")
-            previous=$(git show HEAD~1:"$json_path" 2>/dev/null \
-              | node -p "JSON.parse(require('fs').readFileSync('/dev/stdin','utf8')).version" \
-              || echo "")
-
-            if [ "$current" = "$previous" ]; then return; fi
-
-            name=$(node -p "JSON.parse(require('fs').readFileSync('$json_path','utf8')).name")
-            tag="${name}@${current}"
-
-            echo "Tagging $tag"
-            if [ "$SIGN_TAGS" = "true" ]; then
-              git tag -s "$tag" -m "Release $tag"
-            else
-              git tag "$tag"
-            fi
-            git push origin "$tag"
-          }
-
-          tag_if_changed "apps/claude-code/pr-review"
-          tag_if_changed "apps/claude-code/auto-format"
-          tag_if_changed "apps/claude-code/confluence-publish"
-```
-
-   > Note: uses Node.js (already available) to parse JSON instead of `jq` to stay cross-platform in principle, though this job only runs on ubuntu-latest.
-
-## Verification
-
-Full verification requires a GitHub remote (spec 11). Locally, confirm the workflow file is valid YAML and that the Node.js JSON parsing commands work:
-
-```sh
-node -p "JSON.parse(require('fs').readFileSync('apps/claude-code/auto-format/.claude-plugin/plugin.json','utf8')).version"
-# prints current version of auto-format
-```
-
-## Acceptance criteria
-
-- [ ] `.github/workflows/release.yml` exists and is valid YAML
-- [ ] Workflow triggers only on push to `main`
-- [ ] Tags use `<name>@<version>` scheme
-- [ ] Supports `UNIC_SIGN_TAGS` secret for GPG signing
-- [ ] All three plugins are checked
-
-## Out of scope
-
-- Testing the actual tag creation (requires GitHub remote — spec 11)
-- Adding new plugins to the tag list (handled when new plugins are added)
diff --git a/docs/plans/11-smoke-tests.md b/docs/plans/11-smoke-tests.md
deleted file mode 100644
index 60d401ad..00000000
--- a/docs/plans/11-smoke-tests.md
+++ /dev/null
@@ -1,113 +0,0 @@
-# 11. Cross-Platform Smoke Tests
-**Status: done — 2026-04-29**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** none
-**Depends on:** 10
-**Touches:** `docs/process/smoke-test-checklist.md`
-
-## Context
-
-Before cutting the first official releases, verify that each plugin installs and runs correctly from the new marketplace on a real Claude Code instance. This spec is documentation-heavy: it defines the smoke test checklist and walks through executing it on macOS (local). Windows validation is delegated to CI once the GitHub repo is live.
-
-## Current behaviour
-
-No smoke test documentation exists. Plugins have only been run from their individual repos.
-
-## Target behaviour
-
-`docs/process/smoke-test-checklist.md` contains a step-by-step checklist for validating each plugin from the new marketplace. The checklist has been executed locally on macOS and all items checked off in a `## Results` section of the spec file.
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `docs/process/smoke-test-checklist.md` | Create |
-| `docs/plans/11-smoke-tests.md` | Modify — add `## Results` section |
-
-## Implementation steps
-
-1. Create `docs/process/smoke-test-checklist.md` with sections for each plugin. Template:
-
-   ```markdown
-   # Smoke Test Checklist
-
-   Run these tests from a scratch Claude Code environment (a consumer repo that doesn't already have these plugins installed).
-
-   ## Prerequisites
-
-   - Claude Code CLI installed and authenticated
-   - A test Confluence instance with credentials (for confluence-publish)
-   - A test Azure DevOps PR URL (for pr-review)
-   - A project with Prettier + ESLint configured (for auto-format)
-
-   ## Marketplace setup
-
-   - [ ] `claude marketplace add <unic-agents-plugins-url>` succeeds
-   - [ ] `claude plugins list` shows `pr-review`, `auto-format`, `confluence-publish`
-
-   ## auto-format
-
-   - [ ] `claude plugins install auto-format` succeeds
-   - [ ] Open a consumer project in Claude Code
-   - [ ] Ask Claude to edit a JavaScript file — verify Prettier runs automatically after the edit
-   - [ ] Ask Claude to edit a file with ESLint violations — verify ESLint runs
-   - [ ] On Windows: repeat the above (via CI matrix in `.github/workflows/ci.yml`)
-
-   ## pr-review
-
-   - [ ] `claude plugins install pr-review` succeeds
-   - [ ] Run `/pr-review` with a valid Azure DevOps PR URL
-   - [ ] Verify a review comment is produced
-
-   ## confluence-publish
-
-   - [ ] `claude plugins install confluence-publish` succeeds
-   - [ ] Run `/confluence-publish` with a test Markdown file and a valid page ID
-   - [ ] Verify the Confluence page is updated
-
-   ## Cleanup
-
-   - [ ] `claude plugins uninstall auto-format` succeeds
-   - [ ] `claude plugins uninstall pr-review` succeeds
-   - [ ] `claude plugins uninstall confluence-publish` succeeds
-   ```
-
-2. Execute the checklist locally on macOS. For each item that passes, check it off. For items that require a live GitHub remote or Windows machine (CI items), note "pending CI" in the checklist.
-
-3. Add a `## Results` section at the bottom of this spec file summarising what passed, what is pending CI, and any issues found and resolved.
-
-## Verification
-
-```sh
-cat docs/process/smoke-test-checklist.md  # file exists and is non-empty
-# At least the marketplace setup and one plugin's local items are checked off
-```
-
-## Acceptance criteria
-
-- [ ] `docs/process/smoke-test-checklist.md` exists
-- [ ] Checklist covers all three plugins and the marketplace setup
-- [ ] Local macOS checks have been executed (even if some items are "pending CI")
-- [ ] `## Results` section added to this spec
-
-## Out of scope
-
-- Automated test scripts for the smoke tests (manual checklist is sufficient)
-- Windows smoke tests before CI is live (those run automatically once spec 10 is done and the repo is on GitHub)
-
-## Results
-
-Executed on macOS 2026-04-29.
-
-**Passed (local):**
-- None. No checklist items could be executed locally because all items require a live marketplace URL (the GitHub repo must be published first).
-
-**Pending CI / pending live environment:**
-- All checklist items depended on the GitHub repo being live. The repo was private at the time of the initial spec execution (2026-04-29).
-
-**Issues found:** None. The checklist file is well-formed and covers all three plugins.
-
-**Update — 2026-04-30:**
-GitHub repo is now public at `https://github.com/unic/unic-agents-plugins`. Checklist updated with the real marketplace URL (placeholder removed). All items still need to be checked off manually in a scratch Claude Code environment — none have been executed yet.
diff --git a/docs/plans/12-first-releases.md b/docs/plans/12-first-releases.md
deleted file mode 100644
index 26241c2a..00000000
--- a/docs/plans/12-first-releases.md
+++ /dev/null
@@ -1,82 +0,0 @@
-# 12. First Releases
-**Status: done — 2026-04-30**
-
-**Priority:** P0
-**Effort:** S
-**Version impact:** none (versions were already bumped in specs 05–07)
-**Depends on:** 11
-**Touches:** Nothing new — this spec creates tags
-
-## Context
-
-Specs 05–07 bumped each plugin to its first monorepo version (`pr-review@0.1.1`, `auto-format@0.5.9`, `confluence-publish@2.4.2`). This spec pushes those commits to the GitHub remote and confirms that the release workflow (spec 10) fires and creates the correct tags.
-
-## Current behaviour
-
-The monorepo exists locally with all commits. No GitHub remote has been pushed to yet.
-
-## Target behaviour
-
-- `origin` points to `https://github.com/unic/unic-agents-plugins`
-- All commits pushed to `main`
-- Release workflow fires and creates tags: `pr-review@0.1.1`, `auto-format@0.5.9`, `confluence-publish@2.4.2`
-- Tags are visible in the GitHub Releases / Tags page
-
-## Affected files
-
-No file changes. This spec is a sequence of git/GitHub operations.
-
-## Implementation steps
-
-1. Create the GitHub repository `unic/unic-agents-plugins` (private initially if preferred, can be made public later). Do this via the GitHub web UI or `gh repo create`.
-
-2. Add the remote:
-
-   ```sh
-   git remote add origin https://github.com/unic/unic-agents-plugins.git
-   ```
-
-3. Push:
-
-   ```sh
-   git push -u origin main
-   ```
-
-4. Confirm the CI workflow (spec 09) runs and passes on the push to `main`.
-
-5. Confirm the release workflow (spec 10) runs and creates the three tags.
-
-6. Verify locally:
-
-   ```sh
-   git ls-remote origin 'refs/tags/*'
-   # should list pr-review@0.1.1, auto-format@0.5.9, confluence-publish@2.4.2
-   ```
-
-## Verification
-
-```sh
-gh run list --repo unic/unic-agents-plugins --limit 5  # shows CI + Release runs
-git ls-remote origin 'refs/tags/*'                     # shows all 3 release tags
-```
-
-## Acceptance criteria
-
-- [x] GitHub repo `unic/unic-agents-plugins` exists
-- [x] `main` branch is pushed
-- [x] CI workflow passes (Windows fixed via optionalDependencies)
-- [x] Three release tags created: `pr-review@0.1.1`, `auto-format@0.5.5`, `confluence-publish@2.1.6` (versions differ from spec — see Deviations)
-- [x] Tags visible on GitHub
-
-## Out of scope
-
-- Making the repo public
-- Setting up branch protection rules
-- Adding GitHub Actions secrets for GPG signing (can be done separately)
-
-## Deviations
-
-- **Actual plugin versions differ from spec**: This spec doc stated `auto-format@0.5.9` and `confluence-publish@2.4.2`, but the actual versions after migration (specs 05–07) are `auto-format@0.5.5` and `confluence-publish@2.1.6`. The tags created are `pr-review@0.1.1`, `auto-format@0.5.5`, and `confluence-publish@2.1.6`.
-- **Release workflow did not auto-create tags**: The release workflow uses `fetch-depth: 2` and compared HEAD vs HEAD~1. Since all commits were pushed at once, the version bumps were in older commits and were not detected. Tags were created manually and pushed directly.
-- **Windows CI fix required**: `@ralph-orchestrator/ralph-cli` postinstall fails on Windows (no Windows binary). Moved from `devDependencies` to `optionalDependencies` so postinstall failure is non-fatal on Windows.
-- **Old-repo annotated tags on remote**: Tags `v0.1.0`–`v0.5.4` and `v2.3.5` (from the original plugin repos) were pushed automatically via `push.followTags=true`. These are expected migration artifacts.
diff --git a/docs/plans/13-archive-old-repos.md b/docs/plans/13-archive-old-repos.md
deleted file mode 100644
index a56a3bf4..00000000
--- a/docs/plans/13-archive-old-repos.md
+++ /dev/null
@@ -1,85 +0,0 @@
-# 13. Archive Old Repos
-
-**Status: done — 2026-04-30**
-
-**Priority:** P2
-**Effort:** XS
-**Version impact:** none
-**Depends on:** 12
-**Touches:** Nothing in this repo
-
-## Context
-
-The three original repos (`unic-pr-review`, `unic-claude-code-format`, `unic-claude-code-confluence`) are superseded by the monorepo. Archiving them prevents confusion and new development from happening in the wrong place. Since you are the only user, no redirect or deprecation notice is strictly required, but a final CHANGELOG entry pointing to the new home is good practice.
-
-## Current behaviour
-
-All three old repos are active on GitHub. No deprecation notices exist.
-
-## Target behaviour
-
-Each old repo has a final commit with a deprecation note in `README.md` and `CHANGELOG.md`. All three repos are archived on GitHub (read-only).
-
-## Affected files
-
-These changes are made in the **old repos**, not in this monorepo.
-
-## Implementation steps
-
-1. For each old repo (`unic-pr-review`, `unic-claude-code-format`, `unic-claude-code-confluence`):
-
-   a. In the repo's `README.md`, prepend:
-
-   ```markdown
-   > **Archived.** This plugin has moved to [unic/unic-agents-plugins](https://github.com/unic/unic-agents-plugins).
-   > Install via the new marketplace — see the monorepo README for instructions.
-   ```
-
-   b. In `CHANGELOG.md` (create one for `unic-pr-review` if it doesn't exist), add a final entry under `## [Unreleased]`:
-
-   ```markdown
-   ## [Unreleased]
-
-   ### Changed
-
-   - Migrated to [unic/unic-agents-plugins](https://github.com/unic/unic-agents-plugins) monorepo.
-     Plugin renamed to `<new-name>`. Install from the new marketplace going forward.
-   ```
-
-   c. Commit: `chore: deprecate — migrated to unic-agents-plugins`
-
-   d. Push.
-
-2. Archive each repo on GitHub:
-   - Go to repo **Settings → General → Danger Zone → Archive this repository**
-   - Confirm. The repo becomes read-only.
-
-3. Back in this monorepo, mark this spec done and commit:
-
-   ```sh
-   git add docs/plans/13-archive-old-repos.md
-   git commit -m "chore(spec-13): archive old repos complete"
-   ```
-
-## Verification
-
-```sh
-gh repo view unic/unic-pr-review --json isArchived --jq .isArchived          # true
-gh repo view unic/unic-claude-code-format --json isArchived --jq .isArchived  # true
-gh repo view unic/unic-claude-code-confluence --json isArchived --jq .isArchived  # true
-```
-
-## Acceptance criteria
-
-- [ ] Each old repo has a deprecation note in `README.md`
-- [ ] Each old repo's `CHANGELOG.md` has a final migration entry
-- [ ] All three old repos are archived on GitHub (read-only)
-
-## Out of scope
-
-- Deleting the old repos (archive is safer — keep them for historical reference)
-- Updating any external references or documentation that may link to the old repos
-
-## Deviations
-
-- **`unic-pr-review` had no GitHub remote.** This repo existed only locally and was never pushed to GitHub, so it could not be deprecated (no README/CHANGELOG to update) and could not be archived. Only 2 of the 3 target repos (`unic-claude-code-format` and `unic-claude-code-confluence`) were archived. The acceptance criterion "All three old repos are archived on GitHub" is partially unmet due to this pre-existing condition.
diff --git a/docs/plans/14-fix-release-workflow.md b/docs/plans/14-fix-release-workflow.md
deleted file mode 100644
index 2eca9a48..00000000
--- a/docs/plans/14-fix-release-workflow.md
+++ /dev/null
@@ -1,110 +0,0 @@
-# 14. Fix Release Workflow Tag Detection
-**Status: done — 2026-04-30**
-
-**Priority:** P0
-**Effort:** S
-**Version impact:** none (workflow fix only)
-**Depends on:** 12
-**Touches:** `.github/workflows/release.yml`
-
-## Context
-
-Spec 12 revealed that the release workflow (spec 10) will not create tags when multiple commits are pushed at once. The current logic fetches only the last two commits (`fetch-depth: 2`) and compares the plugin version in `HEAD` vs `HEAD~1`. If the version bump landed in an older commit, the diff is empty and no tag is created. In spec 12, all three tags had to be pushed manually.
-
-The fix is to switch from a diff-based approach to an existence-based one: fetch all tags, then create a tag only if `<name>@<version>` does not yet exist in the remote. This is also idempotent — re-running the workflow never creates a duplicate tag.
-
-## Current behaviour
-
-`.github/workflows/release.yml`:
-- `fetch-depth: 2` — fetches only the last two commits
-- Reads `plugin.json` version from `HEAD` and `HEAD~1`
-- Tags if the two versions differ
-- Silently skips if the bump was more than one commit ago
-
-## Target behaviour
-
-`.github/workflows/release.yml`:
-- `fetch-depth: 0` — fetches full history and all tags
-- For each plugin: reads current version from `plugin.json`, builds the expected tag name (`<name>@<version>`), checks whether that tag already exists
-- Creates and pushes the tag only if it does not exist
-- Idempotent: safe to re-run on any commit
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `.github/workflows/release.yml` | Replace `fetch-depth: 2` with `fetch-depth: 0`; rewrite `tag_if_changed` shell function |
-
-## Implementation steps
-
-1. In `.github/workflows/release.yml`, change the checkout step:
-
-   ```yaml
-   - uses: actions/checkout@v4
-     with:
-       fetch-depth: 0  # need all tags to detect existing releases
-   ```
-
-2. Replace the `tag_if_changed` shell function body. Remove the `previous` variable and diff logic. Replace with a tag-existence check:
-
-   ```bash
-   tag_if_changed() {
-     local plugin_dir="$1"
-     local json_path="$plugin_dir/.claude-plugin/plugin.json"
-
-     if [ ! -f "$json_path" ]; then return; fi
-
-     current=$(node -p "JSON.parse(require('fs').readFileSync('$json_path','utf8')).version")
-     name=$(node -p "JSON.parse(require('fs').readFileSync('$json_path','utf8')).name")
-     tag="${name}@${current}"
-
-     if git rev-parse --verify "refs/tags/$tag" >/dev/null 2>&1; then
-       echo "Tag $tag already exists, skipping"
-       return
-     fi
-
-     echo "Tagging $tag"
-     if [ "$SIGN_TAGS" = "true" ]; then
-       git tag -s "$tag" -m "Release $tag"
-     else
-       git tag "$tag"
-     fi
-     git push origin "$tag"
-   }
-   ```
-
-3. Remove the inline comment `# need HEAD and HEAD~1 to diff` (no longer accurate).
-
-4. Commit:
-
-   ```sh
-   git commit -m "fix(release-workflow): detect existing tags instead of diffing HEAD~1"
-   ```
-
-## Verification
-
-```sh
-# Locally: confirm the workflow file has fetch-depth: 0 and no HEAD~1 reference
-grep "fetch-depth" .github/workflows/release.yml
-# → fetch-depth: 0
-
-grep "HEAD~1" .github/workflows/release.yml
-# → (no output)
-
-# After merging: push a version bump and confirm the workflow creates the tag automatically
-# without manual intervention
-```
-
-## Acceptance criteria
-
-- [x] `fetch-depth: 0` in the checkout step
-- [x] No reference to `HEAD~1` in the workflow
-- [x] `tag_if_changed` uses `git rev-parse --verify "refs/tags/$tag"` to skip existing tags
-- [ ] Workflow re-runs on a commit where no version changed produce no new tags (idempotent) — verified on next push
-- [ ] Next real version bump triggers a tag automatically without manual intervention — verified on next release
-
-## Out of scope
-
-- Adding changelog automation or GitHub Release notes
-- Changing the tag name scheme (`<name>@<version>`)
-- Adding GPG signing secrets (separate concern)
diff --git a/docs/plans/15-release-tools-tests.md b/docs/plans/15-release-tools-tests.md
deleted file mode 100644
index 25958b29..00000000
--- a/docs/plans/15-release-tools-tests.md
+++ /dev/null
@@ -1,89 +0,0 @@
-# 15. Release Tools Test Coverage
-**Status: done — 2026-05-02**
-
-**Priority:** P1
-**Effort:** S
-**Version impact:** none
-**Depends on:** 03
-**Touches:** `packages/release-tools/scripts/`
-
-## Context
-
-`verify-changelog.mjs` already has a test file (`verify-changelog.test.mjs`). The other two scripts with non-trivial logic — `bump-version.mjs` and `sync-version.mjs` — have no tests. This spec adds coverage for both, following the same subprocess-based pattern already established by `verify-changelog.test.mjs`.
-
-`tag.mjs` is excluded: its core logic (read name+version, run sync, call `git tag`) is tested indirectly through integration; the git interaction makes unit testing add little value without a real repo context.
-
-## Current behaviour
-
-- `packages/release-tools/scripts/verify-changelog.test.mjs` exists — 9 test cases
-- `packages/release-tools/scripts/bump-version.test.mjs` — does not exist
-- `packages/release-tools/scripts/sync-version.test.mjs` — does not exist
-- `package.json` test script only references `verify-changelog.test.mjs`
-
-## Target behaviour
-
-- `bump-version.test.mjs` covers: invalid type, dirty tree, missing files, malformed version, no real changelog entries, missing [Unreleased] section, patch/minor/major version arithmetic, CHANGELOG promotion
-- `sync-version.test.mjs` covers: version sync to marketplace + package.json, no-op when up to date, error on missing files, malformed version, field mirroring (license, homepage, keywords)
-- `package.json` test script runs all three test files
-
-## Affected files
-
-| File | Change |
-|---|---|
-| `packages/release-tools/scripts/bump-version.test.mjs` | Create |
-| `packages/release-tools/scripts/sync-version.test.mjs` | Create |
-| `packages/release-tools/package.json` | Modify — extend test script |
-
-## Implementation steps
-
-1. Create `packages/release-tools/scripts/bump-version.test.mjs` with test cases listed in Acceptance criteria. Use the same subprocess-based pattern as `verify-changelog.test.mjs`: create a temp dir via `mkdtempSync`, write fixture files, run the script via `spawnSync` with `cwd: tmpDir` (since `bump-version.mjs` resolves paths from `process.cwd()`). For tests that need a clean or dirty git state, inject cross-platform fake `git` stubs into the temp dir: a POSIX shell script (`git`) for macOS/Linux and a CMD batch file (`git.cmd`) for Windows. Prepend the temp dir to PATH using `path.delimiter` (not a hard-coded `:`), matching the approach in `verify-changelog.test.mjs` but extended to cover Windows. Happy-path bump tests also require a `.claude-plugin/marketplace.json` fixture in the temp dir (consumed by `sync-version.mjs`).
-
-2. Create `packages/release-tools/scripts/sync-version.test.mjs` with test cases listed in Acceptance criteria.
-
-3. Update `packages/release-tools/package.json` test script to include all three test files:
-   ```json
-   "test": "node --test scripts/verify-changelog.test.mjs scripts/bump-version.test.mjs scripts/sync-version.test.mjs"
-   ```
-
-4. Run `pnpm --filter @unic/release-tools test` and verify all tests pass.
-
-5. Commit: `test(release-tools): add bump-version and sync-version tests (spec-15)`
-
-## Verification
-
-```sh
-pnpm --filter @unic/release-tools test
-# All test cases pass; CI validates macOS, Windows, and Linux
-```
-
-## Acceptance criteria
-
-### bump-version.test.mjs
-- [ ] `exits 1 with usage message when bump type is invalid`
-- [ ] `exits 1 when working tree is dirty`
-- [ ] `exits 1 when plugin.json is missing`
-- [ ] `exits 1 when version in plugin.json is malformed (e.g. "not-semver")`
-- [ ] `exits 1 when CHANGELOG.md is missing`
-- [ ] `exits 1 when CHANGELOG.md has no [Unreleased] section`
-- [ ] `exits 1 when [Unreleased] has no real entries (only "(none)")`
-- [ ] `patch bump: increments patch, resets nothing`
-- [ ] `minor bump: increments minor, resets patch to 0`
-- [ ] `major bump: increments major, resets minor and patch to 0`
-- [ ] `bump promotes [Unreleased] section and leaves a fresh empty [Unreleased] — released section date matches YYYY-MM-DD`
-
-### sync-version.test.mjs
-- [ ] `exits 1 when plugin.json is missing`
-- [ ] `exits 1 when .version is missing or not a string in plugin.json`
-- [ ] `exits 1 when marketplace.json has no plugins[] array`
-- [ ] `syncs version from plugin.json to marketplace.json`
-- [ ] `syncs version to package.json when it exists`
-- [ ] `reports no change when versions already match`
-- [ ] `mirrors license, homepage, and keywords fields from plugin.json into marketplace entry`
-
-### package.json
-- [ ] test script references all three test files
-
-## Out of scope
-
-- Tests for `tag.mjs` (git-heavy, integration-only)
-- Extracting a shared reusable fake-binary helper across test files — each test file can inline its own stubs
diff --git a/docs/plans/16-adr-backfill.md b/docs/plans/16-adr-backfill.md
deleted file mode 100644
index c6bfce55..00000000
--- a/docs/plans/16-adr-backfill.md
+++ /dev/null
@@ -1,1428 +0,0 @@
-# 16. Back-fill ADRs from existing specs and tooling
-
-**Status: done — 2026-05-02**
-
-**Priority:** P2
-**Effort:** L
-**Version impact:** none
-**Depends on:** none
-**Touches:** `docs/adr/`, `apps/claude-code/auto-format/docs/adr/`, `apps/claude-code/pr-review/docs/adr/`, `apps/claude-code/unic-confluence/docs/adr/`
-
-## Context
-
-The repo adopted Matt Pocock's engineering skills (`grill-with-docs`, `improve-codebase-architecture`, `diagnose`, `tdd`). Several skills consult `docs/adr/` to avoid re-litigating past decisions and to ground architectural suggestions. Today zero ADRs exist, yet ~74 Ralph specs plus tooling in `packages/`, CI in `.github/workflows/`, and process docs in `docs/process/` collectively encode ~54 architectural decisions worth surfacing.
-
-Without this back-fill the skills operate without grounding and may propose refactors that contradict already-accepted decisions.
-
-`docs/plans/` (forward-looking Ralph specs) and `docs/adr/` (backward-looking decision records) are different artefacts; this spec adds a parallel tree without touching any existing file.
-
-Format: **MADR-lite** — `# NN. Title` / `**Status:** Accepted (YYYY-MM)` / `## Context` / `## Decision` / `## Consequences`.
-
-## Current behaviour
-
-```
-docs/
-└── plans/   ← specs only; no adr/ directory
-apps/claude-code/
-├── auto-format/docs/    ← plans/ only
-├── pr-review/docs/      ← plans/ only
-└── unic-confluence/docs/← plans/ only
-```
-
-No `CONTEXT-MAP.md` or `CONTEXT.md` files exist (created lazily later via `/grill-with-docs`).
-
-## Target behaviour
-
-```
-docs/
-├── adr/
-│   ├── README.md
-│   ├── 0001-pnpm-workspaces-no-turborepo.md
-│   ├── 0002-shared-unic-packages.md
-│   ├── 0003-jsdoc-tsc-nocheck-no-compile.md
-│   ├── 0004-biome-prettier-split.md
-│   ├── 0005-indentation-policy.md
-│   ├── 0006-cross-platform-node-only-scripts.md
-│   ├── 0007-release-tools-bin-commands.md
-│   ├── 0008-tag-scheme-plugin-at-version.md
-│   ├── 0009-plugin-json-version-source-of-truth.md
-│   ├── 0010-pnpm-filter-bump-only-version-path.md
-│   ├── 0011-plugin-migration-via-git-filter-repo.md
-│   ├── 0012-license-files-manually-managed.md
-│   ├── 0013-single-root-marketplace.md
-│   ├── 0014-ci-matrix-three-os-two-node.md
-│   ├── 0015-ci-paths-filter-changed-packages.md
-│   ├── 0016-root-checks-unconditional.md
-│   ├── 0017-verify-changelog-pr-only.md
-│   ├── 0018-release-on-merge-tag-existence.md
-│   ├── 0019-gpg-signing-opt-in.md
-│   ├── 0020-per-plugin-ralph-loops.md
-│   ├── 0021-conventional-commits-package-scope.md
-│   ├── 0022-semver-per-plugin.md
-│   ├── 0023-spec-template-format.md
-│   └── 0024-ralph-atomic-iteration.md
-└── plans/   ← unchanged
-apps/claude-code/
-├── auto-format/docs/adr/
-│   ├── README.md
-│   ├── 0001-hook-always-exits-zero.md
-│   ├── 0002-zero-runtime-dependencies.md
-│   ├── 0003-consumer-owns-formatters.md
-│   ├── 0004-per-project-config-merge.md
-│   ├── 0005-spawnsync-timeout-kill-signal.md
-│   ├── 0006-posix-path-normalization-windows.md
-│   ├── 0007-ci-creates-tags-automatically.md
-│   └── 0008-jsdoc-types-ts-check.md
-├── pr-review/docs/adr/
-│   ├── README.md
-│   ├── 0001-canonical-bot-signature.md
-│   ├── 0002-signature-based-prior-review-detection.md
-│   ├── 0003-target-latest-pr-iteration.md
-│   ├── 0004-incremental-diff-baseline.md
-│   ├── 0005-four-state-thread-classification.md
-│   ├── 0006-reply-not-duplicate-auto-resolve.md
-│   ├── 0007-summary-rewritten-not-appended.md
-│   └── 0008-soft-dependency-pr-review-toolkit.md
-└── unic-confluence/docs/adr/
-    ├── README.md
-    ├── 0001-refuse-publish-without-markers.md
-    ├── 0002-three-strategy-injection-priority.md
-    ├── 0003-structured-macro-for-code-blocks.md
-    ├── 0004-dry-run-read-only.md
-    ├── 0005-ping-check-auth-over-per-page-verify.md
-    ├── 0006-hard-http-timeout.md
-    ├── 0007-clierror-class.md
-    ├── 0008-pure-functions-lib-with-tests.md
-    ├── 0009-bare-integer-page-id-schema.md
-    ├── 0010-no-catalog-for-runtime-deps.md
-    ├── 0011-alias-auto-population.md
-    └── 0012-do-not-add-scope-guard.md
-```
-
-## Affected files
-
-| File | Change |
-| ---- | ------ |
-| `docs/adr/README.md` | Create |
-| `docs/adr/0001-*.md` … `0024-*.md` | Create (24 files) |
-| `apps/claude-code/auto-format/docs/adr/README.md` | Create |
-| `apps/claude-code/auto-format/docs/adr/0001-*.md` … `0008-*.md` | Create (8 files) |
-| `apps/claude-code/pr-review/docs/adr/README.md` | Create |
-| `apps/claude-code/pr-review/docs/adr/0001-*.md` … `0008-*.md` | Create (8 files) |
-| `apps/claude-code/unic-confluence/docs/adr/README.md` | Create |
-| `apps/claude-code/unic-confluence/docs/adr/0001-*.md` … `0012-*.md` | Create (12 files) |
-
-## Implementation steps
-
-### Phase 1 — Scaffold
-
-**1.1** Create `docs/adr/README.md`:
-
-```markdown
-# Architectural Decision Records
-
-Back-filled from existing specs and tooling in 2025-05. Format: MADR-lite.
-
-## Format
-
-~~~markdown
-# NNNN. Title
-
-**Status:** Accepted (YYYY-MM)
-
-## Context
-
-Why this decision had to be made.
-
-## Decision
-
-What was decided.
-
-## Consequences
-
-- Bullet list of implications for future contributors.
-~~~
-
-## Numbering
-
-Files are named `NNNN-slug.md`, zero-padded to 4 digits, per directory.
-Numbers are assigned in the order decisions were recorded, not by importance.
-
-## Amending records
-
-- Never delete an ADR.
-- If a decision is superseded, update the original status to `Superseded by ADR-NNNN` and create a new ADR.
-```
-
-**1.2** Create `apps/claude-code/auto-format/docs/adr/README.md`:
-
-```markdown
-# ADRs — auto-format plugin
-
-Plugin-scoped architectural decisions. Repo-wide decisions live in `docs/adr/` at the monorepo root.
-See the root `docs/adr/README.md` for format and numbering conventions.
-```
-
-**1.3** Create `apps/claude-code/pr-review/docs/adr/README.md`:
-
-```markdown
-# ADRs — pr-review plugin
-
-Plugin-scoped architectural decisions. Repo-wide decisions live in `docs/adr/` at the monorepo root.
-See the root `docs/adr/README.md` for format and numbering conventions.
-```
-
-**1.4** Create `apps/claude-code/unic-confluence/docs/adr/README.md`:
-
-```markdown
-# ADRs — unic-confluence plugin
-
-Plugin-scoped architectural decisions. Repo-wide decisions live in `docs/adr/` at the monorepo root.
-See the root `docs/adr/README.md` for format and numbering conventions.
-```
-
----
-
-### Phase 2 — Repo-wide ADRs
-
-**2.1** Create `docs/adr/0001-pnpm-workspaces-no-turborepo.md`:
-
-```markdown
-# 0001. pnpm 10 workspaces — no Turborepo or Nx
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-A new monorepo needed a package manager strategy. Options included Turborepo, Nx, and plain pnpm workspaces. The repo contains a small number of plugins and shared packages; build orchestration overhead was undesirable.
-
-## Decision
-
-Use pnpm 10 in workspace mode. `pnpm --filter` is the sole dispatch mechanism. A workspace `catalog:` handles shared version pinning. No Turborepo, Nx, or other build orchestrator.
-
-## Consequences
-
-- Adding a new package requires a new directory + `package.json`; no pipeline registration step.
-- CI fan-out is handled by `dorny/paths-filter` rather than task-graph inference.
-- Turborepo caching and remote-cache features are not available.
-- Contributors need pnpm 10; `package.json#packageManager` enforces this.
-```
-
-**2.2** Create `docs/adr/0002-shared-unic-packages.md`:
-
-```markdown
-# 0002. Shared tooling extracted into @unic/* workspace packages
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Multiple plugins share the same lint rules, TypeScript config, and release scripts. Copying these per plugin leads to drift.
-
-## Decision
-
-Extract shared tooling into three `packages/` workspace members:
-
-- `@unic/biome-config` — Biome 2 rules and formatter config
-- `@unic/tsconfig` — base `tsconfig.json` for `tsc --checkJs`
-- `@unic/release-tools` — `unic-bump`, `unic-sync-version`, `unic-tag`, `unic-verify-changelog` bin commands
-
-Plugins depend on these via `workspace:*`. The packages are private (not published).
-
-## Consequences
-
-- Lint and type-checking rules evolve in one place.
-- Adding a new plugin means consuming `workspace:*` deps, not copying scripts.
-- Changes to shared tooling must not break any plugin; cross-plugin test runs are required.
-```
-
-**2.3** Create `docs/adr/0003-jsdoc-tsc-nocheck-no-compile.md`:
-
-```markdown
-# 0003. JSDoc + tsc --checkJs --noEmit — no TypeScript compilation
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Plugin scripts must be runnable by Node directly (no build step). TypeScript provides type safety but the compilation output would be redundant for `.mjs` source.
-
-## Decision
-
-Use `// @ts-check` + JSDoc annotations in `.mjs` files. Run `tsc --checkJs --noEmit` for type-checking. No `tsc` compilation to JS; source files are the runtime artefacts.
-
-## Consequences
-
-- Zero-build deployments: Claude Code installs and runs source directly.
-- Type errors surface at dev time via `pnpm typecheck`, not at deploy time.
-- Some TypeScript features (decorators, enums, `as` casts in body) are unavailable.
-- Contributors unfamiliar with JSDoc types need a brief onboarding.
-```
-
-**2.4** Create `docs/adr/0004-biome-prettier-split.md`:
-
-```markdown
-# 0004. Biome 2 for code/JSON; Prettier reserved for Markdown
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Biome 2 is fast and unifies linting + formatting for JS/TS/JSON. However, Biome does not format Markdown well. Prettier handles Markdown natively.
-
-## Decision
-
-- Biome 2: lint and format `.mjs`, `.js`, `.ts`, `.json`, `.jsonc`
-- Prettier: format `.md` only
-
-Both tools run in `pnpm check` / `pnpm format` at the workspace root.
-
-## Consequences
-
-- Contributors need both `biome` and `prettier` dev dependencies.
-- Markdown formatting is Prettier-governed (2-space indent, 80-col prose wrap by default).
-- Any new file type must be explicitly assigned to one tool to avoid gaps.
-```
-
-**2.5** Create `docs/adr/0005-indentation-policy.md`:
-
-```markdown
-# 0005. Indentation policy: tabs for code, 2-space for data files
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Mixed indentation across contributors and editors causes noisy diffs. A single explicit policy prevents this.
-
-## Decision
-
-- `.mjs`, `.js`, `.ts`: **tabs**
-- `.json`, `.jsonc`, `.yml`, `.yaml`: **2 spaces**
-- `.md`: Prettier default (2 spaces in lists, no tab indent)
-
-Enforced by `@unic/biome-config` and Prettier config at the repo root.
-
-## Consequences
-
-- `editorconfig` at the root mirrors these rules for editor support.
-- Biome and Prettier enforce them in CI; non-conforming files fail `pnpm check`.
-```
-
-**2.6** Create `docs/adr/0006-cross-platform-node-only-scripts.md`:
-
-```markdown
-# 0006. Cross-platform scripting via Node.js APIs — no shell commands
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-CI runs on macOS, Ubuntu, and Windows. Shell-based scripts (`#!/bin/bash`, `cp`, POSIX paths) fail on Windows or require WSL.
-
-## Decision
-
-All release scripts, hooks, and tooling use Node.js built-in APIs (`node:fs`, `node:path`, `node:child_process`, `node:os`) exclusively. No `bash`, `sh`, `cp`, `mv`, or POSIX path separators in scripts.
-
-## Consequences
-
-- Scripts run identically on all three OSes without a shell layer.
-- `node:path.join` + `path.sep` replace POSIX path literals.
-- `child_process.spawnSync` replaces shell pipelines; arguments are passed as arrays.
-- Any future contributor adding a shell-based script must propose an ADR amendment.
-```
-
-**2.7** Create `docs/adr/0007-release-tools-bin-commands.md`:
-
-```markdown
-# 0007. Release scripts exposed as bin commands; cwd-relative, no --package flag
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Release scripts (`bump`, `sync-version`, `tag`, `verify-changelog`) must work identically for any plugin. Two conventions were considered: (a) accept a `--package <name>` flag, (b) read `process.cwd()` and be invoked via `pnpm --filter`.
-
-## Decision
-
-Scripts are registered as `bin` entries in `@unic/release-tools/package.json` (`unic-bump`, `unic-sync-version`, `unic-tag`, `unic-verify-changelog`). They read `process.cwd()` as the package root. Consumers invoke them via `pnpm --filter <name> bump patch`, which sets cwd automatically.
-
-## Consequences
-
-- Plugin `package.json#scripts` entries are short (`"bump": "unic-bump"`).
-- Scripts are reusable across any pnpm-filtered package without extra flags.
-- Scripts will break if invoked outside of a `pnpm --filter` context (they'll read the wrong cwd).
-```
-
-**2.8** Create `docs/adr/0008-tag-scheme-plugin-at-version.md`:
-
-```markdown
-# 0008. Tag scheme: <plugin-name>@<version>
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-A single monorepo releases N plugins independently. A repo-wide `vX.Y.Z` tag cannot distinguish which plugin was released. Namespaced tags (e.g. `auto-format@0.5.5`) allow per-plugin release detection.
-
-## Decision
-
-Every release tag follows the pattern `<plugin-name>@<version>` (e.g. `auto-format@0.5.5`, `pr-review@1.2.0`). The plugin name matches the directory name under `apps/claude-code/`.
-
-## Consequences
-
-- `unic-tag` must construct the tag name from the plugin directory name + `plugin.json` version.
-- The release workflow detects new releases by checking tag existence, not by diffing `HEAD~1`.
-- Standard tools expecting `vX.Y.Z` tags (GitHub Releases auto-title, semantic-release) need configuration.
-```
-
-**2.9** Create `docs/adr/0009-plugin-json-version-source-of-truth.md`:
-
-```markdown
-# 0009. plugin.json is the single source of truth for plugin version
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Each plugin has three files that carry a version: `.claude-plugin/plugin.json`, `marketplace.json`, and `package.json`. Manual edits to all three cause drift.
-
-## Decision
-
-`.claude-plugin/plugin.json` is the authoritative version field. `unic-sync-version` propagates the value to `marketplace.json` and `package.json` after every bump. Humans must only edit `plugin.json`.
-
-## Consequences
-
-- `pnpm --filter <name> bump <type>` is the only sanctioned version-change path.
-- `marketplace.json` must never be hand-edited (enforced by convention; no file lock exists).
-- The sync script must be run before tagging; `unic-tag` calls it defensively.
-```
-
-**2.10** Create `docs/adr/0010-pnpm-filter-bump-only-version-path.md`:
-
-```markdown
-# 0010. pnpm --filter <name> bump <type> is the only sanctioned version-change path
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Version bumps must be atomic: increment the version, update CHANGELOG, sync derived files, and validate the result. Doing these steps ad hoc leads to inconsistency.
-
-## Decision
-
-`pnpm --filter <name> bump <patch|minor|major>` is the only way to change a plugin version. The `unic-bump` script: (1) validates a clean working tree, (2) increments `plugin.json`, (3) runs `unic-sync-version`. Contributors must not hand-edit version fields.
-
-## Consequences
-
-- CI's `verify:changelog` gate catches version bumps that were not accompanied by a changelog entry.
-- `unic-bump` will exit non-zero on a dirty working tree, preventing mixed commits.
-```
-
-**2.11** Create `docs/adr/0011-plugin-migration-via-git-filter-repo.md`:
-
-```markdown
-# 0011. Plugin migration into the monorepo uses git filter-repo (subdirectory rewrite)
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Three plugins were previously maintained in separate repositories. Moving them to the monorepo while preserving full commit history required a strategy. Options: (a) copy files + squash history, (b) `git filter-repo` to rewrite paths + merge with `--allow-unrelated-histories`.
-
-## Decision
-
-Use `git filter-repo --to-subdirectory-filter apps/claude-code/<plugin>` on the source repo, then merge into the monorepo with `git merge --allow-unrelated-histories`. This preserves every commit from the original repo, rewritten under the monorepo subdirectory path.
-
-## Consequences
-
-- Full per-plugin commit history is visible inside the monorepo via `git log -- apps/claude-code/<plugin>`.
-- `git blame` works at the per-plugin level.
-- The migration is one-time; future work happens entirely in the monorepo.
-- `git filter-repo` must be installed locally (not part of standard git).
-```
-
-**2.12** Create `docs/adr/0012-license-files-manually-managed.md`:
-
-```markdown
-# 0012. LICENSE files are maintained manually by the repo owner
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-LICENSE files are legal artefacts. Automated tooling (Ralph, Claude) creating, copying, or deleting them could accidentally relicense packages or create inconsistencies that require legal review.
-
-## Decision
-
-LICENSE files in every package and plugin directory are managed manually by the maintainer. Ralph specs and Claude must never create, copy, or delete LICENSE files. If a spec step requires a LICENSE file to exist, the spec must include a note directing the maintainer to add it manually.
-
-## Consequences
-
-- `AGENTS.md` explicitly prohibits automated LICENSE file changes.
-- Reviewers should reject PRs that add or remove LICENSE files via automation.
-```
-
-**2.13** Create `docs/adr/0013-single-root-marketplace.md`:
-
-```markdown
-# 0013. Single root marketplace.json listing all plugins
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Claude Code discovers plugins via a marketplace manifest. With N plugins in one repo, users could either: (a) add N separate plugin sources, or (b) add one monorepo-level source that lists all plugins.
-
-## Decision
-
-Maintain a single `marketplace.json` at the repo root that lists every plugin. `unic-sync-version` updates the relevant plugin entry automatically. Users add one Claude Code source URL.
-
-## Consequences
-
-- Adding a new plugin requires adding its entry to the root `marketplace.json` template.
-- All plugins are installed as a bundle when a user adds the source; individual opt-out is not supported.
-- `marketplace.json` must never be hand-edited (it is derived from plugin manifests).
-```
-
-**2.14** Create `docs/adr/0014-ci-matrix-three-os-two-node.md`:
-
-```markdown
-# 0014. CI matrix: Ubuntu × macOS × Windows, Node 22 × Node 24
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Plugins must run on all major operating systems (AGENTS.md cross-platform requirement) and must not break when users upgrade Node. Testing on the matrix asserts the cross-platform contract in CI rather than relying on convention.
-
-## Decision
-
-The per-package test job runs a 3×2 matrix: `ubuntu-latest`, `macos-latest`, `windows-latest` × Node 22 LTS, Node 24 LTS.
-
-Windows installs run with `--ignore-scripts` because Windows postinstall scripts are unreliable in CI.
-
-## Consequences
-
-- CI cost and duration scale with 6 environments per changed package.
-- Any script that uses OS-specific APIs will fail on at least one runner.
-- Node version support policy: current LTS + next LTS.
-```
-
-**2.15** Create `docs/adr/0015-ci-paths-filter-changed-packages.md`:
-
-```markdown
-# 0015. CI uses paths-filter to scope test jobs to changed packages
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Running all package tests on every push is expensive as the repo grows. Most pushes touch one plugin or one shared package.
-
-## Decision
-
-Use `dorny/paths-filter` in the CI workflow to detect which packages changed. Only changed packages fan out into the OS × Node matrix. The root checks job (Biome, Prettier) is unconditional and always runs.
-
-## Consequences
-
-- A push touching only `apps/claude-code/auto-format/` does not run `pr-review` or `unic-confluence` tests.
-- A push touching `packages/release-tools/` should ideally trigger all plugins, but the paths filter must be kept up to date manually as new plugins are added.
-- Documentation-only changes may skip the matrix entirely; `pnpm format` still runs.
-```
-
-**2.16** Create `docs/adr/0016-root-checks-unconditional.md`:
-
-```markdown
-# 0016. Root checks (Biome + Prettier) always run, regardless of changed files
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Formatting is a repo-wide hygiene concern. A paths-filter on root checks would allow unformatted files to land if the change set appeared documentation-only.
-
-## Decision
-
-The `root-checks` CI job (Biome lint+format, Prettier on Markdown) runs on every push and every PR, unconditionally. It runs once on `ubuntu-latest` / Node 24 only (formatting is OS-independent).
-
-## Consequences
-
-- Every push pays the cost of a root check (~10–20 s).
-- Contributors cannot bypass the formatter by touching only `docs/` files.
-```
-
-**2.17** Create `docs/adr/0017-verify-changelog-pr-only.md`:
-
-```markdown
-# 0017. verify:changelog runs on PR events only, not on push to main
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-`unic-verify-changelog` enforces that every version bump is accompanied by a dated CHANGELOG entry. Running it on push to `main` would block merges that were already validated on the originating PR.
-
-## Decision
-
-`verify:changelog` is a PR-only gate. It does not run on push to `develop` or `main`. The enforcement happens at review time, not at merge time.
-
-## Consequences
-
-- Direct pushes to `develop` (which are allowed by Gitflow for some maintainers) bypass the changelog gate.
-- The gate relies on PR discipline; it cannot catch hotfixes pushed directly.
-```
-
-**2.18** Create `docs/adr/0018-release-on-merge-tag-existence.md`:
-
-```markdown
-# 0018. Releases triggered automatically on main by tag-existence check
-
-**Status:** Accepted (2025-05) — supersedes HEAD~1 diff approach (spec 14)
-
-## Context
-
-The release workflow must create a `<plugin>@<version>` tag when a new version lands on `main`. Two approaches were considered: (a) diff `plugin.json` against `HEAD~1`, (b) check whether the tag `<plugin>@<version>` already exists.
-
-The HEAD~1 diff approach fails when multiple commits are pushed together (the version bump may not be in the topmost commit). Tag-existence is idempotent and survives re-runs.
-
-## Decision
-
-For each plugin, the release workflow reads the current version from `plugin.json` and checks whether `<plugin>@<version>` already exists as a git tag. If the tag is absent, it creates it (with optional GPG signing). If the tag exists, it skips silently.
-
-## Consequences
-
-- Re-running the release workflow on the same commit is safe (idempotent).
-- Downgrading a plugin version would not create a tag, which is the correct behaviour.
-- The workflow must iterate over all plugins on every push to `main`.
-```
-
-**2.19** Create `docs/adr/0019-gpg-signing-opt-in.md`:
-
-```markdown
-# 0019. GPG tag signing is opt-in via UNIC_SIGN_TAGS repository secret
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Signed git tags provide provenance guarantees but require GPG keys to be set up in the CI environment. Mandatory signing would block contributors who have not configured GPG.
-
-## Decision
-
-`unic-tag` and the release workflow sign tags only when the `UNIC_SIGN_TAGS` environment variable / repository secret is set to a truthy value. When absent, tags are created unsigned.
-
-## Consequences
-
-- Local `pnpm --filter <name> tag` runs without GPG by default.
-- CI can be configured to sign by adding the `UNIC_SIGN_TAGS` secret to the GitHub repository.
-- Verifying tag signatures requires the public key to be distributed separately.
-```
-
-**2.20** Create `docs/adr/0020-per-plugin-ralph-loops.md`:
-
-```markdown
-# 0020. Each plugin has its own Ralph loop with its own ralph.yml and PROMPT.md
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Ralph Orchestrator implements specs one at a time in a loop. A single monorepo-wide loop would mix plugin-specific specs with workspace specs, making iteration focus unclear.
-
-## Decision
-
-Two levels of Ralph loops exist:
-
-1. **Monorepo root loop** (`pnpm ralph` at repo root): implements `docs/plans/` specs (workspace structure, shared packages, CI).
-2. **Per-plugin loop** (`pnpm --filter <name> ralph`): implements `apps/claude-code/<plugin>/docs/plans/` specs.
-
-Each level has its own `ralph.yml` and `PROMPT.md`.
-
-## Consequences
-
-- Adding a new plugin requires creating its own `ralph.yml` and `PROMPT.md`.
-- Running the root loop does not touch plugin-specific specs and vice versa.
-- Status markers (`**Status: done**`) are scoped per loop level.
-```
-
-**2.21** Create `docs/adr/0021-conventional-commits-package-scope.md`:
-
-```markdown
-# 0021. Conventional Commits with package or spec scope
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-With multiple plugins and shared packages in one repo, commit messages without scope make changelogs and `git log` noisy. Conventional Commits with a scope field allow per-package filtering.
-
-## Decision
-
-Commit messages follow Conventional Commits with a mandatory scope:
-
-- Plugin work: `feat(auto-format): …`, `fix(pr-review): …`, `chore(unic-confluence): …`
-- Spec progress: `chore(spec-NN): …`
-- Shared packages: `chore(release-tools): …`, `chore(biome-config): …`
-- Workspace-level: `chore(workspace): …`
-
-## Consequences
-
-- `git log --grep="(auto-format)"` filters to one plugin's history.
-- Release notes can be generated per-plugin by filtering on the scope.
-- Reviewers should reject commits without a scope.
-```
-
-**2.22** Create `docs/adr/0022-semver-per-plugin.md`:
-
-```markdown
-# 0022. SemVer per plugin, not per monorepo
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Plugins are independently released and consumed. A monorepo-level version would conflate unrelated changes across plugins.
-
-## Decision
-
-Each plugin maintains its own SemVer version in `.claude-plugin/plugin.json`. The versioning contract:
-
-- **major**: breaking change to the plugin's CLI interface, exit codes, or file schema
-- **minor**: new feature that is backwards-compatible
-- **patch**: bug fix, documentation update, or internal refactor with no behaviour change
-
-Shared packages (`@unic/*`) are internal and unversioned for external consumers.
-
-## Consequences
-
-- A breaking change in `unic-confluence` does not affect `auto-format` or `pr-review` versioning.
-- Changelogs are per-plugin (`CHANGELOG.md` inside each plugin directory).
-- There is no monorepo-wide version or combined release notes.
-```
-
-**2.23** Create `docs/adr/0023-spec-template-format.md`:
-
-```markdown
-# 0023. Spec template format for Ralph-executable specs
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Ralph Orchestrator needs specs to be unambiguous enough to execute deterministically. An ad hoc format leads to missing context, Ralph pausing for clarification, or incorrect implementations.
-
-## Decision
-
-All specs follow `docs/process/spec-template.md`, which mandates:
-
-- A header block: `Priority`, `Effort`, `Version impact`, `Depends on`, `Touches`
-- `## Current behaviour` with exact before-state snapshots (code or CLI output)
-- `## Target behaviour` with exact after-state
-- `## Implementation steps` with explicit before→after diffs or full file content
-- `## Verification` with shell commands and their expected output
-- `## Acceptance criteria` as checkboxes
-- `## Out of scope` bounding the change
-
-## Consequences
-
-- Specs are longer to write but faster for Ralph to execute without back-and-forth.
-- The "Out of scope" section prevents scope creep during automated implementation.
-- Specs that deviate from the template may cause Ralph to pause or implement incorrectly.
-```
-
-**2.24** Create `docs/adr/0024-ralph-atomic-iteration.md`:
-
-```markdown
-# 0024. Ralph implements one spec per iteration, then commits and stops
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Batching multiple specs into one Ralph iteration makes commits hard to review, complicates revert, and makes it harder to resume after a failure.
-
-## Decision
-
-Ralph's loop implements exactly one spec per run, creates one commit, and stops. Loop resumption reads `**Status: done**` markers in spec files to determine the next spec to implement. The loop never modifies `PROMPT.md` or `ralph.yml` to batch specs.
-
-## Consequences
-
-- Each PR from a Ralph run contains one logical unit of work.
-- The commit message references the spec number and title.
-- Resuming after a failed Ralph run is mechanical: remove the `**Status: done**` marker from the partially-implemented spec and restart.
-```
-
----
-
-### Phase 3 — auto-format plugin ADRs
-
-**3.1** Create `apps/claude-code/auto-format/docs/adr/0001-hook-always-exits-zero.md`:
-
-```markdown
-# 0001. Hook script always exits 0
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-The auto-format hook runs as a Claude Code `PostToolUse` hook after every file edit. If the hook exits non-zero, Claude Code treats the tool call as failed and may retry or abort the session.
-
-## Decision
-
-The hook script always exits 0, even when a formatter fails. Formatter errors are reported to stderr (visible in Claude Code's hook output) but do not block the tool flow.
-
-## Consequences
-
-- Formatter failures are surfaced as warnings, not as tool failures.
-- Claude Code sessions are never blocked by a misconfigured formatter.
-- Silent failures are possible if the user does not monitor hook output; this is an accepted trade-off.
-```
-
-**3.2** Create `apps/claude-code/auto-format/docs/adr/0002-zero-runtime-dependencies.md`:
-
-```markdown
-# 0002. Zero runtime dependencies in the hook script
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-The hook script is installed into users' Claude Code environments. Every runtime dependency adds install time, version pinning concerns, and potential breakage. The repo guideline is "zero runtime deps unless truly essential."
-
-## Decision
-
-The hook script uses only Node.js built-in modules (`node:child_process`, `node:fs`, `node:path`, `node:os`). No third-party packages are listed in `dependencies`.
-
-## Consequences
-
-- Install is instant (`npm install` / `pnpm install` fetches no extra packages).
-- The hook cannot use convenience libraries (e.g. `glob`, `chalk`); all logic is hand-rolled.
-- devDependencies (Biome, TypeScript) are allowed and are not installed in production.
-```
-
-**3.3** Create `apps/claude-code/auto-format/docs/adr/0003-consumer-owns-formatters.md`:
-
-```markdown
-# 0003. The plugin orchestrates; the consumer project owns its formatters
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Bundling Prettier, ESLint, or Biome inside the plugin would lock consumers to specific versions and prevent them from customising formatter config. Detecting and invoking formatters already installed in the consumer project avoids this.
-
-## Decision
-
-The hook script detects which formatters are available in the consumer project (`prettier`, `eslint`, `biome`) by checking `node_modules/.bin` and `package.json` scripts. It invokes whatever it finds. The plugin never bundles or installs formatters.
-
-## Consequences
-
-- Consumers must have their formatters installed; the hook silently skips missing tools.
-- Formatter version and config are always the consumer's own.
-- The plugin works with any combination of Prettier, ESLint, and Biome without code changes.
-```
-
-**3.4** Create `apps/claude-code/auto-format/docs/adr/0004-per-project-config-merge.md`:
-
-```markdown
-# 0004. Per-project config merges over plugin defaults
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Some projects need to disable specific formatters or file extensions. Without an override mechanism, users would have to fork the plugin.
-
-## Decision
-
-The plugin reads an optional `.claude/unic-format.json` in the consumer project root. Settings in this file merge over the plugin defaults, allowing users to opt out of specific tools or add extension sets without modifying the plugin.
-
-## Consequences
-
-- The plugin behaviour is customisable per project without a fork.
-- The config schema must be documented and versioned.
-- The hook must validate the consumer config and fail gracefully on schema errors.
-```
-
-**3.5** Create `apps/claude-code/auto-format/docs/adr/0005-spawnsync-timeout-kill-signal.md`:
-
-```markdown
-# 0005. spawnSync calls include a timeout and killSignal on every formatter invocation
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-A hung formatter (e.g. Prettier on a circular require, ESLint on a very large file) would freeze Claude Code's edit loop indefinitely. A timeout guard is necessary.
-
-## Decision
-
-Every `spawnSync` call in the hook script includes `timeout` (configurable, defaulting to 10 000 ms) and `killSignal: 'SIGTERM'`. If the timeout expires, the process is killed and the hook logs a warning to stderr and exits 0.
-
-## Consequences
-
-- Formatters that take longer than the timeout are killed; the file is left unformatted.
-- The timeout is configurable via per-project config (see ADR-0004).
-- SIGTERM may not stop all processes on Windows; `killSignal` is best-effort there.
-```
-
-**3.6** Create `apps/claude-code/auto-format/docs/adr/0006-posix-path-normalization-windows.md`:
-
-```markdown
-# 0006. Normalize Windows paths to POSIX before extension/glob matching
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Claude Code passes file paths to hooks using the OS path separator. On Windows, paths use backslashes (`\`). Extension matching and glob patterns in the hook script assume POSIX separators.
-
-## Decision
-
-The hook script normalises all incoming file paths to POSIX format (`/`) using `path.normalize` + `.replace(/\\/g, '/')` before any extension or glob matching. Normalization is applied once at entry.
-
-## Consequences
-
-- Extension sets and glob patterns are written as POSIX strings throughout the script.
-- The normalization step adds negligible overhead.
-- Paths passed to `spawnSync` are re-normalized to the OS separator for the subprocess.
-```
-
-**3.7** Create `apps/claude-code/auto-format/docs/adr/0007-ci-creates-tags-automatically.md`:
-
-```markdown
-# 0007. Release tags are created automatically by CI, not manually
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Manual tagging (`pnpm --filter auto-format tag && git push --follow-tags`) is error-prone and easy to forget. The tag is the trigger for the release workflow.
-
-## Decision
-
-The monorepo-level release workflow (`.github/workflows/release.yml`) creates `auto-format@<version>` automatically when `plugin.json` contains a version that has no corresponding git tag. Contributors never run `unic-tag` manually.
-
-## Consequences
-
-- Every version bump merged to `main` is automatically released.
-- Local `unic-tag` is available but not part of the normal release flow.
-- Reverting a version bump after the tag is created requires a manual `git tag -d` + push.
-```
-
-**3.8** Create `apps/claude-code/auto-format/docs/adr/0008-jsdoc-types-ts-check.md`:
-
-```markdown
-# 0008. JSDoc types and tsc --checkJs for type safety without compilation
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Consistent with the monorepo-wide decision (repo ADR-0003). Recorded here for plugin-local reference.
-
-## Decision
-
-Hook script files use `// @ts-check` and JSDoc annotations. Type checking runs via `pnpm typecheck` (`tsc --checkJs --noEmit`). No TypeScript compilation occurs.
-
-## Consequences
-
-- Same as repo-level ADR-0003.
-- Plugin contributors need to write JSDoc for new function parameters and return types.
-```
-
----
-
-### Phase 4 — pr-review plugin ADRs
-
-**4.1** Create `apps/claude-code/pr-review/docs/adr/0001-canonical-bot-signature.md`:
-
-```markdown
-# 0001. Canonical bot signature is the session identity contract
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-The plugin posts review comments on pull requests. Subsequent runs must distinguish their own prior comments from human comments to avoid duplication, enable incremental reviews, and rewrite the summary comment rather than appending.
-
-## Decision
-
-Every comment posted by the plugin ends with a fixed signature line (e.g. `Reviewed by Claude Code`). This exact string is the identity contract: future runs detect prior reviews and comments by substring-matching on this signature.
-
-## Consequences
-
-- The signature string must never change between plugin versions (breaking change).
-- Human reviewers who accidentally include the signature string in their comments will be misidentified as bot comments.
-- The signature is the only identity mechanism; there is no API-level "bot author" field.
-```
-
-**4.2** Create `apps/claude-code/pr-review/docs/adr/0002-signature-based-prior-review-detection.md`:
-
-```markdown
-# 0002. Prior review detected by signature substring match, not by API metadata
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Azure DevOps does not expose a first-class "bot author" field on comments. Alternatives to signature matching (author ID, comment type) are either unavailable or unreliable across ADO versions.
-
-## Decision
-
-The plugin detects a prior review by fetching all PR comments and checking whether any contains the canonical bot signature (see ADR-0001). If found, the run is classified as a re-review.
-
-## Consequences
-
-- Detection is O(N) in the number of PR comments.
-- Works across all ADO versions without API-level bot identity.
-- A human who includes the signature string in a comment will trigger a false re-review detection.
-```
-
-**4.3** Create `apps/claude-code/pr-review/docs/adr/0003-target-latest-pr-iteration.md`:
-
-```markdown
-# 0003. Always target the latest PR iteration, not iterationId=1
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Azure DevOps tracks PR changes as numbered iterations. Inline comments must anchor to lines that exist in the current diff. Comments anchored to `iterationId=1` reference lines from the original push and may become invalid after force-push or rebase.
-
-## Decision
-
-The plugin always fetches the latest iteration ID for the PR and uses it when posting inline comments. `iterationId=1` is never used.
-
-## Consequences
-
-- Comments posted during a re-review anchor to the latest iteration's diff.
-- If a re-review runs between two push events, the iteration ID may change mid-review; comments from the first half of the review may reference a stale iteration. This is an accepted edge case.
-```
-
-**4.4** Create `apps/claude-code/pr-review/docs/adr/0004-incremental-diff-baseline.md`:
-
-```markdown
-# 0004. Incremental diff baseline = prior iteration's sourceRefCommit
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Re-reviewing a PR that has already been reviewed should focus on what changed since the last review, not re-analyse the entire diff. The baseline for "what changed" must be the commit that was HEAD when the previous review was posted.
-
-## Decision
-
-When a prior review is detected, the baseline commit is read from the prior review iteration's `sourceRefCommit.commitId`. Only files changed between that commit and the current HEAD are passed to the review agents.
-
-## Consequences
-
-- Re-review cost and noise are proportional to the delta since the last review.
-- If the PR was rebased and the baseline commit is no longer in the history, the plugin falls back to a full diff.
-```
-
-**4.5** Create `apps/claude-code/pr-review/docs/adr/0005-four-state-thread-classification.md`:
-
-```markdown
-# 0005. Existing threads classified into four states: addressed, disputed, pending, obsolete
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-A re-review must handle existing comment threads intelligently: resolving issues that were fixed, escalating issues that were disputed, and carrying forward issues still outstanding. A shared taxonomy is needed for the reply and summary phases to make consistent decisions.
-
-## Decision
-
-Each existing bot thread is classified into one of four states:
-
-- **addressed**: the issue was fixed in the new diff
-- **disputed**: the reviewer replied disagreeing with the bot comment
-- **pending**: no action taken; issue still exists in the new diff
-- **obsolete**: the relevant code was deleted or moved; the comment no longer applies
-
-## Consequences
-
-- The classification taxonomy must remain stable across plugin versions.
-- Adding a fifth state is a minor version bump (new feature, backwards-compatible).
-- Classification accuracy depends on the review agent's analysis; false classifications are possible.
-```
-
-**4.6** Create `apps/claude-code/pr-review/docs/adr/0006-reply-not-duplicate-auto-resolve.md`:
-
-```markdown
-# 0006. Reply to existing threads instead of opening duplicates; auto-resolve addressed threads
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Re-reviews that open duplicate comments for already-noted issues create noise and make PR conversations hard to follow. Addressed issues should be resolved to signal progress.
-
-## Decision
-
-- For **pending** and **disputed** threads: post a reply noting whether the issue persists or has been escalated.
-- For **addressed** threads: post a reply confirming the fix and resolve the thread.
-- Never open a new thread for an issue that already has an active thread.
-
-## Consequences
-
-- PR comment threads remain linear and readable.
-- Addressed threads are automatically resolved, reducing the reviewer's manual work.
-- Incorrectly classified threads (e.g. false "addressed") will be auto-resolved; the reviewer may need to reopen them.
-```
-
-**4.7** Create `apps/claude-code/pr-review/docs/adr/0007-summary-rewritten-not-appended.md`:
-
-```markdown
-# 0007. Summary comment is rewritten on re-review, not appended
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-A PR that goes through multiple review cycles accumulates stale summary comments if each run appends a new one. A single living summary is easier to read and tracks the current state of the review.
-
-## Decision
-
-The plugin maintains exactly one summary comment (identified by the bot signature). On re-review, the existing summary comment is edited (rewritten) with the updated issue count, severity breakdown, and outstanding items. No new summary comment is posted.
-
-## Consequences
-
-- The PR always shows one summary comment, not a stack of historical ones.
-- The GitHub/ADO comment edit timestamp shows when the summary was last updated.
-- If the summary comment is deleted by a human, the next re-review creates a new one.
-```
-
-**4.8** Create `apps/claude-code/pr-review/docs/adr/0008-soft-dependency-pr-review-toolkit.md`:
-
-```markdown
-# 0008. Soft dependency on pr-review-toolkit — abort with install instructions, never vendor
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-The pr-review plugin uses specialised sub-agents from `pr-review-toolkit`. Bundling the toolkit inside pr-review would couple their release cycles and inflate the plugin size.
-
-## Decision
-
-`pr-review` declares a soft dependency on `pr-review-toolkit`. On startup, the plugin checks whether the toolkit is installed. If not, it aborts and prints installation instructions. It never vendors or copies toolkit code.
-
-## Consequences
-
-- Users must install `pr-review-toolkit` separately.
-- Toolkit upgrades (new agent types, improved prompts) are available to pr-review without a pr-review release.
-- The startup check adds a small latency on every invocation.
-```
-
----
-
-### Phase 5 — unic-confluence plugin ADRs
-
-**5.1** Create `apps/claude-code/unic-confluence/docs/adr/0001-refuse-publish-without-markers.md`:
-
-```markdown
-# 0001. Refuse to publish a page that has no markers; require --replace-all to overwrite
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Publishing to Confluence without placement markers would silently overwrite hand-written page content. This is a destructive operation with no undo in the standard Confluence UI history for non-admins.
-
-## Decision
-
-The publish script refuses to process a Markdown file that contains no placement markers (e.g. `[AUTO_INSERT_START: label]` / `[AUTO_INSERT_END: label]`). To overwrite an entire page unconditionally, the caller must pass `--replace-all` explicitly.
-
-## Consequences
-
-- Authors must add markers to any Markdown file they intend to publish.
-- The `--replace-all` flag is a deliberate escape hatch; it should not be used in automated pipelines.
-- The guard prevents accidental destruction of Confluence page content.
-```
-
-**5.2** Create `apps/claude-code/unic-confluence/docs/adr/0002-three-strategy-injection-priority.md`:
-
-```markdown
-# 0002. Three-strategy injection with explicit priority: markers → anchor macros → append
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Authors need different levels of precision when placing Markdown content into a Confluence page. A single strategy cannot satisfy all cases.
-
-## Decision
-
-The injection engine tries three strategies in priority order:
-
-1. **Plain markers** (`[AUTO_INSERT_START: label]` / `[AUTO_INSERT_END: label]`): highest precision; replaces exactly the bounded section.
-2. **Anchor macros**: places content relative to a named Confluence anchor macro.
-3. **Append**: appends to the end of the page as a last resort.
-
-The first matching strategy wins.
-
-## Consequences
-
-- Authors choose precision vs. convenience by adding the appropriate markers.
-- The engine is deterministic: the same Markdown + same page always produces the same result.
-- Adding a fourth strategy is a minor version bump.
-```
-
-**5.3** Create `apps/claude-code/unic-confluence/docs/adr/0003-structured-macro-for-code-blocks.md`:
-
-```markdown
-# 0003. ac:structured-macro for code blocks instead of <pre>
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Confluence Storage Format supports both `<pre>` (plain preformatted text) and `ac:structured-macro name="code"` (the native Confluence Code macro). The Code macro provides syntax highlighting, a copy button, and a collapse option; `<pre>` provides none of these and loses formatting on Confluence editor reload.
-
-## Decision
-
-All fenced code blocks in Markdown are converted to `ac:structured-macro name="code"` elements in the Confluence Storage Format output. The language attribute is mapped to the macro's `language` parameter.
-
-## Consequences
-
-- Published code blocks look and behave like natively-authored Confluence code blocks.
-- The conversion must handle unknown language identifiers gracefully (fall back to `text`).
-- `<pre>` is never emitted by the converter.
-```
-
-**5.4** Create `apps/claude-code/unic-confluence/docs/adr/0004-dry-run-read-only.md`:
-
-```markdown
-# 0004. --dry-run is purely read-only: never sends a PUT to Confluence
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-`--dry-run` is used in CI smoke tests and pre-commit checks to validate that a Markdown file would publish correctly without actually modifying Confluence.
-
-## Decision
-
-When `--dry-run` is passed, the script performs all parsing, conversion, and marker injection steps but never calls the Confluence API's PUT endpoint. It logs what it would have done and exits 0.
-
-## Consequences
-
-- `--dry-run` is safe to run in any CI context, including production pipelines.
-- The dry run validates the full conversion pipeline but not the Confluence API connection.
-- A successful dry run does not guarantee a successful publish (API auth, page permissions may differ).
-```
-
-**5.5** Create `apps/claude-code/unic-confluence/docs/adr/0005-ping-check-auth-over-per-page-verify.md`:
-
-```markdown
-# 0005. --ping / --check-auth replaces per-page --verify for auth probing
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-An earlier design included a `--verify` flag that fetched N pages to confirm they were accessible. This was slow (N Confluence API round-trips) and gave a vague failure signal when one of N pages was inaccessible.
-
-## Decision
-
-Replace per-page verification with a single `--ping` / `--check-auth` flag that calls a lightweight Confluence API endpoint (e.g. `GET /rest/api/user/current`). One round-trip confirms auth and connectivity.
-
-## Consequences
-
-- Auth check is O(1) regardless of the number of pages configured.
-- The check does not verify page-level access (read/write permissions per page).
-- CI pipelines should run `--check-auth` on startup rather than `--verify`.
-```
-
-**5.6** Create `apps/claude-code/unic-confluence/docs/adr/0006-hard-http-timeout.md`:
-
-```markdown
-# 0006. Hard HTTP timeout on every Confluence API request
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Corporate proxy configurations sometimes cause HTTP requests to stall indefinitely. A hung publish script blocks the user's terminal and any CI pipeline it runs in.
-
-## Decision
-
-Every HTTP request to the Confluence API is wrapped with a hard timeout (configurable, defaulting to 30 000 ms). If the timeout elapses, the request is aborted and the script exits with an error.
-
-## Consequences
-
-- Publish operations fail fast on network issues rather than hanging.
-- The timeout must be configurable for slow enterprise networks.
-- Node's `fetch` API with `AbortController` / `AbortSignal.timeout` implements this without extra dependencies.
-```
-
-**5.7** Create `apps/claude-code/unic-confluence/docs/adr/0007-clierror-class.md`:
-
-```markdown
-# 0007. CliError class instead of console.error + process.exit
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Scattering `console.error(msg); process.exit(1)` throughout the script makes unit testing impossible (process.exit terminates the test runner) and makes error handling inconsistent.
-
-## Decision
-
-All user-facing fatal errors are thrown as `CliError` instances (a custom `Error` subclass with an optional `exitCode` field). The top-level entry point catches `CliError`, prints its message to stderr, and calls `process.exit` exactly once.
-
-## Consequences
-
-- Pure functions in `lib/` can throw `CliError` without calling `process.exit`.
-- Tests can catch `CliError` and assert on its message and exit code without spawning a subprocess.
-- Non-`CliError` exceptions bubble up to the top-level handler as unexpected errors (exit code 1, stack trace printed).
-```
-
-**5.8** Create `apps/claude-code/unic-confluence/docs/adr/0008-pure-functions-lib-with-tests.md`:
-
-```markdown
-# 0008. Pure functions extracted into lib/ and covered by node:test
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-Markdown → Confluence Storage Format conversion and marker injection are deterministic, input→output functions. Keeping them in the main script file makes them hard to test in isolation.
-
-## Decision
-
-All conversion and injection logic lives in `lib/` as pure functions (no I/O, no side effects). The main script imports from `lib/` and handles only I/O. Tests in `test/` use `node:test` to exercise `lib/` functions directly.
-
-## Consequences
-
-- The most complex logic (conversion, injection) is fully covered by unit tests.
-- I/O paths (file reads, HTTP calls) remain thin and are not unit-tested.
-- `lib/` functions must never import from `node:fs`, `node:http`, etc.
-```
-
-**5.9** Create `apps/claude-code/unic-confluence/docs/adr/0009-bare-integer-page-id-schema.md`:
-
-```markdown
-# 0009. confluence-pages.json values are bare integer page IDs
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-`confluence-pages.json` maps local file paths to Confluence page identifiers. An object value (e.g. `{ "id": 123, "space": "ENG" }`) would allow richer metadata but would complicate every read path and invite multi-space scope creep.
-
-## Decision
-
-Values in `confluence-pages.json` are bare integers (Confluence page IDs). No object schema, no space key, no title field. Human-readable aliases (slugified titles) are written alongside the ID as a comment-like key (see ADR-0011).
-
-## Consequences
-
-- Every read path is `config[path]` — no destructuring, no optional field access.
-- The schema cannot be widened to an object without a major version bump.
-- Multi-space support (different Confluence instances for different pages) is explicitly out of scope.
-```
-
-**5.10** Create `apps/claude-code/unic-confluence/docs/adr/0010-no-catalog-for-runtime-deps.md`:
-
-```markdown
-# 0010. Runtime dependencies must not use the pnpm catalog: protocol
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-pnpm workspace `catalog:` references are resolved at install time from `pnpm-workspace.yaml`. When a plugin is installed via `git+https://` (Claude Code's install method), there is no `pnpm-workspace.yaml` in scope and catalog refs break resolution.
-
-## Decision
-
-`dependencies` in `package.json` must use exact version strings or semver ranges — never `catalog:`. `devDependencies` may use `catalog:` because they are not installed in the user's environment.
-
-## Consequences
-
-- Runtime dependencies are pinned or ranged explicitly in `package.json`.
-- Any PR that adds a `catalog:` entry to `dependencies` must be rejected.
-- devDependencies (Biome, TypeScript) continue to use `catalog:` for consistency across the workspace.
-```
-
-**5.11** Create `apps/claude-code/unic-confluence/docs/adr/0011-alias-auto-population.md`:
-
-```markdown
-# 0011. Aliases auto-populate in confluence-pages.json on first publish by numeric ID
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-`confluence-pages.json` is initially configured with bare page IDs, which are not human-readable. Requiring authors to manually add slugified aliases is tedious.
-
-## Decision
-
-On the first successful publish of a page identified by its numeric ID, the script fetches the page title from Confluence, slugifies it, and writes an alias key back into `confluence-pages.json`. Subsequent publishes can reference the page by either the numeric ID or the alias.
-
-## Consequences
-
-- `confluence-pages.json` becomes progressively more readable as pages are published.
-- The script must have write access to `confluence-pages.json` at publish time.
-- An alias collision (two pages with the same slugified title) must be detected and rejected.
-```
-
-**5.12** Create `apps/claude-code/unic-confluence/docs/adr/0012-do-not-add-scope-guard.md`:
-
-```markdown
-# 0012. Explicit "Do not add" scope guard in CLAUDE.md prevents feature creep
-
-**Status:** Accepted (2025-04)
-
-## Context
-
-The unic-confluence plugin has a narrow, well-defined scope: publish local Markdown to Confluence pages using markers. Several adjacent features were proposed and explicitly rejected during design: image upload, page creation, multi-instance support, MCP integration, watch mode, and recursive publish.
-
-## Decision
-
-A `## Do not add` section in the plugin's `CLAUDE.md` pre-rejects these features with brief reasons. Any contributor or automated agent (Ralph, Claude) encountering one of these features must reject the request and refer to this section rather than implementing it.
-
-## Consequences
-
-- Future contributors have a clear signal that these features are not oversight omissions.
-- The ADR system supplements the CLAUDE.md guard: if a feature is re-proposed with new justification, the ADR can be amended rather than the scope guard silently overridden.
-- The scope guard must be reviewed when the plugin reaches 1.0 — some restrictions may be relaxed.
-```
-
-## Verification
-
-Run from the repo root after implementation:
-
-```sh
-# Confirm total file count: 4 READMEs + 24 + 8 + 8 + 12 = 56 files
-find docs/adr apps/claude-code/auto-format/docs/adr apps/claude-code/pr-review/docs/adr apps/claude-code/unic-confluence/docs/adr -name '*.md' | wc -l
-# Expected: 56
-
-# Confirm all ADR files have the required sections
-grep -rL '## Context' docs/adr apps/claude-code/*/docs/adr
-# Expected: only README.md files (no content ADRs should be missing ## Context)
-
-# Confirm Prettier passes on all new Markdown
-pnpm format --check
-# Expected: exit 0
-```
-
-## Acceptance criteria
-
-- [ ] `docs/adr/` exists and contains `README.md` + 24 numbered ADR files
-- [ ] `apps/claude-code/auto-format/docs/adr/` exists and contains `README.md` + 8 numbered ADR files
-- [ ] `apps/claude-code/pr-review/docs/adr/` exists and contains `README.md` + 8 numbered ADR files
-- [ ] `apps/claude-code/unic-confluence/docs/adr/` exists and contains `README.md` + 12 numbered ADR files
-- [ ] All ADR files follow MADR-lite format: status line, `## Context`, `## Decision`, `## Consequences`
-- [ ] `pnpm format --check` passes (Prettier on all new Markdown)
-- [ ] No existing spec files, source files, or config files were modified
-
-## Out of scope
-
-- Editing existing specs (`docs/plans/`) to cross-link to ADRs
-- Creating `CONTEXT.md` or `CONTEXT-MAP.md` (deferred to `/grill-with-docs`)
-- Re-evaluating or amending any of the recorded decisions
-- Adding ADRs for decisions not yet captured in specs or tooling
-- Updating `docs/agents/domain.md` (it already references `docs/adr/` correctly)
diff --git a/docs/plans/17-fix-inbox-ls-collision-check-cross-platform.md b/docs/plans/17-fix-inbox-ls-collision-check-cross-platform.md
deleted file mode 100644
index afda3411..00000000
--- a/docs/plans/17-fix-inbox-ls-collision-check-cross-platform.md
+++ /dev/null
@@ -1,82 +0,0 @@
-# 17. Fix inbox command ls collision check for cross-platform support
-**Status: done — 2026-05-03**
-
-**Priority:** P2
-**Effort:** XS
-**Version impact:** none
-**Depends on:** —
-**Touches:** `.claude/commands/inbox.md`
-
-## Context
-
-The repo mandates that all tooling works on macOS, Windows, and Linux. Step 2 of the inbox capture command uses `ls <path> 2>/dev/null` to detect slug collisions. Both `ls` and the `2>/dev/null` redirect are POSIX-only and break on Windows. The `$(git rev-parse --show-toplevel)` shell substitution inside the same snippet also relies on bash syntax, which is unavailable in cmd.exe environments.
-
-## Current behaviour
-
-Step 2 of `.claude/commands/inbox.md` reads:
-
-```bash
-ls "$(git rev-parse --show-toplevel)/docs/inbox/<slug>.md" 2>/dev/null
-```
-
-This fails on Windows because:
-- `ls` is not a native Windows command
-- `2>/dev/null` is POSIX-only redirect syntax
-- `$(…)` command substitution is not valid in cmd.exe
-
-## Target behaviour
-
-Step 2 uses a cross-platform Node.js snippet. Node resolves the git root internally (no shell substitution), uses `path.join` for OS-correct separators, and checks existence with `fs.existsSync`. Exit code semantics are preserved: `0` = file exists (collision), non-zero = free.
-
-The `allowed-tools` header stays `[Write, Bash]` — no new tool permissions are needed.
-
-## Affected files
-
-| File                           | Change                                      |
-| ------------------------------ | ------------------------------------------- |
-| `.claude/commands/inbox.md`    | Modify — replace Step 2 code block only     |
-
-## Implementation steps
-
-1. In `.claude/commands/inbox.md`, replace the code block in **Step 2** (lines 31–33):
-
-   **Before:**
-   ```bash
-   ls "$(git rev-parse --show-toplevel)/docs/inbox/<slug>.md" 2>/dev/null
-   ```
-
-   **After:**
-   ```bash
-   node -e "
-     const root = require('child_process').execSync('git rev-parse --show-toplevel').toString().trim();
-     const p = require('path').join(root, 'docs', 'inbox', '<slug>.md');
-     process.exit(require('fs').existsSync(p) ? 0 : 1);
-   "
-   ```
-
-   No other changes to the file.
-
-2. Commit:
-
-   ```sh
-   git commit -m "fix(inbox): replace shell ls collision check with cross-platform Node.js"
-   ```
-
-## Verification
-
-- Open `.claude/commands/inbox.md` and confirm the Step 2 code block contains `node -e` with no `ls`, no `2>/dev/null`, and no `$(…)` shell substitution.
-- Run `/inbox test cross-platform fix` in a Claude Code session and confirm it creates `docs/inbox/test-cross-platform-fix.md` without errors.
-- Run `/inbox test cross-platform fix` a second time and confirm the file is created as `docs/inbox/test-cross-platform-fix-2.md` (collision handling works).
-- Clean up: delete both test files.
-
-## Acceptance criteria
-
-- [ ] `.claude/commands/inbox.md` Step 2 uses `node -e` with `execSync`, `path.join`, and `fs.existsSync`
-- [ ] No `ls`, `2>/dev/null`, or `$(…)` in the file
-- [ ] `allowed-tools` remains `[Write, Bash]`
-- [ ] Collision detection still appends `-2`, `-3`, etc. when a slug already exists
-
-## Out of scope
-
-- Updating the inbox skill or any other command files
-- Changing how Step 3 resolves the repo root (standalone `git rev-parse` is cross-platform)
diff --git a/docs/plans/README.md b/docs/plans/README.md
deleted file mode 100644
index dd7b5208..00000000
--- a/docs/plans/README.md
+++ /dev/null
@@ -1,58 +0,0 @@
-# Monorepo Spec Roadmap
-
-This directory contains the Ralph-iterable spec set for the `unic-agents-plugins` monorepo migration and bootstrapping. Each spec is implemented by Ralph in one iteration.
-
-Individual plugins have their own `docs/plans/` for plugin-specific development.
-
-## Execution order
-
-| # | Spec | What ships | Depends on |
-|---|---|---|---|
-| 00 | [bootstrap-validation](00-bootstrap-validation.md) | Seed verification; first commit | — |
-| 01 | [tsconfig-package](01-tsconfig-package.md) | `@unic/tsconfig` package with `tsconfig.base.json` | 00 |
-| 02 | [biome-config-package](02-biome-config-package.md) | `@unic/biome-config` package; root `biome.json` extends it | 01 |
-| 03 | [release-tools-package](03-release-tools-package.md) | `@unic/release-tools` with bump/sync/tag/verify scripts and bin commands | 02 |
-| 04 | [root-docs](04-root-docs.md) | `docs/process/` templates; finalize cross-cutting docs | 03 |
-| 05 | [migrate-pr-review](05-migrate-pr-review.md) | `apps/claude-code/pr-review/` with history; plugin renamed to `pr-review` | 03 |
-| 06 | [migrate-auto-format](06-migrate-auto-format.md) | `apps/claude-code/auto-format/` with history; plugin renamed to `auto-format` | 03 |
-| 07 | [migrate-confluence-publish](07-migrate-confluence-publish.md) | `apps/claude-code/confluence-publish/` with history; plugin renamed to `confluence-publish` | 03 |
-| 08 | [root-marketplace](08-root-marketplace.md) | `.claude-plugin/marketplace.json` listing all three plugins | 05, 06, 07 |
-| 09 | [ci-workflow](09-ci-workflow.md) | `.github/workflows/ci.yml` — path-filtered matrix (3 OS × 2 Node) | 08 |
-| 10 | [release-workflow](10-release-workflow.md) | `.github/workflows/release.yml` — per-plugin version-diff tagging | 09 |
-| 11 | [smoke-tests](11-smoke-tests.md) | Cross-platform install and run validation | 10 |
-| 12 | [first-releases](12-first-releases.md) | Cut `pr-review@0.1.1`, `auto-format@0.5.5`, `confluence-publish@2.1.6` | 11 |
-| 13 | [archive-old-repos](13-archive-old-repos.md) | Final CHANGELOG entries; archive old GitHub repos | 12 |
-| 14 | [fix-release-workflow](14-fix-release-workflow.md) | Fix tag detection to use tag-existence check instead of HEAD~1 diff | 12 |
-| 15 | [release-tools-tests](15-release-tools-tests.md) | `bump-version.test.mjs` + `sync-version.test.mjs` for `@unic/release-tools` | 03 |
-
-## Backlog
-
-Ideas that need a spec before Ralph can run them. When you're ready to act on one, ask Claude to write the spec (`NN-name.md`), review it, then hand it to Ralph.
-
-| Idea | Priority | Notes |
-|---|---|---|
-| Update plugin `CLAUDE.md` files for monorepo context | P1 | All three plugins (`pr-review`, `auto-format`, `confluence-publish`) still have standalone-repo instructions; deferred in specs 05–07 |
-| Bake `pnpm format` into `bump-version.mjs` | P2 | Every bump leaves CHANGELOG.md in a state Prettier disagrees with, requiring a manual `pnpm format` step; discovered during spec 14 |
-| GitHub Release notes automation | P3 | Auto-generate release notes from CHANGELOG.md when the release workflow fires a tag |
-
-**Manual actions pending (not Ralph specs):**
-
-- Execute smoke test checklist (`docs/process/smoke-test-checklist.md`) — requires a scratch Claude Code environment, an Azure DevOps PR URL, and Confluence credentials
-
-## SemVer policy (per plugin)
-
-| Bump | When |
-|---|---|
-| **major** | Breaking change to CLI flags, exit codes, or on-disk file schemas |
-| **minor** | New flag, subcommand, or user-visible feature |
-| **patch** | Bug fix, refactor, docs, internal tooling |
-
-## Ground rules
-
-- **pnpm** for all package operations.
-- **Tabs** in `.mjs`/`.js`/`.ts`; **spaces (2)** in `.json`/`.yml`/`.yaml`.
-- **Conventional commits** with package scope: `feat(auto-format): …`, `chore(spec-NN): …`.
-- **Cross-platform**: Node.js APIs only — no shell, no POSIX path assumptions.
-- **`pnpm --filter <name> bump <type>`** is the only way to change a plugin version.
-- Workspace specs (00–04, 08–13) do not bump any version. Commit with `chore(spec-NN): …`.
-- Plugin migration specs (05–07) bump the plugin's version. Commit with `feat(spec-NN): … (vX.Y.Z)`.
diff --git a/docs/process/ai-development.md b/docs/process/ai-development.md
index a9fb516b..d5e95c5f 100644
--- a/docs/process/ai-development.md
+++ b/docs/process/ai-development.md
@@ -4,22 +4,21 @@ This guide explains the mental model behind the AI-development workflow, the arc
 
 ---
 
-## 1. Two runners, not one
+## 1. The Feature Runner
 
-The most important thing to understand is that this repo has two distinct execution loops, and they are not interchangeable.
+A **Feature Runner** is the skill that implements a Feature's issues end-to-end in one worktree, branch, and pull request (see root `CONTEXT.md`).
 
-|                       | Spec Runner                                                                       | Feature Runner                                             |
-| --------------------- | --------------------------------------------------------------------------------- | ---------------------------------------------------------- |
-| **Input**             | `docs/plans/NN-*.md` Spec                                                         | `docs/issues/<slug>/NN-*.md` Issue                         |
-| **Invocation**        | `pnpm ralph`                                                                      | `/implement-feature`                                       |
-| **Format**            | Prescriptive: before/after snapshots, shell verification commands, explicit steps | Descriptive: `## What to build` + `## Acceptance criteria` |
-| **Worker**            | Agent follows spec as recipe (or `/tdd` for behavioral specs)                     | `/tdd` in non-interactive AFK mode                         |
-| **Completion marker** | `**Status: done**` in spec file                                                   | `Status: resolved` in issue file                           |
-| **Branch**            | Current branch                                                                    | `feature/afk/<slug>` worktree                              |
+New work enters as a GitHub Issue — the canonical tracker for state and ownership (see [docs/agents/issue-tracker.md](../agents/issue-tracker.md)). Once an idea is grilled into a Feature, `/to-prd` and `/to-issues` materialise a `docs/issues/<slug>/` directory with the PRD and the numbered ticket files the Feature Runner reads. GitHub Issues remain the source of truth for triage state; `docs/issues/<slug>/` is the markdown artifact set that captures the grilled scope. Not every GitHub Issue becomes a Feature directory (small fixes never need one).
 
-**When to use which:** The Spec Runner is for building and evolving the repo itself — release tooling, CI configuration, monorepo infrastructure. The Feature Runner is for product work on top of a stable system — new plugin capabilities, improvements to existing features. A rough heuristic: if the work would change something under `packages/` or `.github/`, it belongs in a Spec. If it changes something under `apps/claude-code/<plugin>/`, it belongs in a Feature.
+|                       | Feature Runner                                             |
+| --------------------- | ---------------------------------------------------------- |
+| **Input**             | `docs/issues/<slug>/NN-*.md` Issue                         |
+| **Format**            | Descriptive: `## What to build` + `## Acceptance criteria` |
+| **Worker**            | `/tdd`                                                     |
+| **Completion marker** | `Status: resolved` in issue file                           |
+| **Branch**            | `feature/<name>` (or `feature/afk/<slug>` when AFK)        |
 
-Both runners are backed by the same agent; the difference is in what inputs they receive and how much the agent is expected to figure out on its own.
+`unic-dlc-build` (shipped by `unic-archon-dlc`) is the long-term Feature Runner — see [ADR-0009](../../apps/claude-code/unic-archon-dlc/docs/adr/0009-retire-ralph-adopt-archon-runner.md) and [ADR-0010](../../apps/claude-code/unic-archon-dlc/docs/adr/0010-retire-implement-feature-skill.md). Until it is wired into this repo, the runner is **the developer driving `/tdd` manually**, one issue at a time. Infrastructure work (CI, tooling, packages) and product work (plugin features) both enter through the issue tracker — the split is in the issue content, not in which runner handles it.
 
 ---
 
@@ -28,28 +27,28 @@ Both runners are backed by the same agent; the difference is in what inputs they
 Every piece of work passes through a pipeline before an agent executes it. Each stage has a human-review checkpoint:
 
 ```
-docs/inbox/          ← /inbox: raw capture, no review required
+GitHub Issue / /triage  ← raw capture, no review required
        ↓
-/grill-with-docs     ← human reviews every branch of the design tree
+/grill-with-docs        ← human reviews every branch of the design tree
        ↓
-/to-prd              ← human reviews the synthesized PRD
+/to-prd                 ← human reviews the synthesized PRD
        ↓
-/to-issues           ← human reviews the vertical slice breakdown
+/to-issues              ← human reviews the vertical slice breakdown
        ↓
-/triage              ← human moves issues to ready-for-agent
+/triage                 ← human moves issues to ready-for-agent
        ↓
-/implement-feature   ← agent executes, no human present
+/tdd  (or unic-dlc-build, when wired up)  ← execution
 ```
 
-The pipeline is load-bearing. The quality of the autonomous execution at the bottom depends entirely on the quality of the decisions captured at each stage above it. A vague acceptance criterion that slips through triage will produce a vague implementation — and there is no human in the loop to catch it until the PR review.
+The pipeline is load-bearing. The quality of the execution at the bottom depends entirely on the quality of the decisions captured at each stage above it. A vague acceptance criterion that slips through triage will produce a vague implementation. Under manual `/tdd` you can still catch it interactively; under AFK execution there is no human in the loop until PR review.
 
 ---
 
 ## 3. Why context quality determines AFK quality
 
-When `/tdd` runs inside the Feature Runner, it runs non-interactively. In a normal interactive session, `/tdd`'s planning phase asks the user to confirm interface changes and approve which behaviours to test before writing any code. In AFK mode there is no user to ask.
+`/tdd` is interactive by default: its planning phase asks the user to confirm interface changes and approve which behaviours to test before writing any code. When `/tdd` is run interactively, that conversation is where most ambiguity is eliminated.
 
-The issue's `## Acceptance criteria` replaces that conversation. The planning phase is not skipped — it was completed during the grilling and issue-writing stages. The Feature Runner simply does not repeat it at runtime.
+In AFK mode (the future `unic-dlc-build` path) there is no user to ask. The issue's `## Acceptance criteria` replaces that conversation. The planning phase is not skipped — it must have been completed during the grilling and issue-writing stages.
 
 This means there is a direct line between **grilling quality → PRD quality → issue acceptance criteria quality → implementation correctness**. If any link in that chain is weak, the agent produces a _correct-but-wrong_ implementation: code that satisfies the literal issue description but diverges from what you actually intended.
 
@@ -57,32 +56,9 @@ The grilling session (`/grill-with-docs`) is where that chain is forged. It is n
 
 ---
 
-## 4. The context bundle
+## 4. Writing issues that AFK agents can execute
 
-When the Feature Runner invokes `/tdd` for an issue, it does not pass only the issue file. It assembles a **context bundle** from six sources:
-
-| Input                       | Why it matters                                                                                                                             |
-| --------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ |
-| **Issue file**              | The `## What to build` and `## Acceptance criteria` — the pre-answered plan                                                                |
-| **PRD**                     | The "why" behind the feature; the shared vision from grilling. Without it, the agent reasons from a vertical slice with no broader context |
-| **Sibling issue files**     | Dependency awareness; "what is already resolved" without the runner summarising                                                            |
-| **Scoped CONTEXT.md**       | Domain glossary — ensures test names and interfaces match the project's vocabulary                                                         |
-| **Scoped ADRs**             | Architectural constraints the implementation must respect                                                                                  |
-| **Recent commits (last 5)** | The ideation trail — grilling sessions typically modify CONTEXT.md and ADRs, and those changes land in commits before the runner executes  |
-
-### ADR scoping
-
-Not all ADRs are relevant to all work. Root `docs/adr/` covers monorepo concerns (versioning, tagging, CI) — those are noise for plugin implementation. Per-plugin `docs/adr/` covers domain-specific decisions that directly constrain what the agent builds.
-
-The runner infers scope from the PRD: if it references paths under `apps/claude-code/<plugin>/`, inject that plugin's ADRs and CONTEXT.md. If it references paths outside `apps/` (`.claude/`, `docs/`, `packages/`), inject the root ADRs and CONTEXT.md.
-
-This is why CONTEXT.md and ADRs must be kept current. They are not documentation artifacts — they are runtime inputs to every AFK agent execution.
-
----
-
-## 5. Writing issues that AFK agents can execute
-
-An issue's `## Acceptance criteria` is doing two jobs: it is the definition of done for the human reviewer, and it is the planning conversation substitute for the AFK agent. It must be specific enough for both audiences.
+An issue's `## Acceptance criteria` is doing two jobs: it is the definition of done for the human reviewer, and it is the planning conversation substitute for any AFK agent. It must be specific enough for both audiences.
 
 **Good acceptance criteria** are checkable without ambiguity:
 
@@ -109,37 +85,21 @@ If an issue's acceptance criteria are too vague to verify without judgment, the
 
 ---
 
-## 6. Dependency ordering
+## 5. Dependency ordering
 
 Issues produced by `to-issues` are named with a numeric prefix (`01-`, `02-`, etc.) for human readability. The numbers usually reflect dependency order because `to-issues` publishes blockers first. But **numerical order is not the execution contract**.
 
-The `## Blocked by` field in each issue is the canonical dependency signal. The Feature Runner builds a topological order from `## Blocked by` references before executing anything. If `## Blocked by` and numerical order conflict, the runner halts rather than proceeding silently in the wrong order — because a wrong execution order means downstream issues inherit a broken foundation.
+The `## Blocked by` field in each issue is the canonical dependency signal — see [ADR-0007](../../apps/claude-code/unic-archon-dlc/docs/adr/0007-blocked-by-canonical-sequencing.md). Any Feature Runner (manual or AFK) must respect `## Blocked by` over filename order. If they conflict, the runner halts rather than proceeding silently in the wrong order — because a wrong execution order means downstream issues inherit a broken foundation.
 
 When writing or reviewing issues: always fill in `## Blocked by` accurately. "None — can start immediately" is a valid and important signal. A missing or incorrect `## Blocked by` is more dangerous than a missing acceptance criterion, because the sequencing error compounds silently across every subsequent issue.
 
-The dependency graph also reveals which issues are parallelisable (those with no blockers and no dependents). The Feature Runner serialises all execution by design — parallel issue execution is explicitly out of scope — but understanding which issues are independent helps when manually intervening in a failed run.
+The dependency graph also reveals which issues are parallelisable (those with no blockers and no dependents). The current `/tdd`-per-issue flow serialises execution by default; understanding which issues are independent helps when manually intervening in a failed run.
 
 ---
 
-## 7. Running overnight
+## 6. CONTEXT.md and ADRs as living constraints
 
-The Feature Runner is designed to be composable with `/loop` for unattended overnight execution:
-
-```
-/loop /implement-feature
-```
-
-When the queue empties (no qualifying feature exists — see `.claude/skills/implement-feature/SKILL.md` step 0 for the full qualification rule), the runner outputs `LOOP_COMPLETE` and the loop terminates cleanly. This mirrors the Spec Runner's `completion_promise: LOOP_COMPLETE` in `ralph.yml`.
-
-For overnight runs to succeed, the queue must be in good shape before you start: each target feature must qualify (see SKILL.md step 0) — every issue in `{ready-for-agent, resolved, closed, rejected, ready-for-human}`, no `needs-*` states, no conflicts between `## Blocked by` and numerical order, acceptance criteria specific enough to verify without judgment. A single malformed issue will halt the runner and leave the remainder of the queue unexecuted.
-
-If a `/tdd` invocation fails mid-feature, the failing issue is flipped to `needs-info` with a failure note appended. The runner stops. Subsequent issues in the same feature do not run — they could inherit a broken foundation. The `needs-info` flip prevents `/loop /implement-feature` from auto-selecting the same feature again until a developer triages the failure. Inspect the failure note, fix the issue or the codebase, set the issue back to `ready-for-agent`, and re-run. (Note: a Ctrl+C interrupt — as opposed to a `/tdd` failure — leaves the issue at `ready-for-agent` so a simple re-run resumes it.)
-
----
-
-## 8. CONTEXT.md and ADRs as living constraints
-
-`CONTEXT.md` and `docs/adr/` are not documentation you write once and forget. They are the vocabulary and constraint layer that every agent reads before writing code. Their quality directly affects the quality of every AFK execution.
+`CONTEXT.md` and `docs/adr/` are not documentation you write once and forget. They are the vocabulary and constraint layer that every agent reads before writing code. Their quality directly affects the quality of every execution — manual or AFK.
 
 **Update CONTEXT.md** when a new domain term is introduced or an existing term is redefined. `/grill-with-docs` does this automatically during a grilling session — terms resolved during grilling are written into CONTEXT.md inline. If a term surfaces outside a grilling session, add it manually.
 
@@ -147,35 +107,27 @@ If a `/tdd` invocation fails mid-feature, the failing issue is flipped to `needs
 
 **Never let ADRs drift.** An ADR that no longer reflects the codebase is worse than no ADR — it misdirects the agent. If a decision is superseded, update the original ADR's status to `Superseded by ADR-NNNN` and write the new one.
 
-The commits from your grilling sessions carry this context forward. The Feature Runner injects the last 5 commits into every `/tdd` invocation specifically because grilling sessions modify CONTEXT.md and ADRs — those changes land in commits and the agent needs the ideation trail, not just the final file state.
+The commits from your grilling sessions carry this context forward. When `unic-dlc-build` later runs AFK, recent commits are part of the context input precisely because grilling sessions modify CONTEXT.md and ADRs — those changes land in commits and the agent needs the ideation trail, not just the final file state.
 
 ---
 
-## 9. Keeping docs/plans/ and docs/issues/ in sync
-
-The Spec Runner and Feature Runner evolved independently. Work that was implemented via the Spec Runner (i.e. a Spec in `docs/plans/` was marked `done`) may have a corresponding directory in `docs/issues/<slug>/` that was never updated. The Feature Runner will attempt to implement those stale issues if they have `ready-for-agent` status.
+## 7. Historical: docs/plans/ and the interim `/implement-feature` skill
 
-The convention: when a Spec is marked `**Status: done**`, check for a corresponding `docs/issues/<slug>/` directory. If it exists, mark all issue files in it `closed` and append a note:
+Two earlier execution paths have been retired:
 
-```markdown
-## Comments
-
-> _Closed 2026-05-09 — implemented via Spec Runner (docs/plans/NN-<slug>.md marked done)._
-```
+- **`docs/plans/`** was the intake path for monorepo infrastructure specs (00–17), implemented by `ralph-orchestrator`. All specs are complete and the format is retired as of 2026-05. See [ADR-0009](../../apps/claude-code/unic-archon-dlc/docs/adr/0009-retire-ralph-adopt-archon-runner.md).
+- **`/implement-feature`** was an interim Claude Code skill that wrapped `/tdd` in a non-interactive sub-agent loop. It was retired in favour of converging on a single Feature Runner (`unic-dlc-build`). See [ADR-0010](../../apps/claude-code/unic-archon-dlc/docs/adr/0010-retire-implement-feature-skill.md). [ADRs 0006](../../apps/claude-code/unic-archon-dlc/docs/adr/0006-feature-runner-context-bundle.md) and [0008](../../apps/claude-code/unic-archon-dlc/docs/adr/0008-feature-runner-afk-invocation.md) describe its internals and are superseded.
 
-This is a manual step. There is no automation for it. The `docs/agents/feature-runner.md` reference document records this convention for agents that need to be briefed on it.
+If you encounter `docs/issues/<slug>/` directories whose issues were never closed because they were implemented via a spec or by the retired skill, mark them `closed` with a note referencing the work that covered them.
 
 ---
 
 ## Related
 
 - `docs/process/development-workflow.md` — the 8-phase quick reference
-- `docs/process/ralph-loop-guide.md` — Spec Runner invocation and resumption detail
-- `docs/process/spec-template.md` — spec file format
 - `docs/agents/issue-tracker.md` — issue file conventions
 - `docs/agents/triage-labels.md` — 8-state triage vocabulary
-- `docs/adr/0023-spec-template-format.md` — why specs are prescriptive
-- `docs/adr/0026-tdd-dispatch-by-version-impact.md` — when the Spec Runner uses /tdd
-- `docs/adr/0027-feature-runner-context-bundle.md` — what /tdd receives per invocation
-- `docs/adr/0028-blocked-by-canonical-sequencing.md` — why ## Blocked by beats filename order
-- `docs/adr/0029-feature-runner-afk-invocation.md` — how AFK invocation works
+- `apps/claude-code/unic-archon-dlc/docs/adr/0005-tdd-dispatch-by-version-impact.md` — when to use /tdd vs direct implementation (dispatch by version impact)
+- `apps/claude-code/unic-archon-dlc/docs/adr/0007-blocked-by-canonical-sequencing.md` — why ## Blocked by beats filename order
+- `apps/claude-code/unic-archon-dlc/docs/adr/0009-retire-ralph-adopt-archon-runner.md` — retirement of ralph and docs/plans/
+- `apps/claude-code/unic-archon-dlc/docs/adr/0010-retire-implement-feature-skill.md` — retirement of /implement-feature and /inbox
diff --git a/docs/process/development-workflow.md b/docs/process/development-workflow.md
index bddc319a..27c1155e 100644
--- a/docs/process/development-workflow.md
+++ b/docs/process/development-workflow.md
@@ -1,6 +1,6 @@
 # Development Workflow
 
-This repo follows an adapted 8-phase version of Matt Pocock's 7-phase AI development workflow. The phases move from raw idea capture through AFK execution to QA, using the tools already available here.
+This repo follows an adapted 8-phase version of Matt Pocock's 7-phase AI development workflow. The phases move from raw idea capture through execution to QA, using the tools already available here.
 
 Not every phase is required for every piece of work. A typo fix can go straight to execution. A major feature will touch every phase.
 
@@ -8,15 +8,11 @@ Not every phase is required for every piece of work. A typo fix can go straight
 
 ## Phase 1 — Capture the idea
 
-When an idea surfaces mid-conversation or mid-task, capture it without breaking flow:
+When an idea surfaces mid-conversation or mid-task, capture it without breaking flow by opening a GitHub Issue directly (or running `/triage` and letting it walk the idea through the state machine).
 
-```
-/inbox <one-liner>
-```
-
-This drops a file into `docs/inbox/<slug>.md`. No follow-up needed. Come back to it during triage.
+GitHub Issues are the canonical tracker (see [docs/agents/issue-tracker.md](../agents/issue-tracker.md)) — they hold raw ideas, bug reports, and triage state. Features that get grilled additionally pick up a `docs/issues/<slug>/` directory in Phase 5, where `/to-prd` and `/to-issues` write the PRD and ticket files the Feature Runner consumes. Not every GitHub Issue grows into a Feature directory — small fixes stay as plain GitHub Issues.
 
-If you already have enough context to start grilling immediately, skip the inbox and go straight to Phase 2.
+If you already have enough context to start grilling immediately, skip capture and go straight to Phase 2.
 
 ## Phase 2 — Grill the idea
 
@@ -30,8 +26,6 @@ Use `/grill-with-docs` when the topic involves domain concepts — it is designe
 
 The grilling session walks down every branch of the design tree until the idea is concrete: edge cases surfaced, ambiguities resolved, out-of-scope items named.
 
-**Inbox → grilling transition:** open the inbox file, pass it to `/grill-with-docs` as context, then delete the inbox file once the session completes.
-
 ## Phase 3 — Research (optional)
 
 If the work involves an unfamiliar external API, a complex integration, or anything that would require repeated exploration in fresh context windows, cache the findings:
@@ -72,37 +66,15 @@ Use the triage labels to track state — see `docs/agents/triage-labels.md` for
 
 ## Phase 7 — Execute
 
-There are two execution paths depending on the type of work. Choose based on where the work item lives, not on personal preference — the two runners are not interchangeable.
-
-### Spec Runner — for `docs/plans/` specs
-
-Use the Spec Runner when implementing infrastructure, tooling, or repo-level changes captured as Specs in `docs/plans/`:
-
-```
-pnpm ralph                        # root specs
-pnpm --filter <plugin> ralph      # plugin-specific specs
-```
+### Manual execution with `/tdd` — current default
 
-Specs follow a prescriptive format (before/after snapshots, shell verification commands, acceptance criteria). The Spec Runner implements one Spec per iteration, commits, and stops. See `docs/process/ralph-loop-guide.md`.
+Work through `ready-for-agent` issues one at a time with `/tdd`. For each issue, the `## Acceptance criteria` block stands in for the planning conversation. Mark the issue `resolved` when the implementation lands. Open a PR targeting `develop` once the feature's issues are done.
 
-### Feature Runner — for `docs/issues/` features
+Respect the issue ordering signalled by `## Blocked by` (see [ADR-0007](../../apps/claude-code/unic-archon-dlc/docs/adr/0007-blocked-by-canonical-sequencing.md)) — a downstream issue inherits a broken foundation if a blocker has not landed.
 
-Use the Feature Runner when implementing product features tracked as Issues in `docs/issues/<slug>/`. Once a feature has at least one `ready-for-agent` issue and no unprepped issues (`needs-triage`, `needs-info`, `needs-specs`):
-
-```
-/implement-feature <slug>         # target a specific feature
-/implement-feature                # auto-select next ready feature
-```
-
-The Feature Runner builds a dependency graph from `## Blocked by` references, invokes `/tdd` non-interactively for each issue in topological order, marks each issue `resolved` on completion, and opens a PR targeting `develop` when all issues are done.
-
-Compose with `/loop` for overnight queue draining:
-
-```
-/loop /implement-feature
-```
+### Feature Runner — long-term AFK execution
 
-The runner outputs `LOOP_COMPLETE` when the queue is empty, which terminates the loop cleanly.
+The Feature Runner (`unic-dlc-build`, shipped by `unic-archon-dlc`) is the AFK path going forward; see [ADR-0009](../../apps/claude-code/unic-archon-dlc/docs/adr/0009-retire-ralph-adopt-archon-runner.md) and [ADR-0010](../../apps/claude-code/unic-archon-dlc/docs/adr/0010-retire-implement-feature-skill.md). Until it is wired into this repo, AFK runs are not available — implement manually with `/tdd`.
 
 ### Human execution
 
@@ -118,23 +90,19 @@ Human QA often surfaces new issues or improvement ideas — add them back to the
 
 ## Quick reference
 
-| Phase                 | When                                           | Tool                                          |
-| --------------------- | ---------------------------------------------- | --------------------------------------------- |
-| 1. Capture            | Idea surfaces mid-task                         | `/inbox <one-liner>`                          |
-| 2. Grill              | Before any PRD or spec                         | `/grill-with-docs` or `/grill-me`             |
-| 3. Research           | Unfamiliar external dependencies               | `research.md` (ad hoc)                        |
-| 4. Prototype          | Uncertain design or UX                         | Ad hoc throwaway route                        |
-| 5. PRD                | After grilling                                 | `/to-prd` → `docs/issues/<slug>/PRD.md`       |
-| 6. Issues             | After PRD                                      | `/to-issues` → `docs/issues/<slug>/<NN>-*.md` |
-| 7a. Execute (Spec)    | Specs in `docs/plans/` are ready               | `pnpm ralph` (Spec Runner)                    |
-| 7b. Execute (Feature) | Issues in `docs/issues/` are `ready-for-agent` | `/implement-feature` (Feature Runner)         |
-| 8. QA                 | After execution                                | QA plan (agent-generated, human-verified)     |
+| Phase        | When                                           | Tool                                                         |
+| ------------ | ---------------------------------------------- | ------------------------------------------------------------ |
+| 1. Capture   | Idea surfaces mid-task                         | GitHub Issue (or `/triage`)                                  |
+| 2. Grill     | Before any PRD or spec                         | `/grill-with-docs` or `/grill-me`                            |
+| 3. Research  | Unfamiliar external dependencies               | `research.md` (ad hoc)                                       |
+| 4. Prototype | Uncertain design or UX                         | Ad hoc throwaway route                                       |
+| 5. PRD       | After grilling                                 | `/to-prd` → `docs/issues/<slug>/PRD.md`                      |
+| 6. Issues    | After PRD                                      | `/to-issues` → `docs/issues/<slug>/<NN>-*.md`                |
+| 7. Execute   | Issues in `docs/issues/` are `ready-for-agent` | `/tdd` per issue (Feature Runner via `unic-dlc-build` later) |
+| 8. QA        | After execution                                | QA plan (agent-generated, human-verified)                    |
 
 ## Related
 
-- `docs/inbox/README.md` — inbox conventions
 - `docs/agents/issue-tracker.md` — issue file conventions
 - `docs/agents/triage-labels.md` — 8-state triage vocabulary
-- `docs/process/ralph-loop-guide.md` — Spec Runner detail
-- `docs/process/spec-template.md` — spec file format
 - `docs/process/ai-development.md` — deep guide: mental model, context quality, AFK trust chain, key decisions
diff --git a/docs/process/ralph-loop-guide.md b/docs/process/ralph-loop-guide.md
deleted file mode 100644
index 61966d74..00000000
--- a/docs/process/ralph-loop-guide.md
+++ /dev/null
@@ -1,47 +0,0 @@
-# Ralph Loop Guide
-
-## What is the Ralph loop?
-
-Ralph Orchestrator is a Claude Code plugin that runs specs from `docs/plans/` one at a time. Each iteration implements one spec, commits, and stops. Re-running the loop picks up the next unfinished spec.
-
-## Key files
-
-- **`ralph.yml`** — Loop configuration: which model to use, which `PROMPT.md` to load, concurrency settings.
-- **`PROMPT.md`** — The objective injected into each Ralph iteration. For the monorepo root, it instructs Ralph to scan specs in order and implement the first one not marked `**Status: done**`.
-
-## Starting a loop
-
-From the monorepo root (implements the next pending spec):
-
-```sh
-pnpm ralph
-```
-
-From inside a plugin directory (runs that plugin's own development loop):
-
-```sh
-cd apps/claude-code/pr-review
-pnpm ralph
-```
-
-## Stopping and resuming
-
-- **Stop early**: `Ctrl+C` — safe at any point; the current spec is left in progress.
-- **Resume**: `pnpm ralph` again — Ralph finds the first spec without `**Status: done**` and continues.
-
-## Paused for human review
-
-When Ralph encounters a decision that requires human judgment, it:
-
-1. Adds a `## Questions` section to the bottom of the spec file.
-2. Emits `LOOP_COMPLETE`, which pauses the loop.
-
-To resume: answer the question in the spec file, remove the `## Questions` section, then run `pnpm ralph` again.
-
-## Single-spec-per-iteration discipline
-
-Ralph implements **one spec per run**. This keeps commits atomic and makes it easy to review, revert, or re-run a single change. Do not modify `PROMPT.md` to skip this discipline.
-
-## Writing new specs
-
-Use `docs/process/spec-template.md` as the canonical starting point. Register the new spec in `docs/plans/README.md`.
diff --git a/package.json b/package.json
index fbda189f..c29f569d 100644
--- a/package.json
+++ b/package.json
@@ -13,8 +13,7 @@
     "check": "biome check . && prettier --check \"**/*.md\"",
     "ci:check": "biome ci . && prettier --check \"**/*.md\"",
     "test": "pnpm -r --if-present test",
-    "typecheck": "pnpm -r --if-present typecheck",
-    "ralph": "ralph run -c ralph.yml -H builtin:code-assist"
+    "typecheck": "pnpm -r --if-present typecheck"
   },
   "devDependencies": {
     "@biomejs/biome": "catalog:",
@@ -24,8 +23,5 @@
     "@unic/tsconfig": "workspace:*",
     "prettier": "catalog:",
     "typescript": "catalog:"
-  },
-  "optionalDependencies": {
-    "@ralph-orchestrator/ralph-cli": "catalog:"
   }
 }
diff --git a/pnpm-lock.yaml b/pnpm-lock.yaml
index d616ec38..77a045a9 100644
--- a/pnpm-lock.yaml
+++ b/pnpm-lock.yaml
@@ -9,9 +9,6 @@ catalogs:
     '@biomejs/biome':
       specifier: 2.4.0
       version: 2.4.0
-    '@ralph-orchestrator/ralph-cli':
-      specifier: 2.9.2
-      version: 2.9.2
     '@types/node':
       specifier: 24.12.2
       version: 24.12.2
@@ -47,16 +44,9 @@ importers:
       typescript:
         specifier: 'catalog:'
         version: 5.8.3
-    optionalDependencies:
-      '@ralph-orchestrator/ralph-cli':
-        specifier: 'catalog:'
-        version: 2.9.2
 
   apps/claude-code/auto-format:
     devDependencies:
-      '@ralph-orchestrator/ralph-cli':
-        specifier: 'catalog:'
-        version: 2.9.2
       '@types/node':
         specifier: 'catalog:'
         version: 24.12.2
@@ -72,9 +62,6 @@ importers:
 
   apps/claude-code/pr-review:
     devDependencies:
-      '@ralph-orchestrator/ralph-cli':
-        specifier: 'catalog:'
-        version: 2.9.2
       '@unic/release-tools':
         specifier: workspace:*
         version: link:../../../packages/release-tools
@@ -100,9 +87,6 @@ importers:
         specifier: 17.0.5
         version: 17.0.5
     devDependencies:
-      '@ralph-orchestrator/ralph-cli':
-        specifier: 'catalog:'
-        version: 2.9.2
       '@types/node':
         specifier: 'catalog:'
         version: 24.12.2
@@ -188,179 +172,19 @@ packages:
     cpu: [x64]
     os: [win32]
 
-  '@ralph-orchestrator/ralph-cli@2.9.2':
-    resolution: {integrity: sha512-eRr1dc3SN0d5lnO9jFlc01Y/9EUDkQKXGIaSPNkBwVFZ0DLBf5SUk6fZswch8D4I+sEXsBLYQ423QlZkvNKhCw==}
-    engines: {node: '>=14', npm: '>=6'}
-    hasBin: true
-
   '@types/node@24.12.2':
     resolution: {integrity: sha512-A1sre26ke7HDIuY/M23nd9gfB+nrmhtYyMINbjI1zHJxYteKR6qSMX56FsmjMcDb3SMcjJg5BiRRgOCC/yBD0g==}
 
-  asynckit@0.4.0:
-    resolution: {integrity: sha512-Oei9OH4tRh0YqU3GxhX79dM/mwVgvbZJaSNaRk+bshkj0S5cfHcgYakreBjrHwatXKbz+IoIdYLxrKim2MjW0Q==}
-
-  axios-proxy-builder@0.1.2:
-    resolution: {integrity: sha512-6uBVsBZzkB3tCC8iyx59mCjQckhB8+GQrI9Cop8eC7ybIsvs/KtnNgEBfRMSEa7GqK2VBGUzgjNYMdPIfotyPA==}
-
-  axios@1.15.2:
-    resolution: {integrity: sha512-wLrXxPtcrPTsNlJmKjkPnNPK2Ihe0hn0wGSaTEiHRPxwjvJwT3hKmXF4dpqxmPO9SoNb2FsYXj/xEo0gHN+D5A==}
-
-  balanced-match@4.0.4:
-    resolution: {integrity: sha512-BLrgEcRTwX2o6gGxGOCNyMvGSp35YofuYzw9h1IMTRmKqttAZZVU67bdb9Pr2vUHA8+j3i2tJfjO6C6+4myGTA==}
-    engines: {node: 18 || 20 || >=22}
-
-  brace-expansion@5.0.5:
-    resolution: {integrity: sha512-VZznLgtwhn+Mact9tfiwx64fA9erHH/MCXEUfB/0bX/6Fz6ny5EGTXYltMocqg4xFAQZtnO3DHWWXi8RiuN7cQ==}
-    engines: {node: 18 || 20 || >=22}
-
-  call-bind-apply-helpers@1.0.2:
-    resolution: {integrity: sha512-Sp1ablJ0ivDkSzjcaJdxEunN5/XvksFJ2sMBFfq6x0ryhQV/2b/KwFe21cMpmHtPOSij8K99/wSfoEuTObmuMQ==}
-    engines: {node: '>= 0.4'}
-
-  clone@1.0.4:
-    resolution: {integrity: sha512-JQHZ2QMW6l3aH/j6xCqQThY/9OH4D/9ls34cgkUBiEeocRTU04tHfKPBsUK1PqZCUQM7GiA0IIXJSuXHI64Kbg==}
-    engines: {node: '>=0.8'}
-
-  combined-stream@1.0.8:
-    resolution: {integrity: sha512-FQN4MRfuJeHf7cBbBMJFXhKSDq+2kAArBlmRBvcvFE5BB1HZKXtSFASDhdlz9zOYwxh8lDdnvmMOe/+5cdoEdg==}
-    engines: {node: '>= 0.8'}
-
-  console.table@0.10.0:
-    resolution: {integrity: sha512-dPyZofqggxuvSf7WXvNjuRfnsOk1YazkVP8FdxH4tcH2c37wc79/Yl6Bhr7Lsu00KMgy2ql/qCMuNu8xctZM8g==}
-    engines: {node: '> 0.10'}
-
-  defaults@1.0.4:
-    resolution: {integrity: sha512-eFuaLoy/Rxalv2kr+lqMlUnrDWV+3j4pljOIJgLIhI058IQfWJ7vXhyEIHu+HtC738klGALYxOKDO0bQP3tg8A==}
-
-  delayed-stream@1.0.0:
-    resolution: {integrity: sha512-ZySD7Nf91aLB0RxL4KGrKHBXl7Eds1DAmEdcoVawXnLD7SDhpNgtuII2aAkg7a7QS41jxPSZ17p4VdGnMHk3MQ==}
-    engines: {node: '>=0.4.0'}
-
-  detect-libc@2.1.2:
-    resolution: {integrity: sha512-Btj2BOOO83o3WyH59e8MgXsxEQVcarkUOpEYrubB0urwnN10yQ364rsiByU11nZlqWYZm05i/of7io4mzihBtQ==}
-    engines: {node: '>=8'}
-
-  dunder-proto@1.0.1:
-    resolution: {integrity: sha512-KIN/nDJBQRcXw0MLVhZE9iQHmG68qAVIBg9CqmUYjmQIhgij9U5MFvrqkUL5FbtyyzZuOeOt0zdeRe4UY7ct+A==}
-    engines: {node: '>= 0.4'}
-
-  easy-table@1.1.0:
-    resolution: {integrity: sha512-oq33hWOSSnl2Hoh00tZWaIPi1ievrD9aFG82/IgjlycAnW9hHx5PkJiXpxPsgEE+H7BsbVQXFVFST8TEXS6/pA==}
-
-  es-define-property@1.0.1:
-    resolution: {integrity: sha512-e3nRfgfUZ4rNGL232gUgX06QNyyez04KdjFrF+LTRoOXmrOgFKDg4BCdsjW8EnT69eqdYGmRpJwiPVYNrCaW3g==}
-    engines: {node: '>= 0.4'}
-
-  es-errors@1.3.0:
-    resolution: {integrity: sha512-Zf5H2Kxt2xjTvbJvP2ZWLEICxA6j+hAmMzIlypy4xcBg1vKVnx89Wy0GbS+kf5cwCVFFzdCFh2XSCFNULS6csw==}
-    engines: {node: '>= 0.4'}
-
-  es-object-atoms@1.1.1:
-    resolution: {integrity: sha512-FGgH2h8zKNim9ljj7dankFPcICIK9Cp5bm+c2gQSYePhpaG5+esrLODihIorn+Pe6FGJzWhXQotPv73jTaldXA==}
-    engines: {node: '>= 0.4'}
-
-  es-set-tostringtag@2.1.0:
-    resolution: {integrity: sha512-j6vWzfrGVfyXxge+O0x5sh6cvxAog0a/4Rdd2K36zCMV5eJ+/+tOAngRO8cODMNWbVRdVlmGZQL2YS3yR8bIUA==}
-    engines: {node: '>= 0.4'}
-
-  follow-redirects@1.16.0:
-    resolution: {integrity: sha512-y5rN/uOsadFT/JfYwhxRS5R7Qce+g3zG97+JrtFZlC9klX/W5hD7iiLzScI4nZqUS7DNUdhPgw4xI8W2LuXlUw==}
-    engines: {node: '>=4.0'}
-    peerDependencies:
-      debug: '*'
-    peerDependenciesMeta:
-      debug:
-        optional: true
-
-  form-data@4.0.5:
-    resolution: {integrity: sha512-8RipRLol37bNs2bhoV67fiTEvdTrbMUYcFTiy3+wuuOnUog2QBHCZWXDRijWQfAkhBj2Uf5UnVaiWwA5vdd82w==}
-    engines: {node: '>= 6'}
-
-  function-bind@1.1.2:
-    resolution: {integrity: sha512-7XHNxH7qX9xG5mIwxkhumTox/MIRNcOgDrxWsMt2pAr23WHp6MrRlN7FBSFpCpr+oVO0F744iUgR82nJMfG2SA==}
-
-  get-intrinsic@1.3.0:
-    resolution: {integrity: sha512-9fSjSaos/fRIVIp+xSJlE6lfwhES7LNtKaCBIamHsjr2na1BiABJPo0mOjjz8GJDURarmCPGqaiVg5mfjb98CQ==}
-    engines: {node: '>= 0.4'}
-
-  get-proto@1.0.1:
-    resolution: {integrity: sha512-sTSfBjoXBp89JvIKIefqw7U2CCebsc74kiY6awiGogKtoSGbgjYE/G/+l9sF3MWFPNc9IcoOC4ODfKHfxFmp0g==}
-    engines: {node: '>= 0.4'}
-
-  glob@13.0.6:
-    resolution: {integrity: sha512-Wjlyrolmm8uDpm/ogGyXZXb1Z+Ca2B8NbJwqBVg0axK9GbBeoS7yGV6vjXnYdGm6X53iehEuxxbyiKp8QmN4Vw==}
-    engines: {node: 18 || 20 || >=22}
-
-  gopd@1.2.0:
-    resolution: {integrity: sha512-ZUKRh6/kUFoAiTAtTYPZJ3hw9wNxx+BIBOijnlG9PnrJsCcSjs1wyyD6vJpaYtgnzDrKYRSqf3OO6Rfa93xsRg==}
-    engines: {node: '>= 0.4'}
-
-  has-symbols@1.1.0:
-    resolution: {integrity: sha512-1cDNdwJ2Jaohmb3sg4OmKaMBwuC48sYni5HUw2DvsC8LjGTLK9h+eb1X6RyuOHe4hT0ULCW68iomhjUoKUqlPQ==}
-    engines: {node: '>= 0.4'}
-
-  has-tostringtag@1.0.2:
-    resolution: {integrity: sha512-NqADB8VjPFLM2V0VvHUewwwsw0ZWBaIdgo+ieHtK3hasLz4qeCRjYcqfB6AQrBggRKppKF8L52/VqdVsO47Dlw==}
-    engines: {node: '>= 0.4'}
-
-  hasown@2.0.3:
-    resolution: {integrity: sha512-ej4AhfhfL2Q2zpMmLo7U1Uv9+PyhIZpgQLGT1F9miIGmiCJIoCgSmczFdrc97mWT4kVY72KA+WnnhJ5pghSvSg==}
-    engines: {node: '>= 0.4'}
-
-  lru-cache@11.3.5:
-    resolution: {integrity: sha512-NxVFwLAnrd9i7KUBxC4DrUhmgjzOs+1Qm50D3oF1/oL+r1NpZ4gA7xvG0/zJ8evR7zIKn4vLf7qTNduWFtCrRw==}
-    engines: {node: 20 || >=22}
-
   marked@17.0.5:
     resolution: {integrity: sha512-6hLvc0/JEbRjRgzI6wnT2P1XuM1/RrrDEX0kPt0N7jGm1133g6X7DlxFasUIx+72aKAr904GTxhSLDrd5DIlZg==}
     engines: {node: '>= 20'}
     hasBin: true
 
-  math-intrinsics@1.1.0:
-    resolution: {integrity: sha512-/IXtbwEk5HTPyEwyKX6hGkYXxM9nbj64B+ilVJnC/R6B0pH5G4V3b0pVbL7DBj4tkhBAppbQUlf6F6Xl9LHu1g==}
-    engines: {node: '>= 0.4'}
-
-  mime-db@1.52.0:
-    resolution: {integrity: sha512-sPU4uV7dYlvtWJxwwxHD0PuihVNiE7TyAbQ5SWxDCB9mUYvOgroQOwYQQOKPJ8CIbE+1ETVlOoK1UC2nU3gYvg==}
-    engines: {node: '>= 0.6'}
-
-  mime-types@2.1.35:
-    resolution: {integrity: sha512-ZDY+bPm5zTTF+YpCrAU9nK0UgICYPT0QtT1NZWFv4s++TNkcgVaT0g6+4R2uI4MjQjzysHB1zxuWL50hzaeXiw==}
-    engines: {node: '>= 0.6'}
-
-  minimatch@10.2.5:
-    resolution: {integrity: sha512-MULkVLfKGYDFYejP07QOurDLLQpcjk7Fw+7jXS2R2czRQzR56yHRveU5NDJEOviH+hETZKSkIk5c+T23GjFUMg==}
-    engines: {node: 18 || 20 || >=22}
-
-  minipass@7.1.3:
-    resolution: {integrity: sha512-tEBHqDnIoM/1rXME1zgka9g6Q2lcoCkxHLuc7ODJ5BxbP5d4c2Z5cGgtXAku59200Cx7diuHTOYfSBD8n6mm8A==}
-    engines: {node: '>=16 || 14 >=14.17'}
-
-  package-json-from-dist@1.0.1:
-    resolution: {integrity: sha512-UEZIS3/by4OC8vL3P2dTXRETpebLI2NiI5vIrjaD/5UtrkFX/tNbwjTSRAGC/+7CAo2pIcBaRgWmcBBHcsaCIw==}
-
-  path-scurry@2.0.2:
-    resolution: {integrity: sha512-3O/iVVsJAPsOnpwWIeD+d6z/7PmqApyQePUtCndjatj/9I5LylHvt5qluFaBT3I5h3r1ejfR056c+FCv+NnNXg==}
-    engines: {node: 18 || 20 || >=22}
-
   prettier@3.4.2:
     resolution: {integrity: sha512-e9MewbtFo+Fevyuxn/4rrcDAaq0IYxPGLvObpQjiZBMAzB9IGmzlnG9RZy3FFas+eBMu2vA0CszMeduow5dIuQ==}
     engines: {node: '>=14'}
     hasBin: true
 
-  proxy-from-env@2.1.0:
-    resolution: {integrity: sha512-cJ+oHTW1VAEa8cJslgmUZrc+sjRKgAKl3Zyse6+PV38hZe/V6Z14TbCuXcan9F9ghlz4QrFr2c92TNF82UkYHA==}
-    engines: {node: '>=10'}
-
-  rimraf@6.1.3:
-    resolution: {integrity: sha512-LKg+Cr2ZF61fkcaK1UdkH2yEBBKnYjTyWzTJT6KNPcSPaiT7HSdhtMXQuN5wkTX0Xu72KQ1l8S42rlmexS2hSA==}
-    engines: {node: 20 || >=22}
-    hasBin: true
-
-  tunnel@0.0.6:
-    resolution: {integrity: sha512-1h/Lnq9yajKY2PEbBadPXj3VxsDDu844OnaAo52UVmIzIvwwtBPIuNvkjuzBlTWpfJyUbG3ez0KSBibQkj4ojg==}
-    engines: {node: '>=0.6.11 <=0.7.0 || >=0.7.3'}
-
   typescript@5.8.3:
     resolution: {integrity: sha512-p1diW6TqL9L07nNxvRMM7hMMw4c5XOo/1ibL4aAIGmSAt9slTE1Xgw5KWuof2uTOvCg9BY7ZRi+GaF+7sfgPeQ==}
     engines: {node: '>=14.17'}
@@ -369,9 +193,6 @@ packages:
   undici-types@7.16.0:
     resolution: {integrity: sha512-Zz+aZWSj8LE6zoxD+xrjh4VfkIG8Ya6LvYkZqtUQGJPZjYl53ypCaUwWqo7eI0x66KBGeRo+mlBEkMSeSZ38Nw==}
 
-  wcwidth@1.0.1:
-    resolution: {integrity: sha512-XHPEwS0q6TaxcvG85+8EYkbiCux2XtWG2mkc47Ng2A77BQu9+DqIOJldST4HgPkuea7dvKSj5VgX3P1d4rW8Tg==}
-
 snapshots:
 
   '@biomejs/biome@2.4.0':
@@ -409,179 +230,14 @@ snapshots:
   '@biomejs/cli-win32-x64@2.4.0':
     optional: true
 
-  '@ralph-orchestrator/ralph-cli@2.9.2':
-    dependencies:
-      axios: 1.15.2
-      axios-proxy-builder: 0.1.2
-      console.table: 0.10.0
-      detect-libc: 2.1.2
-      rimraf: 6.1.3
-    transitivePeerDependencies:
-      - debug
-
   '@types/node@24.12.2':
     dependencies:
       undici-types: 7.16.0
 
-  asynckit@0.4.0: {}
-
-  axios-proxy-builder@0.1.2:
-    dependencies:
-      tunnel: 0.0.6
-
-  axios@1.15.2:
-    dependencies:
-      follow-redirects: 1.16.0
-      form-data: 4.0.5
-      proxy-from-env: 2.1.0
-    transitivePeerDependencies:
-      - debug
-
-  balanced-match@4.0.4: {}
-
-  brace-expansion@5.0.5:
-    dependencies:
-      balanced-match: 4.0.4
-
-  call-bind-apply-helpers@1.0.2:
-    dependencies:
-      es-errors: 1.3.0
-      function-bind: 1.1.2
-
-  clone@1.0.4:
-    optional: true
-
-  combined-stream@1.0.8:
-    dependencies:
-      delayed-stream: 1.0.0
-
-  console.table@0.10.0:
-    dependencies:
-      easy-table: 1.1.0
-
-  defaults@1.0.4:
-    dependencies:
-      clone: 1.0.4
-    optional: true
-
-  delayed-stream@1.0.0: {}
-
-  detect-libc@2.1.2: {}
-
-  dunder-proto@1.0.1:
-    dependencies:
-      call-bind-apply-helpers: 1.0.2
-      es-errors: 1.3.0
-      gopd: 1.2.0
-
-  easy-table@1.1.0:
-    optionalDependencies:
-      wcwidth: 1.0.1
-
-  es-define-property@1.0.1: {}
-
-  es-errors@1.3.0: {}
-
-  es-object-atoms@1.1.1:
-    dependencies:
-      es-errors: 1.3.0
-
-  es-set-tostringtag@2.1.0:
-    dependencies:
-      es-errors: 1.3.0
-      get-intrinsic: 1.3.0
-      has-tostringtag: 1.0.2
-      hasown: 2.0.3
-
-  follow-redirects@1.16.0: {}
-
-  form-data@4.0.5:
-    dependencies:
-      asynckit: 0.4.0
-      combined-stream: 1.0.8
-      es-set-tostringtag: 2.1.0
-      hasown: 2.0.3
-      mime-types: 2.1.35
-
-  function-bind@1.1.2: {}
-
-  get-intrinsic@1.3.0:
-    dependencies:
-      call-bind-apply-helpers: 1.0.2
-      es-define-property: 1.0.1
-      es-errors: 1.3.0
-      es-object-atoms: 1.1.1
-      function-bind: 1.1.2
-      get-proto: 1.0.1
-      gopd: 1.2.0
-      has-symbols: 1.1.0
-      hasown: 2.0.3
-      math-intrinsics: 1.1.0
-
-  get-proto@1.0.1:
-    dependencies:
-      dunder-proto: 1.0.1
-      es-object-atoms: 1.1.1
-
-  glob@13.0.6:
-    dependencies:
-      minimatch: 10.2.5
-      minipass: 7.1.3
-      path-scurry: 2.0.2
-
-  gopd@1.2.0: {}
-
-  has-symbols@1.1.0: {}
-
-  has-tostringtag@1.0.2:
-    dependencies:
-      has-symbols: 1.1.0
-
-  hasown@2.0.3:
-    dependencies:
-      function-bind: 1.1.2
-
-  lru-cache@11.3.5: {}
-
   marked@17.0.5: {}
 
-  math-intrinsics@1.1.0: {}
-
-  mime-db@1.52.0: {}
-
-  mime-types@2.1.35:
-    dependencies:
-      mime-db: 1.52.0
-
-  minimatch@10.2.5:
-    dependencies:
-      brace-expansion: 5.0.5
-
-  minipass@7.1.3: {}
-
-  package-json-from-dist@1.0.1: {}
-
-  path-scurry@2.0.2:
-    dependencies:
-      lru-cache: 11.3.5
-      minipass: 7.1.3
-
   prettier@3.4.2: {}
 
-  proxy-from-env@2.1.0: {}
-
-  rimraf@6.1.3:
-    dependencies:
-      glob: 13.0.6
-      package-json-from-dist: 1.0.1
-
-  tunnel@0.0.6: {}
-
   typescript@5.8.3: {}
 
   undici-types@7.16.0: {}
-
-  wcwidth@1.0.1:
-    dependencies:
-      defaults: 1.0.4
-    optional: true
diff --git a/pnpm-workspace.yaml b/pnpm-workspace.yaml
index b4f0e0fd..e906dba1 100644
--- a/pnpm-workspace.yaml
+++ b/pnpm-workspace.yaml
@@ -2,12 +2,8 @@ packages:
   - "apps/*/*"
   - "packages/*"
 
-allowBuilds:
-  "@ralph-orchestrator/ralph-cli": true
-
 catalog:
   "@biomejs/biome": 2.4.0
-  "@ralph-orchestrator/ralph-cli": 2.9.2
   "@types/node": 24.12.2
   prettier: 3.4.2
   typescript: 5.8.3
diff --git a/ralph.yml b/ralph.yml
deleted file mode 100644
index 4ea993ed..00000000
--- a/ralph.yml
+++ /dev/null
@@ -1,35 +0,0 @@
-# Ralph Orchestrator Configuration
-# Generated by: ralph init --backend claude
-# Project: unic-agents-plugins
-# Docs: https://github.com/mikeyobrien/ralph-orchestrator
-
-cli:
-  backend: 'claude'
-
-event_loop:
-  prompt_file: 'PROMPT.md'
-  completion_promise: 'LOOP_COMPLETE'
-  max_iterations: 100 # 15 specs * generous retry budget
-  max_runtime_seconds: 28800 # 8 hours; abort if a spec stalls
-
-# ─────────────────────────────────────────────────────────────────────────────
-# Additional Configuration (uncomment to customize)
-# ─────────────────────────────────────────────────────────────────────────────
-
-# core:
-#   scratchpad: ".ralph/agent/scratchpad.md"
-#   specs_dir: ".ralph/specs/"
-
-# Custom hats for multi-agent workflows:
-# hats:
-#   builder:
-#     name: "Builder"
-#     triggers: ["build.task"]
-#     publishes: ["build.done", "build.blocked"]
-#
-#   reviewer:
-#     name: "Reviewer"
-#     triggers: ["review.request"]
-#     publishes: ["review.approved", "review.changes_requested"]
-
-# Create PROMPT.md with your task, then run: ralph run