feat(unic-pr-review): plugin scaffold + doctor command

.github/workflows/ci.yml

@@ -24,8 +24,9 @@ jobs:
     runs-on: ubuntu-latest
     outputs:
       pr-review: ${{ steps.filter.outputs.pr-review }}
+      unic-pr-review: ${{ steps.filter.outputs.unic-pr-review }}
       auto-format: ${{ steps.filter.outputs.auto-format }}
-      confluence-publish: ${{ steps.filter.outputs.confluence-publish }}
+      unic-confluence: ${{ steps.filter.outputs.unic-confluence }}
       release-tools: ${{ steps.filter.outputs.release-tools }}
       unic-archon-dlc: ${{ steps.filter.outputs.unic-archon-dlc }}
     steps:
@@ -36,6 +37,8 @@ jobs:
           filters: |
             pr-review:
               - 'apps/claude-code/pr-review/**'
+            unic-pr-review:
+              - 'apps/claude-code/unic-pr-review/**'
             auto-format:
               - 'apps/claude-code/auto-format/**'
             unic-confluence:
@@ -46,14 +49,15 @@ jobs:
               - 'apps/claude-code/unic-archon-dlc/**'
 
   test:
-    name: Test ${{ matrix.package }} / ${{ matrix.os }} / Node ${{ matrix.node }}
+    name: Test ${{ matrix.package.name }} / ${{ matrix.os }} / Node ${{ matrix.node }}
     needs: changes
     if: |
       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.pr-review == 'true'
+      needs.changes.outputs.pr-review == 'true' ||
+      needs.changes.outputs.unic-pr-review == 'true'
     strategy:
       fail-fast: false
       matrix:
@@ -70,6 +74,8 @@ jobs:
             changed: ${{ needs.changes.outputs.unic-archon-dlc }}
           - name: pr-review
             changed: ${{ needs.changes.outputs.pr-review }}
+          - name: unic-pr-review
+            changed: ${{ needs.changes.outputs.unic-pr-review }}
         exclude:
           - package:
               changed: 'false'

AGENTS.md

@@ -14,7 +14,8 @@ apps/
 │   ├── pr-review/
 │   ├── auto-format/
 │   ├── unic-confluence/
-│   └── unic-archon-dlc/
+│   ├── unic-archon-dlc/
+│   └── unic-pr-review/
 └── copilot/                  # GitHub Copilot plugins (future)
 packages/
 ├── biome-config/             # @unic/biome-config

CONTEXT-MAP.md

@@ -10,6 +10,7 @@
 - [pr-review](./apps/claude-code/pr-review/CONTEXT.md) — PR review command targeting Azure DevOps
 - [unic-confluence](./apps/claude-code/unic-confluence/CONTEXT.md) — Markdown-to-Confluence publishing command
 - [unic-archon-dlc](./apps/claude-code/unic-archon-dlc/CONTEXT.md) — Archon-powered AI development lifecycle DLC
+- [unic-pr-review](./apps/claude-code/unic-pr-review/CONTEXT.md) — AI-powered PR review with intent checking and Confidence-scored Findings
 
 ## Relationships
 

apps/claude-code/unic-pr-review/.claude-plugin/marketplace.json

@@ -0,0 +1,27 @@
+{
+  "metadata": {
+    "description": "AI-powered PR review with intent checking against Azure Boards and Jira, Confidence-scored Findings, and interactive Approval Loop",
+    "homepage": "https://github.com/unic/unic-agents-plugins"
+  },
+  "name": "unic-pr-review",
+  "owner": {
+    "name": "Unic AG"
+  },
+  "plugins": [
+    {
+      "author": {
+        "name": "Unic AG"
+      },
+      "category": "productivity",
+      "description": "AI-powered PR review with intent checking against Azure Boards and Jira work items, Confidence-scored Findings filtered by confidence threshold, and an interactive per-Finding Approval Loop.",
+      "displayName": "Unic PR Review",
+      "homepage": "https://github.com/unic/unic-agents-plugins",
+      "keywords": ["pr-review", "azure-devops", "jira", "confluence", "code-review", "unic"],
+      "license": "LGPL-3.0-or-later",
+      "name": "unic-pr-review",
+      "source": "./",
+      "tags": ["productivity", "code-review", "azure-devops"],
+      "version": "2.0.0"
+    }
+  ]
+}

apps/claude-code/unic-pr-review/.claude-plugin/plugin.json

@@ -0,0 +1,13 @@
+{
+  "author": {
+    "name": "Unic AG",
+    "url": "https://www.unic.com"
+  },
+  "commands": ["./commands/doctor.md"],
+  "description": "AI-powered PR review with intent checking against Azure Boards and Jira work items, Confidence-scored Findings, and an interactive Approval Loop.",
+  "homepage": "https://github.com/unic/unic-agents-plugins",
+  "keywords": ["pr-review", "azure-devops", "jira", "confluence", "code-review", "unic"],
+  "license": "LGPL-3.0-or-later",
+  "name": "unic-pr-review",
+  "version": "2.0.0"
+}

apps/claude-code/unic-pr-review/AGENTS.md

@@ -0,0 +1,83 @@
+# AGENTS.md — unic-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
+
+`unic-pr-review` is a Claude Code Plugin in the [`unic-agents-plugins`](../../../AGENTS.md) monorepo. It runs an AI-powered PR Review with intent checking against Azure Boards and Jira Work Items, Confidence-scored Findings, and an interactive Approval Loop. See [`CONTEXT.md`](CONTEXT.md) for the domain vocabulary (Plugin, Review, Finding, Confidence, Severity, Intent Brief, Intent Check, Bot Signature, Iteration, Approval Loop, Mode, Provider, Work Item, Notice).
+
+## 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
+- [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/) — ADRs 0001–0009 and a README listing them
+
+## Commands
+
+Plugin-specific pnpm scripts (run from this directory or with `pnpm --filter unic-pr-review <script>` from the repo root):
+
+```sh
+pnpm test                       # run the doctor unit-test suite via node:test
+pnpm typecheck                  # tsc --noEmit over the 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-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/         # Claude Code slash command definitions (.md files)
+scripts/          # Node.js implementation (.mjs, // @ts-check, no compilation)
+scripts/lib/      # Pure-function library modules
+tests/            # node:test suites (*.test.mjs)
+docs/adr/         # Plugin-scoped Architecture Decision Records
+```
+
+## Plugin doctrines
+
+Load-bearing invariants captured as ADRs. All nine must be understood before editing:
+
+- **ADR-0001** — Multi-source intent gathering with shared Atlassian credentials (`.unic-confluence.json` covers both Confluence and Jira)
+- **ADR-0002** — Confidence-scored Findings with explicit Severity thresholds (Critical 90-100, Important 80-89, Minor 60-79; drop below 60)
+- **ADR-0003** — Interactive Approval Loop as the default write path
+- **ADR-0004** — Hard-stop when intent sources are unreachable; empty intent is legitimate
+- **ADR-0005** — `az` CLI for Azure DevOps reads/writes; `node:https` (or global `fetch`) for Atlassian
+- **ADR-0006** — Iteration state lives in the PR's Bot Signature, not on disk; identity is cached from `az devops user show --user me` at startup
+- **ADR-0007** — Re-review uses a delta diff against the prior reviewed Revision
+- **ADR-0008** — Conditional sub-agent spawning by changed-file analysis
+- **ADR-0009** — Pre-PR Mode is a peer of the ADO Modes, not a special case
+
+ADR-0010 (Provider folder bundle layout) is planned but not yet landed; the [docs/adr/README.md](docs/adr/README.md) lists it as planned.
+
+## Conventions
+
+- SPDX header `// SPDX-License-Identifier: LGPL-3.0-or-later` on every `.mjs`
+- Copyright header `// Copyright © 2026 Unic`
+- `// @ts-check` at the top of every `.mjs`; types via JSDoc `@typedef` / `@param`
+- Tabs for indentation, single quotes, no semicolons (Biome)
+- No external runtime npm dependencies — every ADO read/write goes through `az`, every Atlassian call uses `node:https` or global `fetch`
+- Tests use `node:test` + `node:assert/strict`; predicates accept injectable executor / fetch parameters so they are unit-testable without mocking the module system
+
+## Clean-slate doctrine
+
+This Plugin takes **no code, no prompts, no fixtures, and no soft dependency** on `apps/claude-code/pr-review/`. That directory is treated as out of scope. Every module is re-derived from the PRD and ADRs.
+
+## Do not add
+
+- External runtime npm dependencies (zero is the bar — the doctor uses only built-ins)
+- A `LICENSE` file in this directory; per root AGENTS.md the maintainer manages LICENSE files manually
+- An `0010-*.md` ADR file — ADR-0010 is listed as planned in `docs/adr/README.md` but has not yet been merged
+- Any file outside `apps/claude-code/unic-pr-review/` except the one-line entry in the root `CONTEXT-MAP.md`

apps/claude-code/unic-pr-review/CHANGELOG.md

@@ -0,0 +1,34 @@
+# 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
+
+- Test runner now executes credentials.test.mjs in addition to doctor.test.mjs
+- doctor stays fully silent about Jira when jiraUrl is not configured (US-35)
+- realPing degrades gracefully when given a malformed URL instead of crashing the doctor
+- credentials loader distinguishes file-read errors from JSON-parse errors
+
+## [2.0.0] — 2026-05-28
+
+### Added
+
+- Plugin scaffold: `.claude-plugin/plugin.json`, `.claude-plugin/marketplace.json`, `package.json`, `tsconfig.json`
+- Domain vocabulary in `CONTEXT.md` (Plugin, Review, Finding, Confidence, Severity, Intent Brief, Intent Check, Bot Signature, Iteration, Approval Loop, Mode, Provider, Work Item, Notice)
+- `/unic-pr-review:doctor` slash command (`commands/doctor.md`)
+- `scripts/doctor.mjs`: six preflight checks — `az` CLI, `azure-devops` extension, `az devops` login, `az devops user show --user me` identity (for ADR-0006 caching), Confluence reachability, Jira reachability (silent when `jiraUrl` is unset per US 35)
+- `scripts/lib/credentials.mjs`: shared loader for `~/.unic-confluence.json` and `~/.unic-azure.json` with env-var overrides
+- `tests/doctor.test.mjs`: unit tests for all six predicates with stubbed executors and fetchers

apps/claude-code/unic-pr-review/CLAUDE.md

@@ -0,0 +1 @@
+AGENTS.md

apps/claude-code/unic-pr-review/CONTEXT.md

@@ -0,0 +1,83 @@
+# unic-pr-review
+
+A Claude Code Plugin that runs AI-powered PR Reviews with intent checking against Azure Boards / Jira Work Items, emits Confidence-scored Findings, and walks the reviewer through an interactive Approval Loop before any write to the PR.
+
+## Language
+
+**Plugin**:
+The `unic-pr-review` Claude Code Plugin itself; the subject of this context. Ships at `apps/claude-code/unic-pr-review/`.
+_Avoid_: tool, app, integration
+
+**Review**:
+A complete run of the Plugin against either an Azure DevOps Pull Request (ADO Mode) or a local feature branch (Pre-PR Mode) — from intent gathering through aspect fan-out and Approval Loop to the optional write-back.
+_Avoid_: scan, audit, check
+
+**Finding**:
+A single code observation emitted by a Review Aspect agent, carrying `{ severity, confidence, filePath, startLine, endLine, title, body, suggestion? }`.
+_Avoid_: comment, issue, note
+
+**Confidence**:
+A 0-100 integer score attached to every Finding. Findings below 60 are dropped before the reviewer ever sees them (ADR-0002).
+_Avoid_: score, weight, certainty
+
+**Severity**:
+The bucket derived from Confidence — Critical (90-100), Important (80-89), Minor (60-79).
+_Avoid_: priority, urgency, level
+
+**Intent Brief**:
+The structured synthesis of Work Item intent (description, Acceptance Criteria or Repro Steps, plus Confluence context) produced by the Intent Checker and passed verbatim as a preamble to every Review Aspect agent.
+_Avoid_: context, summary, brief
+
+**Intent Check**:
+The per-Acceptance-Criterion verdict block (`addressed` / `partially addressed` / `unaddressed`) rendered at the top of the Review Summary. Omitted when no Work Items are linked (ADR-0004).
+_Avoid_: gap analysis, intent diff, requirements check
+
+**Bot Signature**:
+The load-bearing footer wording `🤖 Reviewed by Claude Code — Iteration N`, owned solely by `scripts/lib/signature.mjs`. Used for Re-review detection and Iteration counting (ADR-0006).
+_Avoid_: marker, tag, watermark
+
+**Iteration**:
+The review count for a single PR, stored in the Bot Signature, incremented on each Re-review.
+_Avoid_: pass, round, run
+
+**Approval Loop**:
+The interactive per-Finding `accept` / `edit` / `skip` flow entered when `--post` is given. Resumable state lives under `<cwd>/.unic-pr-review/<key>/state.json` (ADR-0003).
+_Avoid_: confirmation, prompt, gate
+
+**Mode**:
+One of `pre-pr`, `first-review`, `re-review`, `first-review-fallback`. Selected at runtime by URL presence and Bot Signature detection (ADR-0009).
+_Avoid_: state, phase, kind
+
+**Provider**:
+A folder bundle at `providers/<name>/` implementing the Source Platform contract (`parsePrUrl`, `agents.{fetcher, writer}`, `discoverWorkItems`). v2 ships `providers/azure_devops/`; later releases may add GitHub or GitLab.
+_Avoid_: adapter, backend, driver
+
+**Work Item**:
+A normalised task record `{ id, type, url, raw }` returned by `provider.discoverWorkItems(prMetadata)` — an Azure Boards Work Item, a Jira issue, or a manually pasted URL in Pre-PR Mode.
+_Avoid_: ticket, story, task
+
+**Notice**:
+A prose block at the top of the Review Summary (before the Intent Check) used for warnings, escalations, and fallback conditions.
+_Avoid_: alert, banner, warning
+
+## Relationships
+
+- A **Review** runs in exactly one **Mode**, decided at startup
+- A **Mode** is selected by the active **Provider**'s URL parsing plus **Bot Signature** detection
+- The Intent Checker turns one or more **Work Items** into a single **Intent Brief**
+- Every Review Aspect agent receives the **Intent Brief** and emits zero or more **Findings**
+- Each **Finding** carries a **Confidence** score that determines its **Severity** bucket
+- The **Approval Loop** mediates between **Findings** and PR write-back
+- The **Bot Signature** records the **Iteration** and lets the next Review detect prior runs
+- A **Notice** can appear above the **Intent Check** when something needs the reviewer's attention
+
+## Example dialogue
+
+> **Dev:** "What happens if a PR links no Work Items at all?"
+> **Domain expert:** "The Intent Check block is just omitted from the Review Summary. Per ADR-0004 the absence of intent is a legitimate state — we only hard-stop if a referenced source is unreachable."
+
+> **Dev:** "Why does `doctor` verify `az devops user show --user me` and not just `az devops login`?"
+> **Domain expert:** "ADR-0006 stores Iteration state in the PR's own Bot Signature. To detect our own prior comments at Review time the Plugin caches the authenticated ADO user id at startup. If `user show` can't resolve the identity, Re-review detection breaks — so we check it during doctor instead of failing mid-Review."
+
+> **Dev:** "A Finding came back with Confidence 55. Why didn't I see it in the Summary?"
+> **Domain expert:** "Anything below 60 is filtered out per ADR-0002. That's the noise floor — Minor starts at 60."