v0.12.0

Minor Changes

• #24 ce544d8 Thanks @doidor! - Pick up four battle-tested patterns from the Epichan production harness:

  • security-reviewer role (principle 2) — an optional, read-only specialized reviewer on a
    different model family than the developer, with a Blocking/Warning/Informational severity model
    and an explicit APPROVE / REQUEST CHANGES verdict.
  • no-self-approve skill + a pre_merge gate (principles 9, 10) — agents on a shared bot identity
    must never approve their own PRs; independent approval is required.
  • resolve-conflicts skill (principles 4, 7) — rebase-first conflict repair that verifies
    mergeability before pushing.
  • address-review-comments skill (principles 4, 5) — reply to and resolve every review thread,
    then verify zero threads remain.

  Also enriches the wiki entry/index conventions with optional PR/commit provenance and good-vs-weak
  entry guidance. Bumps knowledgeVersion to 0.7.0 so agentrig update installs the new artifacts.

Patch Changes

• #23 fae869b Thanks @doidor! - Docs: document recreating the harness without the CLI by pointing an agent at the docsite's
  llms.txt. Adds a "Recreate the harness without the CLI (no lock-in)" walkthrough (with a
  ready-to-use prompt) to Getting started, updates the home-page "No lock-in" card and the README, and
  makes the canonical knowledge/templates/ the explicit source for exact, copy-paste artifact
  contents. No behavior or installed-artifact changes — documentation only.

v0.11.3

Patch Changes

• #21 14c7c4b Thanks @doidor! - Remove internal source-repo references from harness content. Drops the "Synthesized from …"
  provenance line in PRINCIPLES.md and the upstream attribution comments in the installed harness
  and eval templates (state-machine.yml, ORCHESTRATION.md, eval/axes.json, eval/score.mjs,
  scripts/repair-worktrees.sh). Bumps knowledgeVersion so agentrig update refreshes these files
  in consuming repos.

v0.11.2

Patch Changes

• #19 dd90c32 Thanks @doidor! - Docs: surface agentrig eval --scaffold (repo-tailored eval scenario generation) much earlier so
  users discover it right away. It now appears in the README quickstart and the "Evaluating the
  harness" intro callout, the docs landing-page feature card, the getting-started "Evaluate" step, and
  a top-of-page pointer in the evals guide — instead of being buried at the bottom of the eval docs.

v0.11.1

Patch Changes

• #17 900c39c Thanks @doidor! - Harness content quality fixes (consistency + dedup), applied to the canonical templates and the
  dogfooded harness:

  • Fix the invalid model_tier: high in every agent role → premium (matches each role's pinned
    model and the cheap/standard/premium vocabulary the state machine defines). Behavior is
    unchanged — only the label was wrong.
  • rules/coding-standards.md now declares priority: 3, as rules/README.md requires of every
    rule.
  • The harness-eval skill gains a triggers: block, for parity/discoverability with the other
    skills.
  • skill-authoring and skill-improver add Write to allowed-tools so they can actually create
    the files they describe.
  • De-duplicate the wiki docs: README.md owns policy (tiers + admission test) and points at
    _TEMPLATE.md for the entry format; index.md is now router/index-only.

  Internal (no user-facing behavior change): the agent prompt copy moved out of
  src/prompts/index.ts into src/prompts/templates/*.md, read at runtime — assembled output is
  byte-identical to before.

v0.11.0

Minor Changes

• #15 3d04ba8 Thanks @doidor! - Update workflow + doctor + new fix command — fewer surgical edits after agentrig update:

  • Marker populator. <!-- AGENTRIG:skills-inventory:start --> ... :end --> in AGENTS.md is
    now rewritten deterministically by both compile and update, so the block always reflects
    the installed skills. The audit check is upgraded to marker-populated — it now fails if the
    block is empty, has unfilled placeholders, or misses any skill under .agents/skills/.
  • agentrig fix + agentrig update --auto-fix. Deterministically repair A1 failures —
    restore broken .agentrig/**/*.yml from canonical, replace unknown model ids (e.g. the
    retired gpt-5) with the safe auto fallback. No agent or network needed.
  • Update validates before exiting. update now runs validateYaml + validateModelIds and
    refuses to leave a broken install in place — pass --auto-fix to self-heal or run
    agentrig fix afterward. Avoids the "audit was PART/FAIL but update exited 0" failure mode.
  • --diff drift classification. Preserved files are tagged 🔴 broken, 🟡 stale,
    🟢 enhancement, or ⚪ mixed so humans (and agents) can decide what's safe to auto-resolve.
  • Update enumerates added + preserved files instead of only printing counts; lists drifted
    files inline when --skip-agent is used.
  • Reconciliation history. .agentrig/state.json now records per-file decisions
    (reconciled[]); a future agentrig update skips re-prompting on files the user chose to
    keep local — unless canonical drifts past the hash that was recorded at decision time.
  • Doctor. Adds install-provenance detection (linked-checkout vs registry), an npm-latest
    comparison, and explicit validation findings so a broken YAML / unknown model id blocks
    doctor from returning 0.

  Knowledge bundle bumped to 0.6.0.

v0.10.0

Minor Changes

• #12 e75cb49 Thanks @doidor! - Rebuild of the eval setup — three honest layers, isolated producer/judge, statistical lift, repo-specific scaffolding.

  Replaces the previous "Harness Score" theater with three bounded layers, each making an explicit claim. See docs/evals.md for the full rubric.

  What's new

  • Layer A1 — install completeness and Layer A2 — quality probes: deterministic, no-model checks split into two scores. A2 catches content drift (parseable YAML/JSON, distinct model families, no unfilled {{PLACEHOLDER}}, every skill frontmatter, axes have issue codes) that A1's structural file-presence checks miss.
  • Layer B — fixture-based dynamic eval: agentrig eval --dynamic now seeds throwaway worktrees from scenarios/<id>/fixture/, runs the producer in isolation, applies a deterministic oracle (oracle.yml) for hard axes, then runs an independent judge (separate provider.startConversation(), different model family enforced) in a dedicated cwd that doesn't see the producer's worktree or reasoning trace.
  • agentrig eval --scaffold [--scaffold-count N] — agent generates repo-tailored scenarios using .agentrig/context.md + bundled generics as templates. Validated post-generation against the live axes.json registry.
  • Paired sign-test lift — score.mjs compare --baseline does a real binomial sign test over --n paired trials, with HELPS/HURTS/INCONCLUSIVE verdicts. Single-trial coin flips extinct.
  • Producer/judge family divergence enforced — score.mjs save rejects same-family pairs unless --allow-same-family is set (and records the override). Pre-flight validateModel() catches bad ids in ~2s instead of after 30s of producer burn.
  • Default models from the role roster — eval --dynamic reads producer from developer.yml and judge from reviewer.yml. The install audit's roles-distinct-families check guarantees the default pair clears divergence enforcement.
  • Bundled scenarios excluded by default — generic scenarios (add-small-feature, fix-failing-test, review-catches-bug) carry bundled: true and the dynamic eval excludes them by default so the eval reflects your repo's signal. --include-bundled opts in.
  • Per-axis veto + weight in axes.json v2 (back-compat with v1). Veto axes (correctness, gate_compliance, finding_correctness, blocking_decision) fail the scenario regardless of aggregate — cosmetic axes can never paper over a real regression.
  • log-gotcha skill + rewritten self-verify skill + inlined pre-handoff checklist in the producer prompt. The harness now nudges agents to surface red→green test evidence and commit wiki entries for non-obvious lessons.
  • Dogfood scenario (agentrig-init-on-empty-repo) — the ONE scenario that tests AgentRig the product itself, deterministically, with no model. Validates that init + compile + eval --static --min 80 keep working in a fresh empty repo.
  • CI workflow .github/workflows/agentrig-eval.yml runs Layer B nightly (harness vs baseline, n=5) and posts a tracking issue with the verdict table.
  • Judge calibration (score.mjs calibrate) — runs the judge over hand-labeled rubric instances in calibration/ and reports % within ±0.5 tier + signed bias. doctor flags any judge below the 80% agreement threshold.

  Bug fixes shipped along the way

  • Critical Rule #2 promoted wiki-logging to the top so agents see it before self-verify.
  • AGENTS.md template no longer has a bare https://github.com/) link.
  • process.exit()-after-console.log truncation on large --json output replaced with process.exitCode.
  • ActivityMonitor leak when provider.startConversation() throws (no more terminal-flood after a failed run).
  • Roster upgraded to claude-opus-4.8 (developer/judge) + gpt-5.5 (reviewer/triager) while preserving the developer-vs-reviewer family invariant.
  • Harness staging into eval worktrees no longer pollutes the producer's diff (now committed via --amend).
  • Report renderer leads with Summary: X/N PASS + per-FAIL evidence + "How to read this" closer; scoped to the current runId so old results never leak in.

  Knowledge migration

  Knowledge version bumped 0.3.3 → 0.5.0. Run agentrig update to refresh artifacts in an existing repo. The new log-gotcha skill, calibration starter set, and bundled-flag on the generic scenarios all propagate automatically.

  Test/audit baseline at release

  • 66/66 tests green
  • Install Completeness: 100% (33/33)
  • Quality Probes: 100% (6/6)
  • Dogfood scenario passes in seconds with no LLM

v0.9.0

Minor Changes

• #8 fc70643 Thanks @doidor! - agentrig init is now non-destructive by default. Previously, init unconditionally clobbered any existing AGENTS.md, .mcp.json, or hand-tailored rule/skill/wiki file at the destination — making it dangerous to adopt AgentRig in a repo that already had agent content (e.g. a curated AGENTS.md from a prior harness).

  Now init:

  • Preserves any existing destination file by default (file content is left verbatim, SHA-identical).
  • Reports preserved files in the install summary (preserved N existing file(s) — pass --force to overwrite: …).
  • Still installs all the canonical machinery around what you have (.agentrig/, skills, projection symlinks, scripts).
  • Compiles your existing AGENTS.md into every projected agent surface — so agentrig init becomes the natural "adopt AgentRig in this existing repo" entry point.

  Pass --force to opt into the previous overwriting behavior. agentrig init --dry-run now shows (new), (preserve existing), or (OVERWRITE) per file.

  agentrig update is unchanged — it still refreshes overwrite-policy machinery as before.

v0.8.0

Minor Changes

• #6 b85dced Thanks @doidor! - First public release as the scoped package @doidor/agentrig: a meta-harness CLI that installs
  best-practice agent harnesses into any repo and projects them to every agent surface (local +
  remote). Includes automated Changesets releases with npm provenance, Node >= 22, and CI/release
  status badges.

v0.7.0

Minor Changes

• #3 2e1de2f Thanks @doidor! - First public release as the scoped package @doidor/agentrig: a meta-harness CLI that installs
  best-practice agent harnesses into any repo and projects them to every agent surface (local +
  remote). Includes automated Changesets releases with npm provenance, Node >= 22, and CI/release
  status badges.

Patch Changes

• #3 2e1de2f Thanks @doidor! - agentrig compile now mirrors the entire AGENTS.md body into the projected .github/copilot-instructions.md and CLAUDE.md, instead of cherry-picking only the Critical Rules and What this repository is sections. Anything the user adds to AGENTS.md (custom sections, repo-specific guidance) now flows through to every downstream agent surface.

  Internally the projection now strips the H1 title, the <!-- AGENTRIG:…:start/end --> marker comments (which are AGENTS.md-internal update-protection), and any lines still carrying unfilled {{PLACEHOLDER}} template tokens.

v0.6.0

Minor Changes

• #1 dc7c740 Thanks @doidor! - First public release as the scoped package @doidor/agentrig: a meta-harness CLI that installs
  best-practice agent harnesses into any repo and projects them to every agent surface (local +
  remote). Includes automated Changesets releases with npm provenance, Node >= 22, and CI/release
  status badges.