wicked-garden 12.27.0

package/CHANGELOG.md

@@ -0,0 +1,712 @@
+# Changelog
+
+## [Unreleased]
+
+## [12.27.0] — 2026-07-07
+
+### Added
+- **npm package**: wicked-garden is now published to npm. `npx wicked-garden install` copies the plugin to `~/.claude/plugins/wicked-garden/` — no git required. `npx wicked-garden status` shows the installed version.
+- **`install.mjs`**: standalone installer with `install`, `update`, `status`, `--version` subcommands. Runs `uv sync` for Python deps if `uv` is available; otherwise defers to `/wicked-garden:setup`.
+- **wicked-loom absorbed**: peer-resolution engine (`manifest`, `resolve`, `compose`, `gate`) now ships in `scripts/loom/` — no separate `wicked-loom` package needed. `_loom.py` dispatches internally; `WICKED_LOOM_BIN` env override still wins.
+- **`/wicked-garden:platform:peer-health`** command: user-facing replacement for `npx wicked-loom doctor` (`--peer`, `--strict`, `--json` flags).
+
+### Changed
+- `plugin.json`: removed `wicked_loom_version` peer requirement.
+- `commands/setup.md`: §2.7b uses internal Python import check; corrected `WICKED_LOOM_CUTOVER=off` doc (FAIL CLOSED, no fallback).
+- `commands/install.md`: removed external `npm i -g wicked-loom` step.
+
+## [12.26.0] — 2026-07-07
+
+### Added
+- **`POST /hooks`** — register a bus event hook (`event_pattern`, `command`). Returns 201 with id and timestamps. Hooks are persisted in the SQLite store.
+- **`GET /hooks`** — list all registered hooks.
+- **`DELETE /hooks/{id}`** — deregister a hook by id; returns 404 if not found.
+- **`POST /council/vote`** — cast a vote on an active council session (`session_id`, `voter_id`, `vote`); appends to the votes JSON array; returns 404 if session not found.
+- **Idempotency key on `wicked.garden.council.voted`** — key format: `garden:wicked.garden.council.voted:{session_id}:{sha256(session_id)[0:16]}:0` per DEC-00010.
+
+### Fixed
+- **Consumer self-consumption** — daemon no longer re-consumes its own `wicked.garden.council.voted` events (belt-and-suspenders: prefix filter + explicit early-return guard).
+- **Thread-safe DB writes** — `_store_event()` and `_save_cursor()` in consumer.py now wrap all writes in `get_write_lock()` (threading.Lock).
+- **Nested HTTP error format** — all `server.py` error responses now return `{"error": {"code": "...", "message": "..."}}` via `_err()` helper (18 endpoints updated).
+- **`wicked_testing_version` pin** — bumped `^0.6.0` → `^0.7.0`; caret probe locks 0.x pins to minor, so wicked-testing 0.7.x was out-of-range and crew commands were blocked.
+
+### Changed
+- **`context: fork` on all 23 agents** — added to every `.md` in `agents/` with fully-qualified `wicked-garden:{domain}:{name}` names, enabling cross-CLI skill discovery.
+- **`context: fork` on 12 skills** — archetype, classify, deliberate, ground, integration-discovery, multi-model, persona, runtime-exec, smaht, swarm, wickedizer, worktrees; name prefix fixed to `wicked-garden:` on the 10 that had bare names.
+
+## [12.25.1] - 2026-06-17
+
+### Bug Fixes
+- fix(peers): `wicked_testing_version` pin `^0.4.0` → `^0.6.0` — the caret probe locks 0.x pins to the minor, so the current wicked-testing **0.6.x** line was out-of-range against `^0.4.0` and crew commands were blocked. Widen the enforced peer floor (mirrors the earlier `^0.3.0` → `^0.4.0` fix). Sibling floors `wicked_vault ^0.4.0` / `wicked_bus ^2.0.0` still cover 0.4.3 / 2.2.3; the declarative `wicked_brain` / `wicked_loom` floors are unchanged.
+
+## [12.21.0] - 2026-06-12
+
+### Changed
+- Evidence-based agent-surface reduction: per-agent blinded lift eval (tests/agents/EVAL_RESULTS.md) found 13 generic 'be a senior X' agents + the product-management skill produce lift=0 vs the base model — removed them (36→23 agents); kept solution-architect/ux-designer/value-strategist (lift+1). References degraded to kept lenses/inline; resolver falls back for unknown roles. Also fixed 5 dead script pointers (doc rot).
+
+## [12.20.0] - 2026-06-12
+
+### Changed
+- refactor(persona): Removed 6 proven-redundant generic built-in personas (lift=0 per tests/persona/EVAL_RESULTS.md); kept platform/qe/agentic as methodology exemplars; generic role names now fall back gracefully; persona:define is the product. Net surface reduction.
+
+## [12.19.1] - 2026-06-12
+
+### Bug Fixes
+- fix(qe): the trust-spine `ProveEndToEndTests` now **RUN and GATE in CI** instead of skipping — they execute against any **resolvable** vault (PATH / `npm i -g` / `node_modules`, mirroring `vault_gate.resolve_vault`), CI installs the published peers, and `WICKED_REQUIRE_E2E=1` (set in `.github/workflows/test.yml`) makes a SKIP of these tests a hard build failure. The PASS / REJECT / independent-attestation path is now actually covered (no more GREEN-BUT-HOLLOW).
+- fix(persona): `delete_persona(name)` resolves name → UUID id before deleting — custom personas are keyed by an auto-generated id in the DomainStore, so delete-by-name silently failed and left the persona in place. Adds a define → delete round-trip test.
+
+### Chores
+- chore(qe): propagate the explicit `--actor "${WICKED_VAULT_ACTOR:-garden-prove}"` to every hard-gate record-then-attest playbook (`skills/archetype/refs/{review,incident,migrate,specify}.md`) so wicked-vault **>= 0.4.0** does not refuse an independent attestation for a weak/ambient worker identity; peer floor `wicked_vault_version` pinned `^0.4.0`. Documented in `.claude/CLAUDE.md`.
+- chore(persona): ran the persona-lift eval (blinded, independent grader) and recorded the result honestly in `tests/persona/EVAL_RESULTS.md` — **lift = 0 on all 3 cases** vs a strong base model. The built-in methodology personas are illustrative/redundant for these textbook cases; the `persona:define` mechanism (inject HOUSE methodology the base model cannot know) is the actual product. Recorded, not asserted.
+- chore: bump 12.19.0 → 12.19.1.
+
+## [12.19.0] - 2026-06-12
+
+### Features
+- feat(persona): triage methodology vs generic personas — sharpen the methodology personas, add a `not_focus` scope guard + GOOD-pattern template to `persona:define`, and demote the generic tail by presentation. Adds `tests/persona/` (deterministic lift suite + eval-case schema).
+- feat(help): regenerate `/help` from the live command tree (drop retired crew/delivery; add prove/archetype/persona/compile/intent/where-am-i) with a `tests/test_help_command_tree.py` lint that fails if help drifts from `commands/`.
+- feat(routers): add requirements/review sprawl routers and intent-revealing descriptions for smaht/wickedizer/ground.
+
+### Bug Fixes
+- fix(qe): `scripts/qe/prove.py` records evidence under an explicit `--actor` (`WICKED_VAULT_ACTOR` → default `garden-prove`) so the hardened **wicked-vault (>= 0.4.0)** trusts an independent attestation and `prove --with-attestations` can reach PASS — without weakening independence (a self-attest by the doer is still refused). `tests/qe/test_prove.py` updated in lockstep. (#945)
+- fix(qe): the trust-spine `ProveEndToEndTests` now RUN against any **resolvable** vault (PATH / `npm i -g` / `node_modules`), not only a sibling checkout or `WICKED_VAULT_BIN` — mirrors `vault_gate.resolve_vault`. CI sets `WICKED_REQUIRE_E2E=1` so a SKIP of these tests is a hard failure (no more GREEN-BUT-HOLLOW).
+- fix(qe): propagate the wicked-vault >= 0.4.0 explicit-`--actor` requirement to every record-then-attest path — the hard-gate playbooks (`skills/archetype/refs/{review,incident,migrate}.md`, plus `specify.md`) now record with `--actor "${WICKED_VAULT_ACTOR:-garden-prove}"` so an independent `attest` is not refused for weak/ambient worker identity. The compiler-emitted integrity gate (record + cross-check, no attest) is intentionally unchanged. Peer floor `wicked_vault_version` pinned `^0.3.0` → `^0.4.0`.
+- fix(persona): `delete_persona(name)` resolves name → UUID id before deleting — custom personas are keyed by an auto-generated id in the DomainStore, so delete-by-name silently failed and left the persona in place. Adds a define → delete round-trip test.
+
+### Chores
+- chore: remove dead `daemon/council.py` pointer; bump 12.18.4 → 12.19.0.
+
+## [12.18.4] - 2026-06-10
+
+### Bug Fixes
+- fix(sentinel): `.githooks/pre-push` resolved its interpreter with `python3 … || python …`, so a deliberate sentinel rejection (non-zero exit) re-ran the script via bare `python` — on machines without one, the block reason was buried under `python: command not found`. The hook now resolves the interpreter once (`command -v python3 || command -v python`) and `exec`s it, preserving the sentinel's exit code; no python at all fails closed with a clear message. (#932)
+
+## [12.18.3] - 2026-06-10
+
+### Bug Fixes
+- fix(peers): `wicked_testing_version` pin `^0.3.0` → `^0.4.0` — the caret probe locks 0.x pins to the minor, so wicked-testing 0.4.0 (current) was rejected and crew commands were blocked. `wicked_brain_version` `^0.14.0` → `^0.15.0` to match the brain codegraph floor garden's ADR 0004 integration relies on (declarative only — no probe enforces it).
+- fix(site): refresh marketing-site content against repo reality — wicked-testing "41 specialist agents" → 40 and "6 CLIs" badge → 5 (0.4.0 roster trim + Copilot removal), family hero stat 81 → 80 agents; bench install script for wicked-bus corrected from a nonexistent Claude-plugin install to `npm i wicked-bus` + `npx wicked-bus-install`.
+
+### Features
+- feat(site): the brain card and memory-bed tour stop now cover the wicked-brain code graph (v0.15.0) — blast-radius, callers, and lineage on demand.
+
+### Chores
+- chore: remove dead wicked-testing tier-coupling (gate-policy path retired at v11); ci: bump GitHub Actions to Node 24-ready majors; docs: link the GitHub Pages site at the top of the README; chore(deps): bump astro (#926).
+
+## [12.18.2] - 2026-06-10
+
+### Bug Fixes
+- fix(brain): auto-start + port-resolution fix — no more false "server not running" notices. (Backfilled entry; released as the version-bump commit `36894f5`.)
+
+## [12.18.1] - 2026-06-10
+
+### Chores
+- chore: remove the superseded in-garden code-relationship graph (now in wicked-brain, ADR 0004) — deletes `scripts/_codegraph.py`, `scripts/codegraph/*` (5 injected-edge extractors), and `tests/codegraph/`; zero live consumers remained. Rewires `search:hotspots` freshness onto `search:index`; marks ADR 0001 superseded. wicked-patch's `codegraph_db.py` adapter is unchanged. No behavior change.
+
+## [12.18.0] - 2026-06-10
+
+### Features
+- feat(codegraph): move the code-relationship graph to wicked-brain (ADR 0004) — blast-radius/lineage/callers are now brain's `graph-*` actions + the `wicked-brain:graph` skill, backed by a codegraph static graph + injected edges (bus/dispatch/capability). `search:blast-radius`/`lineage`/`index` rewired to thin wrappers over brain.
+- feat(codegraph): garden ships its proprietary archetype→playbook edges via the drop-in extractor `.codegraph-extractors/archetype.mjs`, discovered + run by brain's extractor registry (real-repo verified on this repo).
+- docs: ADR 0004 records the inversion of ADR 0001; AGENTS.md is now a thin pointer to `.claude/CLAUDE.md`.
+
+### Bug Fixes
+- fix(sentinel): claim-gate the answer tier at the Stop boundary — it now fires only on a real done/passing/shipped claim with no verdict for HEAD (debounced per sha), instead of on any Bash call during planning. The buggy PostToolUse ref-watch is removed; the pre-push block tier is unchanged.
+
+## [12.17.0] - 2026-06-09
+
+### Features
+- feat: the claim sentinel — state-transition gates without command lists (ADR-worthy core) (6ff25f5)
+
+### Bug Fixes
+- fix(search): index.md — codegraph init required before first index (found on real-repo test) (013d5a6)
+
+### Chores
+- Merge pull request #925 from mikeparcewski/fix/index-init-and-release-prep (5ab3d85)
+- Merge pull request #924 from mikeparcewski/feat/claim-sentinel (589650e)
+
+## [12.16.0] - 2026-06-09
+
+### Features
+- feat: integrate wicked-understanding as the repo-playbooks opt-in layer (full: layer + wired in) (cade19d)
+
+### Chores
+- Merge pull request #923 from mikeparcewski/feat/integrate-wicked-understanding (99d56dc)
+
+## [12.15.0] - 2026-06-09
+
+### Features
+- feat: two required peers (the gate), three opt-in layers + curated-toolkit reframe (ADR 0003) (ad7f312)
+
+### Chores
+- Merge pull request #922 from mikeparcewski/feat/toolkit-optin-peers (14107cf)
+
+## [12.14.0] - 2026-06-09
+
+### Bug Fixes
+- fix(search): hotspots query from codegraph graph instead of brain curl (a7cf0a0)
+
+### Documentation
+- docs: tighten the reframe — concise + fun (0e609e2)
+- docs: reframe wicked-garden as gap-filler for coding-agent harnesses (not 'an SDLC') (931a2a8)
+
+### Chores
+- Merge pull request #921 from mikeparcewski/docs/reframe-gap-filler (4843d73)
+- Merge pull request #920 from mikeparcewski/fix/hotspots-codegraph (c2e821e)
+
+## [12.13.0] - 2026-06-09
+
+### Bug Fixes
+- fix(search): dynamic brain port + cut 5 broken admin wrappers + finish delivery retire (27b64df)
+- fix(ci): find_orphan_agents excludes relative to repo root, not absolute path (92f3b15)
+- fix(search): make blast-radius/lineage injected layer actually work (#916) (305ff3e)
+
+### Refactoring
+- refactor: retire delivery domain and jam post-hoc viewer endpoints (f087c2b)
+
+### Chores
+- Merge pull request #919 from mikeparcewski/cleanup/final-search-trim (149331a)
+- Merge pull request #918 from mikeparcewski/cleanup/retire-delivery-jam-viewers (e953ef4)
+- Merge pull request #917 from mikeparcewski/fix/codeintel-injected-edges-916 (bcd6444)
+
+## [12.12.0] - 2026-06-09
+
+### Refactoring
+- refactor(agentic,jam,persona,smaht): collapse rubric-wrapper commands to inline skill-refs (ADR 0002 pattern) (7eeb8e7)
+- refactor(platform): collapse 8 dispatch-only commands to inline rubrics (0d86a74)
+- refactor(data,delivery): collapse 7 rubric-wrapper commands to inline skill-refs (ADR 0002 pattern) (8ce6f2b)
+- refactor(engineering): collapse 5 rubric-wrapper commands to inline skill-refs (ADR 0002 pattern) (d26cec6)
+- refactor(product): collapse 12 rubric-wrapper commands to inline skill-refs (cleanup ADR 0002 pattern) (264e619)
+
+### Chores
+- Merge pull request #915 from mikeparcewski/cleanup/agent-purge-and-sync (8ab295b)
+- cleanup: purge 3 orphaned agents + re-derive registry + anti-drift tools (ADR 0002 phase 6) (932b4b8)
+- Merge pull request #914 from mikeparcewski/cleanup/agentic-jam-persona-smaht-collapse (61aa0c8)
+- Merge pull request #913 from mikeparcewski/cleanup/data-delivery-collapse (9c5548d)
+- Merge pull request #912 from mikeparcewski/cleanup/platform-collapse (20621c2)
+- Merge pull request #911 from mikeparcewski/cleanup/engineering-collapse (a9cf769)
+- Merge pull request #910 from mikeparcewski/cleanup/product-collapse (b785732)
+
+## [12.11.0] - 2026-06-09
+
+### Features
+- feat(codegraph): add dispatch + capability injected-edge extractors; wire blast-radius/lineage to graph (95c5205)
+- feat(patch): revive wicked-patch via codegraph — adapter builds its --db from the code-graph (3c904f1)
+
+### Chores
+- Merge pull request #909 from mikeparcewski/feat/codegraph-injected-extractors (37b7be2)
+- Merge pull request #908 from mikeparcewski/feat/patch-codegraph-db (49686b8)
+
+## [12.10.1] - 2026-06-09
+
+### Chores
+- Merge pull request #907 from mikeparcewski/chore/remove-dead-commands (a10f2b0)
+- chore: remove dead commands (crew:archive, delivery:process-health) + major-cleanup plan (ADR 0002) (504883f)
+
+## [12.10.0] - 2026-06-09
+
+### Features
+- feat(codegraph): adopt codegraph as the static graph engine + injected-edge extractor (ADR 0001) (88cb6c1)
+- feat: wire real security scanners + make archetype steering actionable (6e1ae61)
+
+### Bug Fixes
+- fix: repair 9 confirmed-broken wicked-garden command surfaces (678f972)
+
+### Chores
+- Merge pull request #906 from mikeparcewski/fix/broken-commands-batch (254677a)
+- Merge pull request #905 from mikeparcewski/feat/security-scanners-and-actionable-steering (1c1329f)
+- Merge pull request #904 from mikeparcewski/feat/code-relationship-graph (805b5a3)
+
+## [12.9.0] - 2026-06-09
+
+### Features
+- feat(qe): real SEMANTIC spec-to-code review — the agent the script always referenced but never existed (631d1b5)
+
+### Chores
+- Merge pull request #903 from mikeparcewski/feat/semantic-reviewer-agent (215e536)
+
+## [12.8.0] - 2026-06-09
+
+### Features
+- feat(qe): prove.py validates OUTPUTS, not just exit codes (regex_match/not_contains/jq_pred) (d7d3dd2)
+
+### Chores
+- Merge pull request #902 from mikeparcewski/feat/prove-output-verifiers (213da9b)
+
+## [12.7.1] - 2026-06-09
+
+### Bug Fixes
+- fix(qe): prove.py --with-attestations actually enforces (was a vacuous hard gate in v12.7.0) (7d0f85b)
+
+### Chores
+- Merge pull request #901 from mikeparcewski/fix/prove-attestation-enforcement (b14cd27)
+
+## [12.7.0] - 2026-06-09
+
+### Features
+- feat(qe): prove.py supports --with-attestations (universal across hard gates) (3df9d57)
+
+### Documentation
+- docs(archetypes): recommend prove.py frictionless gate in all 7 gating playbooks (08e2c05)
+
+### Chores
+- Merge pull request #900 from mikeparcewski/feat/prove-universal (c80bd6a)
+
+## [12.6.0] - 2026-06-09
+
+### Features
+- feat(qe): prove.py — the one-line re-derivation verb (gate without the ritual) (37d586e)
+
+### Documentation
+- docs(build): recommend prove.py as the frictionless gate path (c3132be)
+
+### Chores
+- Merge pull request #899 from mikeparcewski/feat/prove-verb (43d98c5)
+
+## [12.5.2] - 2026-06-09
+
+### Tests
+- test(e2e): fix CI environment assumption — own $HOME, test both configured + unconfigured (9398b08)
+- test(e2e): real-surface harness — gate CLI, bootstrap, routing, resilience (Tiers A+C) (f83cf90)
+
+### Chores
+- Merge pull request #898 from mikeparcewski/test/e2e-three-tier-harness (685c007)
+- ci: install node + wicked peers so tests/e2e gate+resilience RUN (not skip) (8e59ba7)
+
+## [12.5.1] - 2026-06-09
+
+### Bug Fixes
+- fix: vault_gate CLI imports _loom (gate was failing closed via its own documented commands) (d0ce3eb)
+
+## [12.5.0] - 2026-06-09
+
+### Features
+- feat(capabilities): register apm/tracing/logging/telemetry; lock agent caps to the registry (cde4bda)
+
+### Bug Fixes
+- fix(detector): match 'roll out' (spaced) for the ship archetype (54048e0)
+- fix(phase-manager): align hard-gate map to the catalog (drop specify/decide over-reach) (9c042f1)
+- fix: loom gate re-derives against project_dir, not process cwd (#891 regression) (ea8a3f8)
+
+### Documentation
+- docs: reconcile with the v12 loom cutover (loom is the gate; five required peers) (f66010d)
+- docs(spec): record loom migration terminal outcome (§6 remainder stays in garden) (#895) (eb228d0)
+
+## [12.4.0] - 2026-06-08
+
+### Features
+- feat: flow surface — loom-authoritative hard-gate park decision (fail-closed floor) (#893) (a4bab29)
+
+## [12.3.0] - 2026-06-08
+
+### Features
+- feat: wicked-loom contract — loom is the sole gate/resolve path (delete in-process re-derivation) (#891) (7e4492a)
+- feat: wicked-loom cutover — garden shells to loom resolve/gate/flow (additive, flag-gated) (#890) (ae928a7)
+- feat(#878): worktree cwd-leak guard on Bash PreToolUse (#888) (859d2a0)
+
+### Documentation
+- docs: wicked-loom north-star spec (extract the orchestration runtime) (#889) (16c3e70)
+
+## [12.2.0] - 2026-06-01
+
+### Features
+- skill(engineering): large-scale-migration — the map→transform→gate leverage pattern for cross-cutting migrations / refactors / renames / dialect-ports (21b1842)
+
+### Bug Fixes
+- fix(#879): clean up the two rotted test files outside tests/ (#885) (7db380d)
+
+This project follows [Semantic Versioning](https://semver.org/).
+
+The version line restarted at **v11.0.0** with the v11 work-shape archetype
+reframe. The pre-v11 changelog (versions 0.x through 10.x) is preserved in
+the git history but is not maintained as living documentation — those
+releases describe a prior architecture (universal pipeline + rigor tier
+dial) that was replaced wholesale.
+
+---
+
+## [12.1.0] — 2026-05-25
+
+**All four siblings are now required, and the work mode is visible on screen.**
+
+### Required peers (all four)
+
+- **wicked-brain and wicked-bus join wicked-testing and wicked-vault as required peers.** `/wicked-garden:setup` now verifies all four and blocks without them; the SessionStart hook warns. `plugin.json` pins all four (`wicked_brain_version` `^0.14.0`, `wicked_bus_version` `^2.0.0`). The stance is **required at install, resilient at runtime**: a transient outage (brain server down, bus unavailable) degrades gracefully and never bricks a session — it never means a gate treats missing evidence as a pass (that path fails closed). See [`docs/required-peers.md`](docs/required-peers.md).
+
+### Features
+
+- **A work-mode status line** (`scripts/statusline.py`, opt-in via `settings.json` — see getting-started or `/wicked-garden:setup` step 6.6). Renders the live archetype · intent · phase · gate verdict at the bottom of the screen — `🌱 wg │ build·migrate │ intent: feature │ phase: implement │ ⚖ PASS`. Reads existing session state only (one read), fail-soft (degrades to `🌱 wg │ idle`, never blocks a render).
+
+### Documentation
+
+- **Rethought, not patched.** The docs were reoriented from the deleted v6 crew model to v12. README restructured around what/why/how (leads with the re-derived-evidence trust pitch, not the pipeline rationale); ETHOS rewritten (required-infrastructure / resilient-at-runtime; the nine archetypes named; the *"done is not claimed; done is re-derived"* slogan); `docs/domains.md` reframed off the dead orchestrator; new [`docs/required-peers.md`](docs/required-peers.md) + [`docs/compiler.md`](docs/compiler.md); `docs/v11/archetypes.md` corrected (hard gates are enforced at runtime). Stale `crew:*` references removed from setup.
+
+### Notes
+
+- The status line's two bugs (wrong field name, mismatched sanitization) were caught by independent review, not by the tests that shipped CI-green — those were tautological. The fix added a guard test cross-checking the renderer's fields against the real `SessionState` schema. *Honest verdicts beat green dashboards.*
+
+---
+
+## [12.0.0] — 2026-05-25
+
+**"Done" is re-derived, not asserted — and the garden can compile that guarantee onto any repo.**
+
+Every gating archetype's produces-gate stops trusting a self-asserted "done"
+and re-derives it through [wicked-vault](https://www.npmjs.com/package/wicked-vault),
+a new **required** peer. A new compiler emits that same re-derivation as a
+self-contained harness into any repo.
+
+### Breaking changes
+
+- **wicked-vault is now a required peer** (`^0.3.0`; sibling to wicked-bus /
+  wicked-brain / wicked-testing). `/wicked-garden:setup` verifies it and blocks
+  without it, and the SessionStart hook warns. Install with
+  `npx wicked-vault-install`. Offline/dev kill-switch: `WICKED_VAULT_BIN=""`
+  (see CONTRIBUTING.md).
+
+### Features
+
+- **Vault-backed gates for all gating archetypes** (`scripts/qe/vault_gate.py`).
+  The gate runs `wicked-vault cross-check` — re-hashing the recorded evidence
+  and re-running its verifier — instead of `evidence_tracker`'s
+  satisfied-when-claimed model. A claimed-but-false "tests pass" is REJECTED; a
+  missing vault **fails closed** rather than passing on a self-assertion. Hard
+  gates (review / incident / migrate) require an *independent* attestation —
+  the evaluator is not the agent that did the work.
+- **The compiler emit stage + `/wicked-garden:compile`** (`scripts/compiler/`).
+  Detects a repo's test/lint/build commands and emits a self-contained,
+  vault-backed gate into `<repo>/.wicked/` (contract + a stdlib-only `gate.py`
+  + a claims-vs-evidence lint + README) that runs with **no wicked-garden
+  runtime present** (resolves the vault via `npx`). Optionally installs the
+  triggers that fire it — a git pre-push hook and a GitHub Actions workflow
+  whose toolchain setup is derived from the detected ecosystem. Multi-claim
+  contracts (test/lint/build), each pinned to its own verifier.
+
+### Notes
+
+- 422 tests; proven on an unseen repo (memos) and self-hosted on wicked-garden
+  (which surfaced and fixed a detector scoping bug).
+- Follow-up: #879 — two rotted test files CI never collected, found while
+  self-hosting the compiler.
+
+---
+
+## [11.1.3] — 2026-05-08
+
+**Council command + 4 sibling commands had broken refs the validator missed.**
+
+The validator at v11.1.2 caught script-path drift in YAML/JSON fields
+and `${CLAUDE_PLUGIN_ROOT}/...` strings, but missed two reference
+shapes that hide inside markdown code fences and prose:
+
+- **Bare `scripts/<domain>/<file>.py` paths** — references that don't
+  use the `${CLAUDE_PLUGIN_ROOT}` prefix.
+- **`from <module> import …` Python imports** — references that name
+  a module deleted in a v11 cleanup.
+
+`commands/jam/council.md` had a documented `from crew.hitl_judge import …`
+example pointing at a script deleted in PR #866. The council itself ran
+fine; the post-synthesis HITL judge step was broken. 9 other places
+across `commands/delivery/`, `commands/engineering/`, `commands/smaht/`
+had similar breaks (mostly references to the deleted
+`scripts/crew/crew.py::find-active` auto-resolver).
+
+### Fixed
+
+- `commands/jam/council.md` — replaced the HITL judge code example with
+  inline prose heuristics that the agent applies. Council itself
+  unchanged; only the gate-feeding step removed.
+- `commands/delivery/process-health.md`, `commands/smaht/state.md`,
+  `commands/smaht/briefing.md` — replaced the v6 `crew.py find-active`
+  invocations with v11 explicit-`--project` semantics.
+
+### Added
+
+- `scripts/ci/validate.py` now checks two additional reference shapes:
+  bare `scripts/.../*.py` paths anywhere in commands/agents markdown,
+  and `from <module> import` patterns whose module name doesn't exist
+  under `scripts/` or `hooks/scripts/`. Stdlib + common third-party
+  + obvious-template-placeholder names are whitelisted to keep the
+  signal-to-noise ratio high.
+
+The validator is now strict enough that a future v11 cleanup will
+catch its own drift on the next CI run rather than the next user run.
+
+390 / 390 tests still passing.
+
+---
+
+## [11.1.2] — 2026-05-08
+
+**marketplace.json plugin version sync.**
+
+The marketplace registration's `plugins[0].version` field had drifted —
+stuck at `8.8.1` across the three v11 releases (`v11.0.0`, `v11.1.0`,
+`v11.1.1`) because nobody bumped it. The plugin manifest reported
+`11.1.1`; the marketplace listing reported `8.8.1`. Anyone consulting
+the marketplace registration saw a stale number.
+
+### Fixed
+
+- `.claude-plugin/marketplace.json` `plugins[0].version` bumped to
+  `11.1.2` to match `.claude-plugin/plugin.json`.
+
+### Added
+
+- `scripts/ci/validate.py` now enforces the version-parity invariant.
+  When `plugin.json` and `marketplace.json` disagree on the plugin's
+  version, validation fails with a named error: *"marketplace.json
+  plugins[name=X].version = '8.8.1' does not match plugin.json version
+  = '11.1.1'. Bump both together."* This catches the same drift class
+  on the next release attempt rather than the release after.
+
+No behavior change beyond the version bump and the validator addition.
+Tests still 390/390 passing.
+
+---
+
+## [11.1.1] — 2026-05-08
+
+**Closes the wicked-brain + wicked-bus persistence verification gap.**
+
+Spent a session probing the plugin's relationship with both companion
+plugins. Brain persistence turned out to be functional (11,582 indexed
+items, 108 memories, search round-trip verified). Bus integration had
+a real bug: v11 archetype events were never registered in
+`BUS_EVENT_MAP`, so v11 wicked-garden was invisible to the bus despite
+the bus itself being healthy (19,790+ events from companion plugins).
+
+### Fixed
+
+- v11 archetype events now register + emit (PR #874). Five new event
+  types in `scripts/_bus.py::BUS_EVENT_MAP`:
+  - `wicked.archetype.created` — archetype-mode project init
+  - `wicked.archetype.advanced` — phase approved; carries next_phase
+  - `wicked.archetype.completed` — final phase approved
+  - `wicked.archetype.hard_gate_passed` — confirmed_by + evidence
+  - `wicked.archetype.classified` — LLM classifier result persisted
+- `phase_manager.create_project` emits `wicked.archetype.created` on
+  archetype-mode projects.
+- `phase_manager.approve_phase` emits `wicked.archetype.advanced` on
+  every archetype approval, plus `wicked.archetype.hard_gate_passed`
+  when `confirmed_by` was supplied, plus `wicked.archetype.completed`
+  when `is_complete()` flips true.
+- `scripts/classify/persist.py` emits `wicked.archetype.classified`
+  when the LLM classifier writes its result.
+
+All emits fail-open: bus unavailable never blocks the disk write.
+
+### Verified
+
+- **Brain persistence**: search returns 42 hits for `Session goal` —
+  the prompt-submit hook's `_write_brain_memory()` writes are reaching
+  the FTS5 index. Brain server runs on port 4243; 72 MB SQLite DB at
+  `~/.wicked-brain/projects/wicked-garden/.brain.db`.
+- **Bus round-trip**: real CLI invocation
+  (`phase_manager create … --archetype-mode build` + two approves)
+  advances `newest_event_id` by 3, with per-type counters incrementing
+  as expected.
+
+### Tests
+
+390 passing (was 384 in v11.1.0; +6 new in `TestBusEmits`).
+
+---
+
+## [11.1.0] — 2026-05-08
+
+**Closes the structural gaps in the v11.0.0 reframe.** v11.0.0 was the
+research-grade reframe shipped in two hours; v11.1.0 closes 11 of the
+12 items on the post-v11.0.0 critique. The remaining item (wicked-brain
+persistence) needs real session traffic to validate, not a marathon.
+
+### Added
+
+- **LLM-based archetype classifier** via the new `wicked-garden:classify`
+  skill (PR #869). The skill prompts the model to reason through the
+  user's prompt, classify into archetype(s), identify boolean signals,
+  and persist to SessionState via `scripts/classify/persist.py`. The
+  prompt-submit hook prefers persisted classifications over the regex
+  detector — regex is now Tier 2 fallback. When regex returns only
+  triage, the hook emits a `<wg classify-due />` directive inviting the
+  model to classify properly. New SessionState fields: `archetypes_v11`,
+  `signals_v11`, `classified_at`.
+- **Runtime hard-gate enforcement** (PR #870). Five archetype phases —
+  `migrate:cutover`, `incident:mitigate`, `review:remediate-or-accept`,
+  `specify:validate`, `decide:record` — now require non-empty
+  `--confirmed-by` AND `--confirmation-evidence` to advance. The audit
+  trail records both fields with `hard_gate=True`. The doctrine of
+  "hard:* gate" is now mirrored in `phase_manager` code, not just
+  playbook prose.
+- **5 restored QE utilities** under `scripts/qe/` (PR #868), framed as
+  v11 library tools the archetype playbooks call when relevant:
+  - `verdict_schema.py` (~250 LOC) — validates review-archetype
+    verdict artifacts. Slimmed from the v6 `gate_result_schema.py`.
+  - `verdict_audit.py` (~140 LOC) — append-only audit log of verdicts.
+    Slimmed from v6 `gate_ingest_audit.py` + `dispatch_log.py`. No
+    HMAC; v11 review enforces banned-reviewer checks at validation.
+  - `conditions_manifest.py` — track CONDITIONAL findings to
+    resolution. Idempotent re-init preserves prior resolutions.
+    `all_resolved()` answers the contract question.
+  - `content_sanitizer.py` — strip prompt-injection patterns from
+    reviewer free-text fields. Floor not wall.
+  - `evidence_tracker.py` — track per-archetype produces contracts
+    (e.g. `shipped-code` + `test-report` for build).
+- **v6→v11 project state migration tool**
+  (`scripts/setup/migrate_v6_projects.py`, PR #871). Detects v6-v10
+  projects in the DomainStore via phase-list pattern matching,
+  proposes the closest v11 archetype, and (with `--apply`) translates
+  state in place. Original phase plan preserved in
+  `extras.v11_migration_source` for audit.
+- **End-to-end integration test** for the full migrate lifecycle —
+  catalog hydration → state transitions per phase → hard-gate refusal
+  → hard-gate advance → `is_complete` (PR #870).
+- **58-prompt calibration corpus** (PR #872). Exercises all 9
+  archetypes plus paraphrase, multi-archetype, ambiguous edge cases.
+  Three locked-in tests assert overall recall ≥ 85%, overall precision
+  ≥ 80%, per-archetype recall ≥ 70%. Actual: **100% per-archetype
+  recall**, 90%+ precision after phrase tuning.
+- **GitHub Actions workflow** `.github/workflows/test.yml` runs the
+  full pytest suite + smokes phase_manager CLI archetype-mode + smokes
+  the archetype detector + smokes verdict_schema validation + asserts
+  the migrate:cutover hard gate refuses without confirmation and
+  advances with it. End-to-end CI for v11.
+
+### Changed
+
+- `hooks/scripts/prompt_submit.py` — `_build_archetype_directive` now
+  accepts `state`, prefers persisted LLM classification, emits
+  `<wg classify-due />` directive when regex returns only triage. Tag
+  format carries `classified="llm"` or `classified="regex"`.
+- `.claude-plugin/archetypes.json` — phrase lists tuned to 100% recall
+  on the calibration corpus. Added paraphrase coverage for build
+  (refactor / patch / wire up), migrate (rename the / expand the /
+  retire the / take the / column out), incident (production is broken
+  / postmortem / memory leak / paging us), ship (ship to / ship the),
+  review (take a look at / look at this pr), decide (pick a / pick
+  between / which database / which queue / which framework), and
+  explore (what should we / how might we / brainstorm / approaches to
+  / not sure how / ideas for).
+- `agents/crew/reviewer.md`, `agents/product/requirements-analyst.md`,
+  `commands/setup.md`, `commands/smaht/briefing.md` — replaced
+  references to deleted v6 modules (traceability.py, qe-evaluator
+  naming sweep, _stack_signals, archetype_detect, affected_repos)
+  with v11 equivalents (evidence_tracker, conditions_manifest, the v6
+  migration script, an explanatory note).
+- `skills/archetype/refs/build.md` and `refs/review.md` — playbooks
+  now reference the restored `scripts/qe/*` utilities at the right
+  phase boundaries.
+
+### Removed
+
+- `daemon/` (~8600 LOC, PR #871) — the v6-v10 projector daemon was
+  wired to gate-result schemas, dispatch-log HMACs, and projection
+  resolvers, all of which v11.0.0 deleted. v11 treats the bus as an
+  audit substrate, not a projection enforcement layer.
+- `.github/workflows/benchmark.yml` — enforced p95 SLO on
+  `_load_gate_result` (deleted in v11.0.0). Trigger paths no longer
+  exist; the workflow was dormant clutter.
+- 227 lines of dead code in `hooks/scripts/prompt_submit.py` (PR #867)
+  — `_assemble_current_chain`, `_consume_facilitator_reeval`,
+  `_consume_phase_start_gate`, plus their call sites and event-type
+  references that pointed at deleted v6 modules.
+
+### Test summary
+
+384 tests passing across 8 directories: tests/, tests/crew/, tests/qe/,
+tests/calibration/, tests/hooks/, tests/fixtures/, plus utilities.
+Calibration: 100% per-archetype recall on the 58-prompt corpus.
+
+---
+
+## [11.0.0] — 2026-05-07
+
+**Reframe: work-shape archetypes replace the universal pipeline.**
+
+The fixed `clarify → design → test-strategy → challenge → build → test → review`
+pipeline with a rigor-tier dial is gone. Each prompt now classifies into one
+or more **work-shape archetypes** (`triage` · `explore` · `specify` · `decide` ·
+`ship` · `review` · `incident` · `build` · `migrate`), each of which owns its
+own phase shape, produces contract, HITL discipline, and cost band.
+
+### What's new
+
+- **`.claude-plugin/archetypes.json`** — canonical catalog. Each archetype
+  declares its phases, produces, HITL level, cost band, maturity, and
+  detection signals (phrase list + boolean flags).
+- **`scripts/crew/archetypes_v11.py`** — detector + steering engine. CLI
+  shim for shell + agent integration. Stdlib-only.
+- **`skills/archetype/`** — agent-facing entry point. Slim `SKILL.md`
+  (101 lines) + 9 ref playbooks (62–97 lines each), one per archetype.
+  Each playbook documents the archetype's phases, produces contract,
+  HITL discipline, run procedure, exit condition, and anti-patterns.
+- **`commands/archetype/`** — 9 slash commands, one per archetype, for
+  direct user invocation.
+- **`hooks/scripts/prompt_submit.py`** — UserPromptSubmit hook now emits
+  a slim `<wg archetype="X" score="Y" />` system-reminder when a prompt
+  routes to a work shape. `simple-edit` intent stays silent.
+- **`scripts/crew/scope_delta.py`** — scope-delta heuristic to catch
+  silent project-sized additions to wave plans.
+- **`scripts/crew/phase_manager.py`** — slim project-state manager
+  (~370 lines, down from ~6000+). State CRUD + `--archetype-mode` for
+  new projects. No gate machinery.
+
+### What's gone
+
+The v6–v10 universal-pipeline machinery has been deleted:
+
+- **Agents**: `crew/contrarian`, `crew/facilitator`, `crew/gate-adjudicator`,
+  `crew/gate-evaluator`, `crew/independent-reviewer`, `crew/phase-executor`,
+  `crew/process-facilitator`, `crew/qe-orchestrator`. Kept:
+  `crew/implementer`, `crew/researcher`, `crew/reviewer` (general purpose).
+- **Commands**: all of `crew/` except `crew/archive` (general project
+  management).
+- **Skills**: `propose-process`, `facilitator-score`, all of `skills/crew/`.
+- **Plugin configs**: `gate-policy.json`, `phases.json`, `autonomy-policy.json`,
+  `finding-classification.json`.
+- **Scripts**: gate-result schema validators, gate-dispatch, dispatch-log
+  HMAC, conditions-manifest, content-sanitizer, consensus-gate, autonomy
+  policy table, reconcile / reconcile_v2, propose-process facilitator
+  rubric, validate-plan / validate-reeval-addendum, reeval-addendum,
+  archetype_detect (v6.3 target-kind classifier), challenge-manifest,
+  rigor-escalator, hitl-judge, gate-adjudicator, semantic-alignment,
+  convergence, swarm-trigger, factor-questionnaire, plus ~50 more
+  legacy scripts in `scripts/crew/`.
+- **Docs**: `MIGRATION-v7.md`, `V7-AC-TRACEABILITY.md`, `crew-workflow.md`,
+  `cross-phase-intelligence.md`, `autonomy.md`, `spc.md`, all of
+  `docs/v9/`, `docs/audits/`, `docs/calibration/`, `docs/cluster-a/`,
+  `docs/composition/`, `docs/council/`, `docs/evidence/`, `docs/research/`.
+
+### Migration note
+
+There is no migration. v11 is a clean break from v6–v10. Projects authored
+under the prior architecture continue to load (the project state schema
+is forward-compatible) but the gate / rigor / phase machinery is no
+longer enforced. New projects should be created with
+`phase_manager create … --archetype-mode <archetype>`, or simply by
+letting the UserPromptSubmit hook auto-route.
+
+### Pre-v11 history
+
+Versions 0.x through 10.x lived in the git log under tags `v0.1.0`
+through `v9.2.18`. Major themes across that span:
+
+- **v0–v3**: initial plugin scaffold, command + agent surface, hooks.
+- **v4–v5**: smaht context-assembly orchestrator (HOT/FAST/SLOW/SYNTHESIZE
+  tiered router). Replaced in v6.
+- **v6**: facilitator-driven dynamic phase plan, gate-policy×rigor matrix,
+  archetype-aware gate-adjudicator (target-kind classifier).
+- **v7**: AC traceability, semantic reviewer, convergence-verify gate.
+- **v8**: dropped `/wicked-garden:mem:*` (memory moved to wicked-brain).
+- **v9**: bus-as-truth cutover for gate-critical artifacts; drop-in
+  plugin contract; 14 default-on `WG_BUS_AS_TRUTH_*` tokens.
+- **v10**: intent variable replacing the 5-classifier cascade;
+  steering-not-blocking principle; slim-body contract for commands
+  and skills.
+- **v10.x dogfood PRs (#854–#860)**: surface fixes that pointed at the
+  deeper structural issue v11 solved — hardcoded deliverable filenames,
+  gate-result schema field aliases, gate-vocabulary unification,
+  reeval-log auto-stub, phase-plan authority hydration, CONDITIONAL
+  condition-resolution mechanism, scope-delta HITL trigger.
+
+For each of those releases, see the corresponding tag in git history.