lifecycle: validate the registry files in check, reconcile planned.md

[Jammy2211]

Why

scripts/lifecycle.py check reported OK over a planned.md in which 9 of 13 entries were wrong. cmd_check validated only two conditions, both about active.md and the state folders — it never opened planned.md or parked.md, and never resolved a single prompt: path.

The registry is the first thing a task-selection pass reads, so the rot has a measured cost: a session picked the two highest-leverage planned.md entries and spent most of itself discovering both were already fixed on main.

What this adds

Three registry checks over active.md / planned.md / parked.md:

• every prompt: path resolves — and resolves exactly, not via the pre-lifecycle PyAutoMind/<work-type>/<target>/ fallback;
• it resolves into a state folder its registry implies — planned → draft/, active → active/, parked → either, since parked.md holds both scoped and started-then-parked tasks;
• no slug is listed in two registries at once.

Output shape and the exit-code contract (0 OK, 1 drift) are unchanged.

What this reconciles

Six of the thirteen entries were work that had already shipped — verified on each upstream main, not inferred:

Entry	Shipped as
notebook-kernel-cwd-auto-simulate	PyAutoHands build_util.py:302 routes through run_notebook.py, setting resources['metadata']['path']
auto-simulate-guard-wrong-simulator-target	autolens_workspace — 236 resolvable should_simulate guards audited against the 54 simulators' declared outputs, 0 mismatches
build-testpypi-rehearsal-mode (M1)	PyAutoHands release.yml — rehearsal dispatch input, resolve_mode job, downstream jobs gated if: rehearsal != 'true'
heart-ci-linkage (M0)	PyAutoHeart heart/checks/ci_status.{sh,py}; the script's own comment says it "replaces the old gh run list --limit 1" — verbatim the defect the entry described
heart-release-validation (M2)	PyAutoHeart pyauto-heart validate --ingest → heart/validate.py, validation_report.json, release-integrate.yml
heart-release-profile-wheel-integration (M3)	PyAutoHeart heart/validate.py carries the named release profile and gates fidelity on profile == release

The entire M0–M3 release-validation milestone chain shipped without one registry entry being retired.

These are removed rather than given fabricated complete/ records — they shipped under other tasks' PRs, and dated records with merge evidence nobody verified would put a worse lie somewhere more trusted. Three legacy paths normalised to draft/. planned.md goes 13 → 7 entries, all paths exact.

Tests

tests/test_lifecycle_check.py — the first test for lifecycle.py. Fictional, hermetic fixtures per the KEEP-copied-into-the-template rule (test_spawn_privacy.py), with every leg driven by input that must trip it. Mutation-checked: reverting either fix fails the specific test guarding it.

Two checks were themselves wrong before these tests pinned them — the parked.md state rule, and the trailing-parenthetical path form (- prompt: ....md (carries the phase table)).

python3 scripts/lifecycle.py check → OK. pytest tests/ → 99 passed.

Known limitation

A local check cannot catch the class that cost the most. All six shipped entries were found by reading upstream code; five were catchable here only because their prompt files happened to be missing. Had those files existed, check would still print OK. Catching that class needs each entry's issue: cross-checked against GitHub, which makes check non-hermetic — deliberately out of scope, recorded in the prompt.

Prompt: draft/maintenance/pyautomind/registry_integrity_check.md

---

Generated by Claude Code