maintenance: self-heal stale complete/index.md in Lifecycle Drift CI

[Jammy2211]

Overview

The Lifecycle Drift workflow fails lifecycle.py index --check on every push to main while complete/index.md is stale, emailing the human once per push. Three storms so far (2026-07-24 → PR #97; 2026-07-30 morning → aba2ce5; 2026-07-30 evening → fa35972, ~25 emails in 2h). The fold-into-record --apply pattern cannot cover ad-hoc mutations of complete/ (the third break was a manual git mv of 10 phase prompts after the index regen, in the same commit). With many concurrent agent sessions, prevention alone will keep losing — the check must repair, not alarm.

Plan

• On push to main, a stale index is repaired in CI: regenerate with index --apply, verify convergence, commit as the github-actions bot, push (rebase-retry against concurrent main advances).
• The run fails — and emails — only if the repair itself fails (non-convergence or push rejected after retries).
• Pull-request runs keep the current read-only fail-with-message behaviour.
• The semantic lifecycle.py check stays a hard fail (state contradictions cannot be auto-fixed).

Detailed implementation plan

Affected Repositories

• PyAutoMind (primary, only)

Branch Survey

Repository	Current Branch	Dirty?
./PyAutoMind	main	clean (in sync with origin/main)

Suggested branch: feature/lifecycle-drift-self-heal

Implementation Steps

1. .github/workflows/lifecycle_drift.yml:
   • Raise permissions: contents: write (needed for the heal push; checkout's persisted GITHUB_TOKEN credentials are used).
   • Keep the lifecycle check step unchanged (hard fail).
   • Replace the index freshness step with a self-healing script:
     • index --check passes → exit 0.
     • Not a push event → fail with the current "run index --apply" message (PR behaviour unchanged).
     • Else: index --apply, then index --check again — if still drifting, fail (regen does not converge; genuine bug).
     • git config the github-actions[bot] identity, git add complete/index.md, commit "lifecycle: self-heal stale complete/index.md".
     • Push with up to 3 attempts: on rejection, git pull --rebase origin main, re-run index --apply (new records may have landed), git commit --amend --no-edit, re-verify, retry. Exhausted retries → fail.
   • Loop safety: pushes made with the default GITHUB_TOKEN do not trigger new workflow runs, so the heal commit cannot re-trigger; the pre-push convergence check guards regardless.
   • Update the workflow's header comment to describe the self-healing contract.
2. Validation:
   • bash -n / YAML-lint the workflow locally; workflow_dispatch a green run post-merge.
   • Controlled end-to-end test post-merge: push a deliberately index-stale (but semantically valid) tweak to a completion record directly to main, watch the workflow push a bot heal commit instead of failing, then confirm the next push is green.

Key Files

• .github/workflows/lifecycle_drift.yml — the only file changed.
• scripts/lifecycle.py — read-only reference (cmd_index --check/--apply semantics); no changes expected.

Original Prompt

Click to expand starting prompt

Make PyAutoMind's Lifecycle Drift CI self-healing: on index drift, run lifecycle.py index --apply, commit and push the regenerated index as a bot commit (with loop guard), failing/emailing only if the repair push itself fails.

Context: third drift break today (fa35972 — ad-hoc git mv of 10 phase prompts into complete/2026/07/ after record --apply) caused ~25 failure emails 17:35–19:27 UTC; the fold-into-record-apply pattern cannot cover ad-hoc mutations of complete/. Full prompt: active/lifecycle_drift_self_heal.md.