evolvelisted

# evolve — repo-medic self-evolution Self-contained skill. Distills "pitfalls hit recently" into rules injectable into SKILL.md, so future runs stop hitting the same pitfalls. ## 🛑 MANDATORY WORKFLOW — check all before declaring done ### Phase 0: Collect sources - [ ] **Read** `references/lesson-schema.md` (lesson extraction format) - [ ] **Set the time window**: default `--since 7d` (user may change it to 1d / 30d / since-last-tag) - [ ] **Scan git log**: extract commit message + body (using `git log --since=... --pretty=full`) - [ ] **Scan work-notes**: read `docs/work-note/*.md` and regex-match the `**Symptom**:` / `**Cause**:` / `**Fix**:` fields - [ ] **Optional**: scan KB public-knowledge for lesson-type content - [ ] **Run** `python scripts/extract.py --since 7d --repo .` to extract candidates - [ ] 🛑 **GATE**: extraction complete + candidate list presented to the user ### Phase 1: Distill (candidate → lesson) - [ ] **Deduplicate**: merge entries with similar Symptom/Cause - [ ] **Rank by frequency**: prioritize anything appearing ≥ 2 times (high ROI) - [ ] **Tag by category**: give each lesson one tag (DB / deploy / config / frontend / general) - [ ] **Filter scope**: skip anything unrelated to repo-medic (user business logic, third-party bugs, etc.) - [ ] **Abstract into rules**: turn concrete cases into reusable "when X happens, do Y" patterns - [ ] 🛑 **GATE**: after distillation, show the user ≤ 10 candidates (> 10 = not converged, re-extract) ### Phase 2