What problem does it solve? AI agents accumulate knowledge during tasks, but persisting every observation bloats or corrupts skill files. This Skill provides a disciplined end-of-task pipeline that decides whether new knowledge is important enough to persist, verifies it against external evidence, and safely integrates it into SKILL.md without regressions. ## Core Features & Use Cases - Five-step memory pipeline: importance gating, external verification, conflict detection, gated lean updates, and a separate git commit for auditability. - Verification-first persistence: knowledge is only written when confirmed by a green test/build/lint/eval, entailment against a cited file, or explicit user confirmation. - Promote-or-discard gating: changes are staged in SKILL.md.next and promoted only if validate-skill.mjs passes with no correct-to-wrong regressions. - Use Case: After fixing a subtle bug in a crawler, run this pipeline to check whether the lesson is non-obvious and verified, then replace the outdated passage in the relevant SKILL.md and commit it separately. ## Quick Start At the end of your task, ask the agent to run the meta-skill-evolution pipeline for each skill involved and persist only verified important learnings.
