learn-from-mistakes

Record retrospective logs after human sign-off to capture corrections, missed cases, and implementation gaps for pattern tracking. Entries are written to .claude/learnings.md with the date, feature, description, and a pattern tag for trend analysis.

Track implementation gaps by logging human corrections and missed cases after sign-off. This accumulates insight over time, and when the same mistake occurs three or more times, it surfaces a concrete suggestion to update AGENTS.md or improve existing skills.

Pattern aggregation works by logging retrospective entries with short pattern tags into .claude/learnings.md. When recurring patterns appear three or more times, the system surfaces concrete suggestions to update AGENTS.md or improve skills for future work.

Trigger a retrospective log after a human marks a chunk complete, or when the user says "log what went wrong", "retrospective", "what did we miss", or "learn from this". It runs post sign-off to capture corrections without fixing anything directly.

No, logging retrospective entries does not fix anything. The goal is to accumulate insight over time for pattern tracking. It captures corrections and implementation gaps, surfacing suggestions to update AGENTS.md or improve skills when patterns recur.

Yes, retrospective logs aggregate patterns over time. When the same class of mistake occurs three or more times, the system surfaces a concrete suggestion to update AGENTS.md or improve an existing skill based on the captured corrections and missed cases.