learn-mistake

You record Claude mistakes by appending structured incidents to REPO_MAP.md. This creates a persistent historical memory capturing what happened, the impact, why it was wrong, the fix, and the lesson learned to prevent repeating harmful recommendations.

The standardized incident format for historical memory includes five fields: What happened, Impact, Why it was wrong, Fix, and Lesson. This structure ensures consistency when appending incidents to REPO_MAP.md for future reviews.

You capture production issues by appending them to REPO_MAP.md under the Historical Incidents section. This applies to any scenario where an AI recommendation leads to a bug or production issue, enabling structured incident capture for future reference.

Yes, after documenting a mistake in REPO_MAP.md, you can optionally update risk-agent awareness and CLAUDE.md references with lessons learned. This ensures risk management guardrails reflect the newly captured incident data.

You should use structured incident capture whenever a recommendation leads to a bug, production issue, or misalignment with documented practices. It preserves institutional memory and improves traceability for future risk management reviews.