retrospective-codify

To turn trial-and-error learnings into reusable AI rules, map task failures to successful resolutions, extract explicit guidance, and output the deduplicated knowledge as lint rules, skills, or AGENTS.md/CLAUDE.md updates to prevent recurring mistakes.

Updating CLAUDE.md with task knowledge requires consolidating initial attempts, final solutions, and insights into a predefined template. The process detects existing knowledge duplicates and appends or updates the rules to enforce deduplicated knowledge management.

Preventing AI agents from repeating coding mistakes involves codifying task completion insights into structured lint rules or AGENTS.md files. This bridges failures and successes, providing explicit guidance that restricts future agents from falling into known pitfalls.

Yes, you can create lint rules from project retrospective sessions by extracting lessons from task trial-and-error. The workflow automatically classifies the output destination and formats the deduplicated insights into a predefined template suitable for lint rule enforcement.

Deduplication when adding new rules works by detecting overlaps between freshly extracted insights and existing knowledge. It then proposes targeted appendings or updates to established skills and rules, ensuring the knowledge base remains concise without redundant entries.

Codify task learnings into a new skill when the insight represents a distinct workflow, and update AGENTS.md for project-specific behavioral guidance. The workflow automatically classifies and routes the deduplicated knowledge to the appropriate predefined template.