retrospective-codify

To turn task learnings into reusable rules, you map initial failures to final solutions and document the bridging insights as actionable templates, ast-grep rules, or CLAUDE.md entries to prevent repeating past mistakes.

The best way to document project learnings as ast-grep rules is to run a retrospective workflow that extracts insights from completed tasks and classifies them into static rules, ensuring deduplication against existing knowledge for unique value.

Failure-to-success mapping for knowledge management works by recording initial failed attempts, capturing the final working solution, and isolating the specific bridging insights needed to encode this experience into repeatable skill templates.

Yes, you can update CLAUDE.md files from task retrospectives by codifying extracted learnings into structured guidance entries, performing quality checks to avoid duplicates, and writing the new rules directly into the configuration.

You should codify learnings into a new skill versus a static rule when the documented knowledge requires complex reasoning or repeatable multi-step logic, rather than simple static pattern matching or basic CLAUDE.md guidance updates.

Yes, codifying retrospective learnings prevents duplicate knowledge entries by enforcing structured artifact creation and performing deduplication checks against existing rules to ensure any new documentation provides unique value.