reflect

To capture learnings for CLAUDE.md, use a post-conversation review process to identify missteps and successful patterns, then propose concrete edits. This surfaces actionable improvements to tighten guidance for future sessions.

The best way to identify gaps in skill guidance is running an end-of-conversation review. This examines the completed workflow to find misalignments and outputs recommended edits to existing skills to maintain consistency across future runs.

Yes, a post-conversation review can propose new skills. By evaluating the current conversation's workflow, it identifies unmet needs and surfaces concrete changes needed, including suggesting entirely new skills to handle uncovered tasks.

To generate actionable edits for skill development without external actions, apply a structured review that specifies actionable criteria. It outputs recommended edits and next steps for CLAUDE.md or skills internally, without executing any external changes.

The conversation review process works both during and after active sessions. Applying it mid-conversation allows you to review guidance, identify misalignments, and propose concrete edits to refine CLAUDE.md or skills on the fly.

From an end-of-conversation reflection, you should expect outputs specifying what worked, what didn't, and two concrete improvements. These recommended edits and next steps refine CLAUDE.md, existing skills, or new skills for future sessions.