reflect

name	reflect
description	Reviews the current conversation to extract learning opportunities, skill improvements, and memory candidates. Presents findings for user approval.
user-invocable	true

Session Reflection

Your task is to review the current conversation and extract actionable improvements.

Steps

1. Scan the conversation for:

   • Tasks completed (what was done)
   • Errors encountered (what went wrong)
   • User corrections (where Claude was wrong or suboptimal)
   • Repeated patterns (actions done more than once)
   • Tool usage patterns (which tools were used most, any inefficiencies)

2. Categorize findings into:

   Learning opportunities:

   • Mistakes that could be prevented by a rule in CLAUDE.md or a rule file
   • Domain knowledge that should be saved to memory
   • Patterns specific to this project that Claude should remember

   Skill creation candidates:

   • Multi-step workflows that were performed manually and could become a skill
   • Recurring tasks that would benefit from automation

   Skill improvement candidates:

   • Existing skills that failed or needed manual correction
   • Skills that could be more concise or effective

   Memory candidates:

   • Project-specific facts discovered during the session
   • Debugging insights or solutions to tricky problems
   • Architecture patterns or file relationships

3. Present findings as a prioritized list with:

   • Category (Learn / New Skill / Improve Skill / Memory)
   • Description of the finding
   • Suggested action (specific rule text, skill outline, or memory entry)

4. Ask the user what to implement.

5. Execute approved actions:

   • Write new rules to .claude/rules/
   • Update existing skills
   • Create new skills in .claude/skills/
   • Save knowledge to auto memory