A Claude Code skill that reviews a finished session for the moments you had to
correct or steer the agent, and turns the ones worth keeping into durable rules:
global or project CLAUDE.md, auto-memory, or planning docs. Nothing is written
without your approval.
git clone https://github.com/dankochetov/self-improve.git ~/.claude/skills/self-improveThen run /self-improve at the end of a session.
Add this to your ~/.claude/CLAUDE.md so Claude offers it instead of waiting for
you to remember:
## Session wrap-up
- When a work session included corrections, steering, or discoveries worth keeping, offer to run /self-improve before wrapping up. Skip the offer after trivial or purely conversational sessions.