reflection

Session reflection captures mistakes and corrections, then appends extracted atomic rules to project skill files. By explicitly triggering a reflection command, you create a feedback loop that updates target project conventions for subsequent sessions.

Learning from user feedback works by analyzing session corrections when explicitly triggered, extracting durable rules, and appending them under a Learned Rules section. This process updates the target project's skill directory to enforce refined behavior in future sessions.

You should trigger a reflection update at the end of a Claude Code session after providing corrections. It runs only when explicitly requested, ensuring safe, targeted updates to project skill files rather than automatic background changes.

Updating project rules preserves existing skill formats while appending new atomic rules under Learned Rules. It generates clear semantic commit messages to maintain a clean history and audit trail within the target project directory.

Yes, committing learned rules to git is handled automatically during the reflection process. The skill extracts durable rules from session feedback, updates the target files, and commits changes with semantic messages to preserve a clean history.

The best way to stop repeating mistakes is to run a reflection command after a session. This extracts corrections as atomic rules, appends them to your project's skill files, and creates a feedback loop for future automation.