reflect

You can extract learnings from session history by analyzing long-running conversations to surface corrections, successes, and recurring patterns. The reflect skill processes this session data to identify actionable improvements and persists them directly into skill files.

The best way to analyze long-running conversations is to manually trigger a reflection command after a session. This assesses the source data, classifies the extracted learnings by confidence, and writes updates directly to the repository or local skill directories.

Yes, the reflect skill supports manual confirmation before applying any changes. It proposes skill updates based on the session analysis and waits for your approval before writing the persistent learnings to the repository in-place.

Yes, session reflection is applicable to long-running conversations across projects. It analyzes the session history to detect recurring patterns and corrections, then persists those learnings to improve targeted skills regardless of the project scope.

Extracted session learnings are stored as persistent updates written directly into skill files. You can choose to save these updates in-place within the repository or persist them to a local skills directory at ~/.claude/skills.

You should use manual activation after completing a long-running conversation or session. Running the reflect command at this point ensures that all corrections and successes are captured and classified by confidence before being persisted to skill files.