| name | reflect |
| description | Capture non-obvious lessons from recent activity. Dispatches the reflector agent to scan commits, edits, and dispatches in scope and propose memory-entry candidates ready to accept. |
| user-invocable | true |
| allowed-tools | Read, Bash, Grep, Glob |
Reflect
Catch lessons before they evaporate. The reflector agent scans recent activity, finds non-obvious moments worth carrying forward, and proposes structured memory-entry candidates per the workspace's feedback-memory format.
When to use
- After a PR merges and you noticed something non-obvious during the work
- After multi-step debugging where a pattern emerged
- At end of session when the work was substantive (not routine)
- Whenever you corrected the system on something specific — the correction is the lesson
Skip when the session was routine. Most sessions don't produce lessons; that's fine.
What this does
-
Parses $ARGUMENTS for scope:
- No arg → last session (recent commits + recent edits + recent agent dispatches)
--pr <N> → activity scoped to PR N
--last <N> commits → last N commits on current branch
--since YYYY-MM-DD → activity since date
-
Dispatches the reflector agent (via the Task tool with subagent_type: reflector) with the scoped context.
-
The agent returns 0-N memory-entry candidates with triggering moments, lessons, and ready-to-paste MEMORY.md pointer lines.
-
You accept / reject / edit each candidate. Files are written only after your acceptance.
Arguments: $ARGUMENTS