post-session-reflection

Post-Session Reflection

Use this skill when the user signals the session is winding down (phrases like "wrap up", "what did we do", "summarize", "done for today"), or immediately after a multi-tool research collage that warrants a recap.

Output format — exactly three bullets

- **Produced**: <artifacts — tool outputs, files, commits, plan docs>
- **Decided**: <non-obvious choices made this session, with the reason>
- **Next**: <explicit next step, owner, and trigger condition>

Rules

1. Three bullets, no more, no less. If you cannot compress to three, drop the least important item; do not add a fourth.
2. Each bullet is one sentence. Long bullets hide decisions.
3. "Produced" stays concrete — link file paths and commit SHAs when they exist, not abstractions.
4. "Decided" favours the non-obvious. Trivial defaults ("we used the default timeout") do not belong here. A fork in the road does.
5. "Next" is actionable. A task the next session can pick up without re-reading the transcript. Include the trigger (e.g. "when BTC >= $X" or "before PR #Y merges").

Example

- **Produced**: `MetricsDecorator` + `HealthMonitor` (commits b7c6991, bb1b8aa) and 24 new unit tests in `tests/unit/observability/`.
- **Decided**: Prometheus-compatible dict snapshot over text format for now — keeps the surface JSON-serialisable until we add the HTTP endpoint in Phase 2.
- **Next**: Wire `HealthMonitor.start()` into `build_production_service` and expose `venue_health_check_total` via `/cryptozavr:health` when Supabase is reachable again.

When NOT to use

• Mid-session debugging (use venue-debug instead).
• Quick Q&A where no artifacts were produced.
• Planning sessions — use the feature-dev skill's plan format there.