Capture

MANDATORY PREPARATION

Invoke /agent-workflow — it contains workflow principles, anti-patterns, and the Context Gathering Protocol. Follow the protocol before proceeding — if no workflow context exists yet, you MUST run /teach-maestro first.

Capture the current session's work into a persistent summary. This creates a record that survives session boundaries — future commands can reference what happened here.

What to Capture

Read the conversation history and identify:

1. Commands run — which Maestro commands were invoked and in what order
2. Decisions made — architectural choices, trade-offs accepted, patterns adopted
3. Files changed — what was created, modified, or deleted
4. Issues found — bugs, gaps, or risks identified
5. Next steps — what should happen next (be specific)

Output Format

Generate a session summary file at .maestro/sessions/{date}_{topic}.md:

# Session: {topic}
Date: {YYYY-MM-DD}

## Commands Run
- /diagnose → Score: 18/25
- /fortify → Added retry logic to API handlers
- /evaluate → Verified with 3 test scenarios

## Decisions
- Chose retry-with-backoff over circuit breaker (simpler, sufficient for current load)
- Kept synchronous error handling (async not justified yet)

## Files Changed
- `src/api/handler.ts` — added retry wrapper
- `src/middleware/auth.ts` — added input validation
- `tests/api.test.ts` — new test file

## Open Issues
- Rate limiting not yet implemented (deferred to next session)

## Next Steps
1. Run `/guard` to add rate limiting
2. Run `/evaluate` with adversarial test cases

Also append a decision entry to .maestro/decisions.jsonl recording this capture.

Session Capture Checklist

• All commands from this session listed
• Key decisions documented with rationale
• Files changed listed with brief descriptions
• Open issues captured (nothing lost)
• Next steps are specific and actionable
• Summary saved to .maestro/sessions/

Recommended Next Step

After capturing, your next session should start with /recap to restore context, then proceed with the next steps listed above.

NEVER:

• Store raw prompts or full LLM outputs in the session file
• Overwrite an existing session file — create a new one
• Skip the "Next Steps" section — this is what makes capture valuable
• Capture without reading the conversation history first
• Auto-capture without the user invoking this command