retro

Sprint retrospective - review what worked, what didn't, improve process

SKILL.md

GitHub

Retrospective analysis. Review a time period, extract patterns, improve the system.

Steps

Step 1: Determine scope

If user specified a time period, use that. Otherwise default to "this week."

Step 2: Gather data (parallel reads)

Read simultaneously:

  • Recent daily notes (last 5-7 days)
  • .claude/logs/verdicts.jsonl (session quality trends)
  • .claude/logs/incident-log.md (issues and blocks)
  • .claude/logs/failure-log.md (tool failures)
  • .claude/knowledge-nominations.md (pending learnings)
  • Task Board.md (completion rate)

Step 3: Analyze patterns

What went well?

  • Tasks completed on time
  • Smooth workflows (no blocks)
  • Learnings successfully captured
  • Quality verdicts trending positive

What didn't go well?

  • Repeated failures (same error type)
  • Blocked commands that should have been allowed
  • Tasks that took much longer than expected
  • Context flushes (/clear) needed frequently
  • Quality verdict blocks

What to change?

  • Are there process bottlenecks?
  • Are hooks too strict or too lenient?
  • Are agents missing capabilities?
  • Are commands missing or underused?

Step 4: Extract improvements

For each identified improvement:

  1. Is it a knowledge-base rule? → Promote directly
  2. Is it a process change? → Add to Scratchpad for user review
  3. Is it a tool/config change? → Create task on Task Board
  4. Is it a pattern to watch? → Nominate to knowledge-nominations

Step 5: Write retro report

Add to daily note:

## Retrospective — [period]

### Went Well
- [bullets]

### Didn't Go Well
- [bullets]

### Action Items
- [ ] [specific improvement with owner]

### Metrics
- Tasks completed: [X]
- Quality verdict pass rate: [X]%
- Incidents: [X] (CRITICAL: [X], HIGH: [X])
- Context flushes: [X]

Step 6: Create action items

Add any actionable improvements to Task Board.md → This Week.