| name | retro |
| description | Metrics-driven retrospective — commits, files changed, lines +/-, tasks completed vs failed, drift items, blockers carried over. Reads git log, .pipeline/ sprint records, QA reports, handoff archives. Gaps and failures listed FIRST, then patterns, then what worked. Concrete recommendations for next session. Use after sprint completion or significant work session. |
| disable-model-invocation | true |
| argument-hint | [sprint-N | session | all] |
Produce an honest, metrics-backed retrospective. Gaps and failures first, strengths second. Specific numbers, not vibes.
Git history
git log --oneline --since="7 days ago" 2>/dev/null || git log --oneline -30 2>/dev/null || echo "no git"
Branch info
- Branch: !
git branch --show-current 2>/dev/null
- Commits in last 7 days: !
git rev-list --count --since="7 days ago" HEAD 2>/dev/null || echo "unknown"
1. Determine scope
Based on $ARGUMENTS:
sprint-N: Read .pipeline/build/sprints/N/ completion records + SPRINT-REPORT.md, and the QA report at .pipeline/review/sprints/N/QA-REPORT.md.session: Use git log from today (or since last handoff timestamp from .claude/handoff-*.md).all: Read all available sprint data + full git history on branch.- No argument: Default to
session.
2. Gather metrics
Collect what's available (skip what doesn't exist):
From git:
- Total commits in scope
- Files changed (count + list)
- Lines added / removed
- Commit frequency pattern (bursts vs steady)
From pipeline (if .pipeline/ exists):
- Tasks attempted vs completed vs failed vs blocked
- Drift items found (from QA-REPORT.md or sprint reports)
- Verify verdicts (
implemented | partial | missing | drift | manual) and build per-task verdicts (verified / drifted / paused / contradiction / synthetic)
- Wave count and wave sizes
From handoff archives (if .claude/handoff-*.md exist):
- Blockers that persisted across sessions
- Remaining items that carried over (scope creep signal)
3. Generate retrospective
Write output in this structure (to stdout, not a file, unless user asks to save):
## Retrospective: <scope>
**Period:** <date range>
**Branch:** <branch>
### Metrics
| Metric | Value |
|--------|-------|
| Commits | N |
| Files changed | N |
| Lines +/- | +N / -N |
| Tasks completed | N/M |
| Drift items | N |
| Blockers | N |
### Gaps & Failures (address first)
<Numbered list. Each: what went wrong, why, impact. Be specific — file names, task IDs, error messages.>
### What Drifted
<Items where implementation diverged from spec/plan. Include: what was expected, what happened, whether it was resolved.>
### Patterns
<Recurring themes: types of bugs, common drift causes, bottleneck areas, tools that helped/hindered.>
### What Worked
<Specific things that went well. Approaches, tools, decisions that paid off.>
### Recommendations
<Concrete suggestions for next session/sprint. Each: what to do differently + expected impact.>
4. Offer to save
Ask user if they want the retro saved. If yes, write to:
.planning/retros/retro-<date>.md (if .planning/ exists)- Or
.claude/retros/retro-<date>.md (fallback)
