retro

Every completed milestone deserves a retrospective. This skill analyzes hard data from git history and GitHub to identify what went well, what went wrong, and what to change. No vague observations — every finding must be backed by data, and every recommendation must be actionable.

Step 1: Gather Data

Run these commands to collect metrics for the milestone period:

git log --oneline --since="MILESTONE_START" --until="now"

git shortlog -sn --since="MILESTONE_START"

gh issue list --state closed --milestone "MILESTONE_NAME"

gh pr list --state merged --search "milestone:MILESTONE_NAME"

For each PR:

gh pr view NUMBER --json additions,deletions,changedFiles,commits

Step 2: Analyze Metrics

Calculate and report these metrics in tables.

Velocity Metrics:

Quality Metrics:

Process Health:

Step 3: Identify Patterns

Based on the data, answer these questions:

1. What worked well? — Which practices produced the best metrics? (e.g., "Issues created via work-breakdown had 40% smaller PRs")
2. What went wrong? — Where did metrics deviate from targets? (e.g., "3 PRs exceeded 200 lines — scope creep on feature X")
3. Root causes — For each "went wrong" item, ask WHY twice. Do not stop at symptoms.

Step 4: Generate Action Items

For each root cause identified, produce ONE specific, actionable improvement.

Format:

### Action Items
1. **[Problem]**: [Root cause] → **Action**: [Specific change to make]
   - Owner: [skill to update / process to change]
   - Measure: [How to verify this improved next milestone]

Rules:

• Maximum 3 action items per retro (focus beats breadth)
• Each must be implementable within the framework (update a skill, add a hook, change a threshold)
• Each must have a measurable success criteria for the next retro to verify
• If a skill's SOP contributed to a problem, propose a specific edit to that skill

Step 5: Store and Compare

• Store the retro report in auto memory — write to memory/retro_learnings.md and update MEMORY.md index with a one-line summary of key action items
• If a previous retro exists in memory/retro_learnings.md, read it first to compare metrics
• If a previous retro exists, compare: did the action items from last time improve the metrics?
• Track improvement trends across milestones

Output Format

## Retrospective: [Milestone Name]
**Period**: YYYY-MM-DD to YYYY-MM-DD | **Issues**: N closed | **PRs**: N merged

### Velocity
[table from Step 2]

### Quality
[table from Step 2]

### Process Health
[table from Step 2]

### What Worked Well
1. [Data-backed observation]

### What Went Wrong
1. [Data-backed observation] — Root cause: [why]

### Action Items
1. [Specific actionable improvement with measurement]

### Comparison with Previous Retro
[If exists: did prior action items improve metrics?]

Next Steps

Report to user: "Retro complete. Velocity: [summary]. Quality: [summary]. Top 3 action items: [list]"

Suggested next steps (user decides):

• Action items need implementation → "Update affected skills"
• Ready for next milestone → "Run roadmap-steer"