/phase2-performance

Phase 2: Performance and Improvement Analysis

After registration audit completes, automatically analyze improvement opportunities.

Step 1: Check Skill Performance Metrics

For each plugin being updated, invoke `pensive:skill-review` to analyze execution history:

# If updating specific plugin
pensive:skill-review --plugin <plugin-name> --recommendations

# If updating all plugins
pensive:skill-review --all-plugins --recommendations

Look for:

• Unstable skills (stability_gap > 0.3)
• Recent failure patterns
• Performance degradation trends
• Low success rates (< 80%)

Step 2: Surface Recent Failures

Query skill execution logs for failures:

/skill-logs --plugin <plugin-name> --failures-only --last 7d

Extract:

• Common error messages
• Recurring failure patterns
• Environmental dependencies causing issues

Step 3: Check for Workflow Improvements

1. Check if `sanctum:workflow-improvement` skill has been invoked recently 2. Review git history for recent fixes to commands/skills/agents in this plugin 3. Check issue tracker for open improvement issues

Command to check recent workflow fixes:

git log --oneline --grep="improve\|fix\|optimize" \
  --since="30 days ago" -- plugins/<plugin-name>/

Step 4: Generate Improvement Recommendations

Based on Steps 1-3, create recommendations in this format:

## Improvement Recommendations for <plugin-name>

### Critical (Immediate Action)
- [ ] Skill: <skill-name> - Stability gap: 0.45 - Review error handling
- [ ] Command: <command-name> - 5 failures in last week - Missing validation

### Moderate (Schedule for Next Sprint)
- [ ] Agent: <agent-name> - Performance degradation detected - Review token usage
- [ ] Skill: <skill-name> - Low success rate (72%) - Improve documentation

### Low Priority (Backlog)
- [ ] Hook: <hook-name> - Occasional timeouts - Add async handling

Step 5: Create Action Items (Automatic)

Critical and Moderate issues are automatically logged to GitHub issues.

For each Critical or Moderate recommendation:

1. Check for duplicates in existing issues first 2. Create GitHub issue with appropriate labels:

• Critical: `high-priority`, `plugin:<name>`, component label
• Moderate: `medium-priority`, `plugin:<name>`, component label

3. Report created issues to user at end of command

Low Priority items are reported but NOT auto-created (remain in backlog documentation).

To skip automatic creation: use `--no-auto-issues` flag.

Also create TodoWrite items for immediate tracking:

improvement:<plugin-name>:skill-<name>-stability
improvement:<plugin-name>:command-<name>-validation
improvement:<plugin-name>:agent-<name>-performance