| name | system-review |
| description | Run a system review to check project health.
Use when issues drop below 5, an EPIC closes, or 7+ days since last review.
Triggers on "system review", "project health", "review system".
|
| allowed-tools | Read, Edit, Write, Bash, Grep, Glob, Task, WebSearch |
System Review Skill
Trigger Conditions
Run when ANY occur:
- Open GitHub issues drop below 5
- An EPIC issue is closed (label:
epic)
- 7 days since last review
- Manual request
Review Checklist
Phase 1: Registry Reconciliation
grep -c "status: implemented" docs/research/registry/techniques.yaml
grep -c "status: planned" docs/research/registry/techniques.yaml
Phase 2: Documentation Sync
Phase 3: Issue Health
gh issue list --state open --limit 50
Phase 4: Generate Report
Create GitHub issue titled "System Review: YYYY-MM-DD" with findings and action items.
TZ='America/New_York' date '+%Y-%m-%d'
gh issue create --title "System Review: $(TZ='America/New_York' date '+%Y-%m-%d')" \
--label "maintenance" --body "## Findings\n\n[report here]"
Anti-rationalization — System review
| Excuse | Counter |
|---|
| "Skip the review, I just looked at this last week" | A week is enough for new alerts, dep advisories, and CI flakes. Run all phases. |
| "No new issues, the review is wasted time" | The review's value isn't in finding new issues — it's in confirming the system isn't drifting silently. Empty reviews are good signal. |
| "I'll skip Phase X, it's never useful" | If a phase is never useful, file an issue to remove it. Don't silently skip — the next reviewer will skip a different phase. |
Red flags
- Phase X "passed" with no evidence captured (dates, alert counts, CI status)
- Stale-issue review missed an issue >90 days old
- New CodeQL alert appeared but not triaged in the review
- Review issue created but never closed (review work lingers)
Verification checklist
