| name | patrol |
| description | Report-only maintenance sweep — dependency audit, test-suite health, security quick-scan, workboard grooming. Never changes code; produces a findings report and proposed backlog tickets. Use for recurring maintenance ("check the project's health") or schedule it with /loop. |
| argument-hint | [deps | tests | security | board | all] |
/patrol — Report-Only Maintenance Sweep
L1 autonomy: read and report, never fix. Safe to run on a schedule
(/loop 1d /patrol) or in CI — it cannot break anything.
- Scope from
$ARGUMENTS (default all). Run the relevant checks — spawn
one agent per area, in parallel, each explicitly instructed READ-ONLY:
deps — ecosystem audit (npm audit / pip-audit / cargo audit):
exploitable vulns relevant to actual usage, plus major-version drift on
direct dependencies.tests — run the full suite 2× and report: failures, flakes (pass/fail
flicker), runtime trend, coverage holes on recently changed files.security — spawn security-auditor scoped to changes since the last
patrol (or last 20 commits) — quick pass, CRITICAL/HIGH only.board — workboard grooming report: stale in_progress/blocked tickets,
unfixed MEDIUM/LOW findings aging in ticket sections, epics stuck in
verifying, drift between BOARD.md and ticket frontmatter.
- Report one compact digest: per-area verdict (✅ / ⚠️ / 🔴), findings by
severity, and a proposed actions list.
- Propose, don't do. For each action worth taking, offer to create a backlog
ticket (
/board add style — batch, one confirmation). Fixing happens later
via /ship or /debug-findings, initiated by the user.
- Append a one-line entry to
workboard/BOARD.md's activity log:
patrol: <date> — <verdicts summary> so patrol history is visible.
If the project has no workboard, run the checks anyway and just report.
