patrol

Patrol Cycle

Execute a full Gas Town system patrol in sequence without pausing for confirmation between steps. Report a single summary at the end.

Patrol Sequence

Run all 6 components in this exact order:

1. hooks — Verify all hook scripts exist and are executable (~/.claude/hooks/)
2. inboxes — Scan inbox directories for pending items; count unprocessed files
3. convoys — Check convoy health: active count, throughput, last heartbeat
4. worker pools — Inspect utilization; detect zombie/stale workers
5. cleanup — Remove stale wisps, temp files, and orphaned lock files
6. health — Aggregate system health summary

Issue Classification

For every anomaly found, classify as one of:

• KNOWN_FIX — Matches a known failure pattern; auto-resolve immediately
• NOVEL — Unfamiliar issue; escalate to escalation inbox with diagnostic context

Known Fixes (auto-resolve without confirmation)

Pattern	Detection	Action
Stale wisps	Files in wisp dir older than 1 hour	Delete stale files
Zombie processes	Process in worker pool with no heartbeat >5 min	kill -9, respawn
Stale heartbeats	Heartbeat timestamp >10 min old	Reset heartbeat, mark worker idle
Broken symlinks	find returns symlinks with missing targets	Remove broken links
Empty inbox dirs	Inbox exists but contains zero files	Log as healthy, no action
Orphaned lock files	Lock file with no associated live PID	Remove lock file

Novel Issue Escalation

When a NOVEL issue is found:

1. Write a diagnostic record to ~/.claude/escalations/ with timestamp
2. Include: component name, error description, relevant file paths, suggested investigation steps
3. Continue patrol — do not halt on NOVEL issues

Output Format

After completing all 6 steps, output a single Markdown table:

| Component    | Status  | Last Checked        | Action Taken              |
|--------------|---------|---------------------|---------------------------|
| hooks        | OK      | 2026-02-28 14:32:01 | none                      |
| inboxes      | OK      | 2026-02-28 14:32:02 | 3 items processed         |
| convoys      | WARN    | 2026-02-28 14:32:03 | reset stale heartbeat     |
| worker pools | OK      | 2026-02-28 14:32:04 | none                      |
| cleanup      | OK      | 2026-02-28 14:32:05 | 2 stale wisps removed     |
| health       | OK      | 2026-02-28 14:32:06 | 1 NOVEL issue escalated   |

Status values: OK | WARN | ERROR | ESCALATED

Headless / Long-Running Behavior

• Never pause for confirmation between steps
• After 20 patrol cycles, output a clean handoff summary: total issues found, KNOWN_FIX resolved, NOVEL escalated, next recommended action
• If a step takes >30s, log timeout and move to next step — do not block the cycle
• Do not open editors, browsers, or interactive processes during patrol