| name | system-troubleshoot |
| description | Zero-token diagnosis of common system issues. Run when things look stuck. |
| tags | ["ops","troubleshoot","health","diagnosis","zero-cost"] |
| version | 1.0.0 |
system-troubleshoot
Diagnose why the system isn't working. Zero token cost — pure bash + SQLite.
When to use
- Tasks stuck in pending (not promoting to in_progress)
- Executor seems dead or duplicated
- Robert says "workers aren't doing anything"
- Stability monitor detects tasks stuck >30 min
- Any time you're unsure about system health
How to run
bash /root/.openclaw/scripts/system-troubleshoot.sh
What it checks (9 checks)
- Executor running? (0 = dead, >1 = orphans)
- Pending tasks stuck? (quarantined agents, dead blocked_by)
- In-progress stuck >15 min? (killed executor left zombies)
- Concurrency full? (MAX_CONCURRENT=2 met, shows oldest)
- Gateway down? (OOM, config error)
- Bridge responding? (port conflict, crash)
- Codex health? (OAuth expired, rate limited)
- Orphan processes? (duplicates of any monitored process)
- Log spam? (executor log >10MB)
Output format
For each issue: PROBLEM → CAUSE → FIX (actionable command)
Summary: N issue(s) found
Self-improvement
This skill gets better each time someone discovers a new failure mode:
- Diagnose the issue manually
- Add a check to system-troubleshoot.sh
- Chart the failure pattern for future reference