OpenClaw Reliability Pack

Single entrypoint for OpenClaw reliability triage.

Best for

• No-reply/channel routing incidents
• Gateway/runtime instability and tool hangs
• Pre-release or post-change regression decisions

Not for

• Feature implementation unrelated to reliability
• Broad refactors before baseline diagnosis
• Silent fixes without evidence and verification gate

60-second preflight

Before routing, capture:

• primary symptom,
• blast radius (who/what affected),
• when issue started,
• last change before failure,
• current severity (low/medium/high).

Router

Pick one path first, then execute it fully.

Path A: Channel issues

Use for:

• no replies
• listener/startup errors
• reply/quote/mention rendering breaks
• thread/topic misrouting
• duplicate sends

Run: openclaw-channel-doctor

Path B: Core runtime issues

Use for:

• openclaw status unreachable
• missing scope/operator.read
• gateway closed 1000
• hanging tool calls
• cron misbehavior
• session wedges

Run: openclaw-troubleshooter-pro

Path C: Change-risk validation

Use for:

• “safe to release?”
• “did this update regress quality?”
• “ship or rollback?”

Run: agent-regression-guard

Standard reliability flow

1. Classify symptom (A/B/C).
2. Run read-only baseline checks first.
3. Apply smallest safe fix.
4. Verify with explicit gate.
5. Return one status:
   • resolved
   • mitigated
   • blocked
   • rollback_recommended

Output contract

Always return:

• chosen path (A/B/C),
• commands run,
• key evidence,
• current risk level (low/medium/high),
• next action.

Escalation rule

Escalate when any remain true:

• repeated failure after restart,
• critical path still broken,
• uncertain root cause after baseline checks.

Escalation payload must include:

• symptom,
• exact commands run,
• key logs/errors (redacted),
• top 2 suspected causes,
• next highest-signal test.

Author

Vassiliy Lakhonin