Openclaw Self Heal

Autonomously diagnose and fix OpenClaw system errors affecting UI, channels, gateway, memory, or pairing, escalating only if needed.

Security Scan

The skill's instructions, file references, and commands are coherent with a self-healing OpenClaw utility — it asks for no extra credentials or installs and performs only expected diagnostic and restart actions.

Purpose & Capability

Name and description match the actions in SKILL.md and references: commands to check status, logs, restart the gateway, check port conflicts, inspect config and session files, and escalate for token/config edits are all normal for a self-heal tool.

Instruction Scope

Instructions reference system logs and OpenClaw-specific paths (~/ .openclaw, /tmp/openclaw, macOS DiagnosticReports) and include process control (restart, pkill, kill port owner) and an approval step (openclaw pair --approve). These are within scope for repair but include operations that can be destructive or change authorization state; the skill explicitly says to escalate before config edits or credential rotation.

Install Mechanism

Instruction-only skill with no install spec and no code files — nothing is written to disk by an installer, which reduces supply-chain risk.

Credentials

No environment variables, secrets, or external credentials are requested. The instructions do inspect local config and token state via CLI/logs (expected for this purpose) but do not request unrelated credentials or external service keys.

Persistence & Privilege

always is false and there is no install-time persistence. However the skill is allowed autonomous invocation (platform default) and the runtime instructions permit the agent to perform sensitive actions (process killing, restarts, and approving pairing). Autonomous invocation is not flagged by itself, but you should be aware the agent could perform these actions without additional prompts if allowed.

Assessment

This skill appears internally consistent for diagnosing and fixing OpenClaw components. Before enabling it consider: (1) Back up ~/.openclaw/config.json and session files so you can recover if a fix is mistaken. (2) If you don't want the agent to take autonomous destructive actions, disable autonomous invocation (prevent model-driven runs) or require user-invocation only. (3) Note the agent can run pkill, kill processes using ports, restart services, and run openclaw pair --approve — these can affect availability and authorization; require manual approval for credential or config edits. (4) Test the skill in a safe/staging environment first. (5) Monitor the skill's logs (memory/YYYY-MM-DD.md) to review what was changed after any run.

Diagnose and fix OpenClaw issues autonomously. Fix first, report after (unless destructive).

Quick Triage (run first)

openclaw status 2>&1
openclaw logs --limit 100 2>&1

Parse output for: error, warn, SIGTERM, failed, pairing required, disconnected, crash.

See references/diagnostics.md for symptom → fix mappings.

Fix Philosophy

1. Try the safe fix first — restart gateway, reload config, clear cache
2. Fix → verify → report — don't ask Jake until you've tried
3. Escalate only for: data loss risk, credential resets, config changes Jake must approve, or repeated failure after 2 attempts
4. Always log what you did to memory/YYYY-MM-DD.md

Common Commands

# Gateway control
openclaw gateway status
openclaw gateway restart
openclaw gateway stop && sleep 3 && openclaw gateway start

# Status + logs
openclaw status
openclaw status --deep
openclaw logs --limit 200
openclaw logs --follow   # use briefly, then kill

# Security audit
openclaw security audit

Verification

After any fix, confirm resolution:

openclaw status 2>&1 | grep -E "(Gateway|Channel|Error|CRITICAL|WARN)"

If still broken after restart, consult references/diagnostics.md for deeper fixes.

Reporting

After resolving: brief summary to Jake — what broke, what you did, current state. If you couldn't fix it: tell Jake what you found + what you tried, with the exact error.