openclaw-doctor（OpenClaw doctor 搜索）

!openclaw status --all 2>&1 | head -20!openclaw gateway status 2>&1 | head -10

OpenClaw Doctor — Comprehensive Diagnostic Audit

You are an agent diagnosing another agent's infrastructure. Run a full 10-domain audit, classify every finding by severity, and produce a structured report with actionable fixes.

This skill is read-only — never apply fixes, only recommend them.

Quick Start

1. Track progress through all 10 domains using the checklist
2. Run each domain's commands, interpret output, classify findings
3. Consult references/config-reference.md for recommended values
4. Consult references/severity-rules.md for classification rules
5. Present the unified Health Report at the end

Audit Checklist

- [ ] Domain 1: Gateway Health
- [ ] Domain 2: Security
- [ ] Domain 3: Channels
- [ ] Domain 4: Auth & Models
- [ ] Domain 5: Memory System
- [ ] Domain 6: Context & Compaction
- [ ] Domain 7: Heartbeat & Cron
- [ ] Domain 8: Hooks & Skills
- [ ] Domain 9: Workspace Integrity
- [ ] Domain 10: System & Network

Audit Protocol

Run all 10 domains. Collect findings. Never stop early — the full picture matters.

Domain 1 — Gateway Health

openclaw status --all
openclaw gateway status
openclaw update status

Evaluate:

• Gateway service loaded, running, PID alive
• Latency < 500ms
• Version current vs latest (> 5 behind = WARN, not running = CRITICAL)
• Port 18789 responding, no "Address already in use"

Domain 2 — Security

openclaw security audit --deep
stat -f "%Lp %N" ~/.openclaw/openclaw.json
stat -f "%Lp %N" ~/.openclaw/auth-profiles.json
stat -f "%Lp %N" ~/.openclaw/credentials/ 2>/dev/null
stat -f "%Lp %N" ~/.openclaw/state/ 2>/dev/null

Evaluate:

• Security audit critical/warning/info counts
• Config files should be 600 (not 644 = CRITICAL)
• Credentials/state dirs should be 700
• API keys hardcoded in config vs env vars
• Gateway token auth enabled (no auth = WARN)

Domain 3 — Channels

openclaw channels status --probe

Evaluate:

• Per-channel: enabled / configured / running / probe passes
• Probe failure on configured channel = CRITICAL
• No channels at all = WARN
• DM policy open vs pairing (open = INFO)

Domain 4 — Auth & Models

openclaw models status --probe

Evaluate:

• Primary model probe succeeds (fail = CRITICAL)
• Fallback models configured (none = WARN)
• Image model configured (none = WARN)
• Sub-agent model set to cheaper model (same as primary = INFO)
• Single provider, no diversity = INFO

Domain 5 — Memory System

openclaw memory status --deep
openclaw config get agents.defaults.compaction.memoryFlush
openclaw config get agents.defaults.memorySearch

Evaluate:

• Index health: dirty flag, chunk count, file count
• Flush enabled with thresholds (disabled = WARN)
• Dirty index + 0 chunks = WARN (broken)
• Memory search provider configured (none = WARN)
• Agent name in index matches current agent (mismatch = WARN)

Domain 6 — Context & Compaction

openclaw config get agents.defaults.contextPruning
openclaw config get agents.defaults.compaction

Evaluate against references/config-reference.md:

• Pruning mode set (none = WARN; adaptive recommended)
• keepLastAssistants set (unset = INFO)
• reserveTokensFloor >= 20000 (< 20000 = WARN)
• memoryFlush.softThresholdTokens in 4000-8000 (outside = INFO)

Domain 7 — Heartbeat & Cron

openclaw config get agents.defaults.heartbeat
openclaw cron list
openclaw cron status

Read ~/.openclaw/workspace/HEARTBEAT.md to check if it has actual tasks.

Evaluate:

• Heartbeat enabled + HEARTBEAT.md empty = WARN (burning tokens)
• Heartbeat interval >= pruning TTL = WARN (cache expires before heartbeat)
• Cron scheduler not running when jobs exist = WARN
• No heartbeat / no cron = INFO (may be intentional)

Domain 8 — Hooks & Skills

openclaw hooks list
openclaw skills list
openclaw plugins list

Evaluate:

• Hooks: count ready vs error (errors = WARN)
• Skills: count ready vs blocked vs disabled
• Plugins: loaded vs error (errors = WARN)
• Many skills blocked by same missing dep = INFO

Domain 9 — Workspace Integrity

Check ~/.openclaw/workspace/ for required files:

Additional checks:

• IDENTITY.md filled in vs template placeholders (template = WARN)
• Config backup accumulation (~/.openclaw/openclaw.json.bak* > 5 = INFO)

Domain 10 — System & Network

tailscale status
tailscale serve status 2>/dev/null
launchctl list 2>/dev/null | grep openclaw
du -sh ~/.openclaw/
du -sh ~/.openclaw/logs/ 2>/dev/null

Evaluate:

• Tailscale daemon running, version match (mismatch = WARN)
• LaunchAgent loaded
• Log dir total size (> 10MB error log = WARN)
• Total .openclaw/ disk usage (> 1GB = WARN)

Report Template

After all 10 domains, present this:

## OpenClaw Health Report — {YYYY-MM-DD}

**Version:** {from status --all}  |  **Gateway:** {running/stopped}  |  **Uptime:** {if available}

### Summary
| Severity | Count |
|----------|-------|
| CRITICAL | N     |
| WARNING  | N     |
| INFO     | N     |
| PASS     | N     |

### Findings

#### CRITICAL
- [C1] {Domain}: {finding} — `{fix command}`

#### WARNING
- [W1] {Domain}: {finding} — `{fix command}`

#### INFO
- [I1] {Domain}: {finding} — {recommendation}

#### PASS
- {Domain}: All checks passed

### Value-Add Opportunities
- {opportunity} — {impact} — {effort estimate}

### Quick Fix Script

Review before running — generated from CRITICAL and WARNING findings

{fix commands, one per line, commented with finding ID}

Report rules:

• Every domain appears (findings or PASS)
• CRITICAL and WARNING include fix commands
• INFO includes recommendations
• Quick Fix Script only has CRITICAL + WARNING fixes
• Note when a fix should use openclaw-admin change discipline

Execution Guidelines

• Parallel where possible: Run independent commands together
• Platform awareness: On Linux use stat -c "%a %n" instead of stat -f "%Lp %N"
• Graceful failures: If a command fails, note as INFO and continue — never abort
• No mutations: Never run config set, --fix, gateway restart, or any write operation
• Cross-reference: Use references/config-reference.md and references/severity-rules.md

References

• references/config-reference.md — Recommended values, file permissions, CLI commands
• references/severity-rules.md — Classification rules for CRITICAL/WARNING/INFO/PASS

Related Skills

• openclaw-admin — Apply fixes using verify-apply-verify-restart-test discipline
• openclaw-maintain — Daemon ops, updates, cron, log rotation
• openclaw-extend — Add plugins, channels, nodes, webhooks