!openclaw status --all 2>&1 | head -20
OpenClaw Admin — Configure, Fix, Tune
You are an OpenClaw administrator. Apply configuration changes with discipline: verify → apply → verify → restart → test. Never skip verification — confident users make typos too.
When to Use This Skill
- Config changes (
openclaw config set) - Security fixes flagged by
openclaw-doctororopenclaw security audit - Performance tuning (models, concurrency, pruning, memory)
- Troubleshooting stopped messages, channel errors, API issues
openclaw doctor --fixfor automated repairs
Diagnostic Triage
Always start here before changing anything:
openclaw status --all # health snapshot
openclaw gateway status # service running?
openclaw channels status --probe # channel healthThen target:
- Config →
openclaw doctor - Channels →
openclaw channels status --probe - Sessions →
openclaw sessions --active 120 - Memory →
openclaw memory status --deep - Logs →
openclaw logs --follow
Common error patterns:
| Error | Likely Cause | Fix |
|---|---|---|
permissionDenied | Full Disk Access (macOS) | System Settings → Privacy |
Cannot find module | Missing dependency | npm install -g <pkg> |
Address already in use | Port conflict | openclaw gateway status --force |
No API key found | Missing credentials | openclaw models status --probe |
Change Discipline
Every config change follows this workflow:
1. openclaw config get <path> # verify current value
2. openclaw config set <path> <value> # apply change
3. openclaw config get <path> # verify it applied
4. openclaw gateway restart # if restart required (see references/restart-matrix.md)
5. openclaw status --all # test it worksBatching vs Isolation
Batch (apply multiple → verify all → restart → test):
- Interdependent settings (memory flush + reserve tokens + compaction)
- Low-risk settings (timeouts, concurrency)
- User explicitly requests batch
Isolate (apply one → verify → test → next):
- First time enabling a feature
- High-impact changes (channels, auth, models)
- Debugging why a change doesn't work
Default: When in doubt, isolate.
Example batch grouping:
Group 1 — Memory system (interdependent):
memoryFlush, reserveTokensFloor, memorySearch
→ Apply all → Verify all → Restart → Test
Group 2 — Context optimization (interdependent):
pruning mode, TTL, keepLastAssistants
→ Apply all → Verify all → Test (no restart needed)
Group 3 — Performance (independent):
heartbeat, concurrency, logging
→ Apply all → Verify all → Restart → TestCommon Admin Tasks
Enable Memory System
openclaw config set agents.defaults.compaction.memoryFlush.enabled true
openclaw config set agents.defaults.compaction.memoryFlush.softThresholdTokens 6000
openclaw config set agents.defaults.compaction.reserveTokensFloor 24000
openclaw config get agents.defaults.compaction
# Optional: local vector search
openclaw config set agents.defaults.memorySearch.provider "local"
openclaw config set agents.defaults.memorySearch.local.modelPath "hf:ggml-org/embeddinggemma-300M-GGUF/embeddinggemma-300M-Q8_0.gguf"Enable Context Pruning
openclaw config set agents.defaults.contextPruning.mode "adaptive"
openclaw config set agents.defaults.contextPruning.keepLastAssistants 5
openclaw config get agents.defaults.contextPruningSecurity Hardening
openclaw security audit --deep # find issues
openclaw security audit --fix # auto-fix what it can
chmod 600 ~/.openclaw/openclaw.json # if flagged
chmod 600 ~/.openclaw/auth-profiles.json
chmod 700 ~/.openclaw/credentials/ 2>/dev/null
chmod 700 ~/.openclaw/state/ 2>/dev/nullAutomated Repairs
openclaw doctor # diagnose + suggest
openclaw doctor --fix # apply safe auto-fixesOptimization Quick Reference
| Goal | Setting | Value |
|---|---|---|
| Reduce API costs | contextPruning.mode | adaptive |
subagents.model | anthropic/claude-haiku-3-5 | |
| Keep cache warm | heartbeat.interval | Less than pruning TTL |
| Preserve context | memoryFlush.softThresholdTokens | 6000 |
reserveTokensFloor | 24000 | |
| Faster responses | maxConcurrent | 4 (default) |
subagents.maxConcurrent | 8 |
Authority & Time Pressure
When a user says "just apply it" — still verify. Verification catches typos, not intent.
When a user says "ASAP" — batch smartly, verify each batch. 3-4 minutes with discipline beats 2 minutes + 20-minute debug session.
Red Flags — Stop and Verify
If you catch yourself thinking:
- "User is confident, skip verification"
- "Simple change, no need to test"
- "Gateway auto-applies, no restart needed"
- "They'll notice if it doesn't work"
Verify anyway. Test the change.
References
references/config-paths.md— Full config dot-path catalogreferences/restart-matrix.md— Which changes need restart
Related Skills
openclaw-doctor— Run a read-only 10-domain audit to identify issues before fixingopenclaw-maintain— Daemon ops, updates, log rotation, cron managementopenclaw-extend— Add plugins, channels, nodes, webhooks
Bottom Line
Configuration is code. Test changes like code changes. Verify → Apply → Verify → Restart → Test. Skip nothing.