Verifying Your Setup
After completing the Getting Started steps, follow this guide to confirm the plugin is loaded and governance is active.
1. Check the plugin loaded
After restarting the gateway, check the OpenClaw logs for the plugin initialization message:
[openbox] plugin loaded
If you configured the LLM gateway, you should also see:
[openbox] LLM gateway started on http://127.0.0.1:18919/v1
If neither message appears, see Troubleshooting - Plugin not loading.
2. Check the LLM gateway is running
If you configured the LLM gateway, verify it is responding:
curl http://127.0.0.1:18919/health
Expected response:
{"status":"ok"}
If the health check fails, the gateway may not have started - check the logs for [openbox] LLM gateway server error messages. See Troubleshooting - Gateway not starting.
3. Run a tool call and check the logs
Start your agent and perform a simple action - for example, ask it to read a file. In the OpenClaw logs, you should see governance evaluation output:
[openbox] session registered verdict=allow
[openbox] tool=Read verdict=allow reason="" ms=120
This confirms:
- The plugin is intercepting tool calls
- OpenBox Core is reachable
- Your API key is valid
- The session was registered
If the verdict shows block, your policies in the OpenBox dashboard are actively governing the action.
4. Check the OpenBox dashboard
Open the OpenBox dashboard and navigate to the sessions view. You should see a new session entry corresponding to your agent run, with:
- Session timeline showing tool calls and LLM inferences
- Governance verdicts for each action
- OTel span data attached to activities
If no session appears in the dashboard, check that your openboxUrl and openboxApiKey are correct. See Troubleshooting - No events in dashboard.
5. Test a block verdict
To confirm governance enforcement is working end-to-end, create a test policy in the OpenBox dashboard that blocks a specific tool (e.g. block Bash tool calls). Then ask your agent to run a shell command.
You should see in the logs:
[openbox] tool=Bash verdict=block reason="Blocked by test policy" ms=95
And the agent should receive the block reason instead of executing the command.
Remove the test policy when you are done.
What to check if something is wrong
| Symptom | Likely cause | See |
|---|---|---|
No [openbox] plugin loaded in logs | Missing config, plugin not installed | Plugin not loading |
| Gateway health check fails | Port conflict, missing llmBaseUrl/llmApiKey | Gateway not starting |
verdict=allow but expected block | Policy not configured in dashboard | OpenBox Dashboard |
| No session in dashboard | Wrong openboxUrl or invalid API key | No events in dashboard |
| Tool calls proceed but no log output | Plugin not loaded or hooks not registered | Plugin not loading |