3.10.5 Handle on-board tasks
Some tasks fit the board better—device state, on-board models, or on-board skills. The OpenClaw page exposes what is available based on device state.
Handing work to OpenClaw
After Moss routes a task on-board, you land on On-device Agent. Typical flow:
- Describe needs in AI Dock, e.g. “Check critical services on the board.”
- Moss validates requirements.
- If ready, Moss explains the plan and asks for confirmation.
- After you confirm, OpenClaw (or the relevant page) continues.
- Results return to Studio for follow-up or debugging.
When the network jitters
If plain SSH commands time out or flap, Studio shows the failure reason. Retry, stabilize the network, or ask Moss whether the On-device Agent path makes sense.
High-risk actions still require your confirmation.
Task output
After OpenClaw runs, key outcomes appear in Studio. For example:
- Inspection summaries
- Execution logs
- Failure reasons
- Next steps that need your confirmation
If the device is offline, Studio guides you to restore connection from the page state.
When not to delegate to OpenClaw
| Scenario | Reason |
|---|---|
| You explicitly say “do this on the PC” | Follow your instruction |
| Work is only local files | OpenClaw targets the board |
| Deep analysis needed | Outline with Moss first |
| Quick win | Finish in Terminal or Workbench |