Skip to main content
This runbook provides an end-to-end flow for running the iCommerce agent inside a sandbox.

​
1) Provision a Sandbox

Save the sandbox_id from the response.

​
2) Verify Connectivity

​
3) Execute Write Actions (Explicit)

​
4) Capture Outputs

  • Stream output from the execution response
  • Store logs and results for audit trails

​
5) Stop the Sandbox

Step 5 is the one that costs money if you skip it. A sandbox bills until it is stopped or its timeout expires, and a run that fails between step 3 and step 5 leaves one up — so stop it in a finally, or from a cleanup job that sweeps sandboxes older than your longest expected run.
Step 3 is deliberately explicit: write actions need --apply, and without it the agent answers questions and changes nothing. Keep it off while you are still reading the outputs from step 4 — a read-only rehearsal against real data is the cheapest way to find out the agent misunderstood the task.

​
Next steps

Run the agent in a sandbox

The narrative version of this runbook.

Sandbox API flow

The underlying create/execute/stop sequence.

Agent sessions

For work that outlives one sandbox.

Troubleshooting

When connectivity or auth fails.
Last modified on August 31, 2026