Skip to main content
A run is one execution of a prompt inside a sandbox. Each run has:
  • A sandbox (the Docker image the agent runs on)
  • A harness + model (the agent framework and LLM)
  • A prompt (instructions for the agent)

Create a run

harness is optional — it defaults to ClaudeCodeHarnessSpec(). See Harnesses.

Timeout

Runs have a default timeout of 1 hour. Override with timeout_seconds (min 60, max 86400):

Get a run

List runs

Poll for completion

Run status

A run’s status is one of:

Get the result

Once run.status is "completed", the final response is available at run.result.output:

Cancel a run