Security.RunDevelopers
v1 PreviewOpenAPI ↗
API & MCP reference/Fix a failed or stuck run
HOW-TO GUIDES

Fix a failed or stuck run

Find the failing step before retrying or changing the command.

A request is rejected before a run starts

  • 401: check the API URL and bearer token.
  • 409: a reused idempotency key may have a different body, or an object is in the wrong state.
  • 422: inspect the validation response. Common causes are an unknown command, an absolute file path, or a deadline above 120 seconds.
  • 429: wait for Retry-After, then retry the same request and key.
  • 503: inspect the selected tool’s readiness and ask the operator to repair the missing pack or dependency.

The run is failed

Read its error, native exit code and stderr. Check filenames first: a staged src/check.py is not the same path as check.py. Check the chosen command against the tool’s installed entrypoints.

The run is timed_out

Read any retained output to see how far it got. Reduce the target set or input size, lower the tool’s concurrency, or raise the deadline up to 120 seconds. Starting a new intended run requires a new idempotency key.

The result is outcome_unknown

Do not assume the command did not run. Review the run and retained evidence before repeating a side-effecting operation. Retrying a submission with the same key retrieves the existing run rather than replaying it.

Cleanup is not destroyed

Ask the operator to check termination and reconciliation. Do not treat cleanup failure as ordinary successful completion.

Cancel a run you no longer need

Send the cancel request, then retrieve the run until it is terminal. Partial output may still be available.

Search documentation, capabilities and native tools.

Copy this page

Your browser couldn’t copy automatically. Copy the selected Markdown below.