When a Tool Times Out, Don’t Guess and Retry

A tool call can time out after the external action has already happened. Retrying immediately may repeat the action; treating the timeout as failure may leave the outcome unclear. The safer design is to make uncertainty a visible state.

For a workflow that changes something outside the system, try this sequence:

1. Mark the action as submitted.
2. If confirmation arrives, record it as complete.
3. If the call times out, mark the result unknown—not failed.
4. Check for the existing action before retrying. If it can’t be verified, route it to a named owner.

A useful test simulates the action succeeding while its response is lost. The expected result: no blind second attempt, and a clear recovery path. This doesn’t guarantee reliable execution; it makes the uncertain handoff easier to inspect and resolve.