Hosted `shell` Continuations Require Missing `shell_call_output`

Hey everyone, Thanks for the detailed report. Are you still seeing this issue?

If yes, could you share the x-request-id from a failed continuation request and the execution timestamp, including timezone? Please share the failed request ID, not a later retry that succeeded.

I tested the hosted shell flow with previous_response_id and wasn’t able to reproduce the failure: the first response included matching shell_call / shell_call_output items, and the follow-up completed successfully.

One workaround I verified is to avoid relying on stored history and replay the prior response output items explicitly, then reuse the returned shell container with container_reference. In other words: include the previous response.output in the next request’s input, append the next user message, and set the shell tool environment to the prior container_id. Hope this helps. Thank you!