Hi OpenAI team and community,
I’m currently integrating the computer-use-preview model using the Responses API, and I’ve run into a blocking issue regarding user confirmation prompts, specifically when handling actions like “screenshot”.
The API sometimes returns a message like this:
{
"type": "reasoning",
"summary": [
{
"type": "summary_text",
"text": "Confirming action X with user"
}
]
}
And the associated computer_call object contains:
{
"type": "computer_call",
"action": {
"type": "screenshot"
},
"pending_safety_checks": []
}
The problem:
- There’s no pending_safety_checks to acknowledge, and
- The documentation doesn’t explain how to confirm or proceed from this “Confirming action with user” request.
This leads to the process being stuck, as I cannot send any acknowledgment or continuation signal back to the API. Unlike typical safety checks (which require acknowledged_safety_checks), this case offers no clear path forward.
If you attempt to repeat the request, it will end the task as completed, with no further computer_call issued, effectively halting execution.
Is there an undocumented parameter or specific response pattern I should be sending back to acknowledge this kind of prompt?
Would really appreciate guidance or an example on how to correctly handle this scenario.
As reference, this is an example of the complete output sent by the API in such a case:
[
{
"id": "resp_xxxxxxxxxxxxxxxxxxxxx",
"object": "response",
"created_at": 1748107803,
"status": "completed",
"background": false,
"error": null,
"incomplete_details": null,
"instructions": null,
"max_output_tokens": null,
"model": "computer-use-preview-2025-03-11",
"output": [
{
"id": "rs_xxxxxxxxxxxxxxxxxxxxx",
"type": "reasoning",
"summary": [
{
"type": "summary_text",
"text": "Preparing to confirm message input"
}
]
},
{
"id": "cu_xxxxxxxxxxxxxxxxxxxxx",
"type": "computer_call",
"status": "completed",
"action": {
"type": "screenshot"
},
"call_id": "call_xxxxxxxxxxxxxxxxxxxxx",
"pending_safety_checks": []
}
],
"parallel_tool_calls": true,
"previous_response_id": "resp_xxxxxxxxxxxxxxxxxxxxx",
"reasoning": {
"effort": "medium",
"generate_summary": "concise",
"summary": "concise"
},
"service_tier": "default",
"store": true,
"temperature": 1,
"text": {
"format": {
"type": "text"
}
},
"tool_choice": "auto",
"tools": [
{
"type": "computer_use_preview",
"display_height": 1127,
"display_width": 1717,
"environment": "mac"
}
],
"top_p": 1,
"truncation": "auto",
"usage": {
"input_tokens": 3995,
"input_tokens_details": {
"cached_tokens": 0
},
"output_tokens": 87,
"output_tokens_details": {
"reasoning_tokens": 0
},
"total_tokens": 4082
},
"user": null,
"metadata": {}
}
]
Thanks!