Python: Responses API: missing status field on function_call / function_call_output items

[ltwlf]

When using AG-UI client-side tools with AzureOpenAIResponsesClient, follow-up requests that include tool call history fail with 400 Bad Request because the Responses API requires a status field on function_call and function_call_output items.

The root cause is in RawOpenAIResponsesClient._prepare_content_for_openai:

• function_call items get "status": None (line ~1168) — API rejects null
• function_call_output items omit status entirely (line ~1194) — API requires it

The API expects status to be "in_progress", "completed", or "incomplete".

This only surfaces with AG-UI client-side (declaration-only) tools because server-side tools use previous_response_id chaining and never re-serialize tool history as input.

Workaround: monkey-patch to add "status": "completed" on both item types.

Versions: agent-framework-core 1.0.0rc3, agent-framework-ag-ui 1.0.0b260304, Python 3.13

[shengkwan]

@ltwlf I also experienced the same issue, details as follows:

Title

_prepare_content_for_openai hardcodes status: None for function_call items, causing Azure OpenAI Responses API 400 error

---

Description

Summary

When using OpenAIResponsesClient (or AzureOpenAIResponsesClient) with conversation history that contains function_call content items, the _prepare_content_for_openai method in _responses_client.py hardcodes "status": None in the serialized payload. The Azure OpenAI Responses API rejects this with a 400 Bad Request because status must be a string — one of "in_progress", "completed", or "incomplete".

Environment

• agent-framework version: ==1.0.0rc1 (latest beta at time of report)
• openai SDK version: (as installed with the above)
• Python version: 3.13.3
• Platform: Windows 11
• Azure OpenAI endpoint: Azure AI Foundry (Responses API v1)

Steps to Reproduce

1. Create a CustomHistoryProvider (implementing BaseHistoryProvider) that returns Message objects containing Content.from_function_call() items with additional_properties={"status": "completed"}.
2. Start a conversation that triggers tool/function calls, so the conversation history stored in the DB includes function_call + function_result pairs.
3. On the next turn, the agent loads the conversation history via the history provider and sends it to the Azure OpenAI Responses API.
4. The API returns a 400 Bad Request.

Error Message

openai.BadRequestError: Error code: 400 - {
  'error': {
    'code': 'invalid_payload',
    'message': 'Invalid payload',
    'type': 'invalid_request_error',
    'details': [
      {
        'code': 'ValidationError',
        'message': 'type: Value is "null" but should be "string"',
        'param': '/input/37/status',
        'type': 'error'
      },
      {
        'code': 'ValidationError',
        'message': 'const: Expected "in_progress"',
        'param': '/input/37/status',
        'type': 'error'
      },
      {
        'code': 'ValidationError',
        'message': 'const: Expected "completed"',
        'param': '/input/37/status',
        'type': 'error'
      },
      {
        'code': 'ValidationError',
        'message': 'const: Expected "incomplete"',
        'param': '/input/37/status',
        'type': 'error'
      }
    ]
  }
}

Root Cause

In agent_framework/openai/_responses_client.py, the _prepare_content_for_openai method handles function_call content at approximately line 1024:

case "function_call":
    if not content.call_id:
        logger.warning(f"FunctionCallContent missing call_id for function '{content.name}'")
        return {}
    fc_id = call_id_to_id.get(content.call_id, content.call_id)
    if not fc_id.startswith("fc_"):
        fc_id = f"fc_{fc_id}"
    return {
        "call_id": content.call_id,
        "id": fc_id,
        "type": "function_call",
        "name": content.name,
        "arguments": content.arguments,
        "status": None,          # <--- BUG: Always None
    }

The status field is hardcoded to None. However, the Azure OpenAI Responses API requires status to be a string with one of three valid values: "in_progress", "completed", or "incomplete".

Inconsistency with text_reasoning handling: For text_reasoning content items, the method correctly reads the status from additional_properties:

case "text_reasoning":
    ret: dict[str, Any] = {"type": "reasoning", "summary": []}
    ...
    props: dict[str, Any] | None = getattr(content, "additional_properties", None)
    if props:
        if status := props.get("status"):
            ret["status"] = status    # <--- Correctly reads from additional_properties

This same pattern should be applied to function_call items.