Python: [Bug]: OpenAIChatClient fails with 'str' object has no attribute 'output' when provider returns non-OpenAI conformant/structured response

[Shaik-Vaheed-UP]

Description

What happened?

When using the OpenAI Responses client (OpenAIChatCompletionClient) with structured outputs
(response_format), if the backend returns a response body that is a plain string instead of
a structured OpenAI Response object, raw_response.parse() returns a str and the
framework crashes while iterating response.output:

Exception: ChatClientException: ("<class 'agent_framework_openai._chat_client.OpenAIChatClient'> service failed to complete the prompt: 'str' object has no attribute 'output'", AttributeError("'str' object has no attribute 'output'"))
  File "agent_framework_openai/_chat_client.py", line 727, in _get_response
    response = raw_response.parse()
  File "openai/_legacy_response.py", line 139, in parse
    parsed = self._options.post_parser(parsed)
  File "openai/resources/responses/responses.py", line 2942, in parser
    return parse_response(...)
  File "openai/lib/_parsing/_responses.py", line 61, in parse_response
    for output in response.output:
AttributeError: 'str' object has no attribute 'output'

What did you expect to happen?

A robust, actionable error reporting that the provider returned a non-OpenAI-compliant
response (ideally including the raw payload for debugging) — not an internal AttributeError
crash.

Steps to reproduce the issue

1. Configure the Python Agent Framework OpenAIChatClient against an OpenAI-compatible
   backend / proxy / gateway that may return a plain string or non-spec body (e.g. an HTML or
   text error page on failure/throttling).
2. Issue a request that uses structured output (options={"response_format": SomeModel}).
3. Cause the backend to return a non-conformant body.
4. The client crashes with the stack trace above.

Remediation suggestion

Defensively check the parsed response type in _inner_get_response / _parse_response_from_openai
before accessing response.output, and raise a targeted ChatClientException that includes
the raw payload when the response is not a valid Response / ParsedResponse. This mirrors the
hardening already applied elsewhere for OpenAI-compatible provider schema drift.

This is the same class of problem as #6234 (system_fingerprint), but on the Responses
(structured outputs) code path.

Code Sample

from agent_framework_openai import OpenAIChatCompletionClient
from pydantic import BaseModel


class _Response(BaseModel):
    summary: str


# Point to a proxy / self-hosted / custom OpenAI-compatible backend:
client = OpenAIChatClient(base_url="https://proxy-or-selfhosted-openai-endpoint/v1")

# Structured output request (response_format) as used in the validation agent:
response = await client.get_response(
    messages=messages,
    options={"response_format": _Response},
)
# If the backend replies with a plain string / HTML error body, the crash above occurs.

Error Messages / Stack Traces

Exception: ChatClientException: ("<class 'agent_framework_openai._chat_client.OpenAIChatClient'> service failed to complete the prompt: 'str' object has no attribute 'output'", AttributeError("'str' object has no attribute 'output'"))
File "/home/site/wwwroot/.python_packages/lib/site-packages/agent_framework_openai/_chat_client.py", line 727, in _get_response
response = raw_response.parse()
File "/home/site/wwwroot/.python_packages/lib/site-packages/openai/_legacy_response.py", line 139, in parse
parsed = self._options.post_parser(parsed)
File "/home/site/wwwroot/.python_packages/lib/site-packages/openai/resources/responses/responses.py", line 2942, in parser
return parse_response(
File "/home/site/wwwroot/.python_packages/lib/site-packages/openai/lib/_parsing/_responses.py", line 61, in parse_response
for output in response.output:
AttributeError: 'str' object has no attribute 'output'

Package Versions

agent-framework: 1.6.0

Python Version

Python 3.12

Additional Context

Related to the schema-mismatch bug tracked in #6234 (system_fingerprint). The crash occurs at agent_framework_openai/_chat_client.py line 727 (raw_response.parse()), then in the OpenAI SDK at openai/lib/_parsing/_responses.py line 61 (for output in response.output). Likely caused by a non-OpenAI / OpenAI-compatible backend returning a plain text or HTML body (e.g. on error/throttling). Suggest hardening with a type guard and surfacing the raw payload.

[Shaik-Vaheed-UP]

The issues is with OpenAIChatCompletionClient

[github-actions]

🤖 Automated triage reproduction notes (agent-authored — trust but verify)

Agent analysis

Repro: OpenAIChatClient._parse_response_from_openai() in python/packages/openai/agent_framework_openai/_chat_client.py:1971 crashes with AttributeError when the response parameter is a str instead of an OpenAI Response object. Trigger: raw_response.parse() (line 733) returns a non-Response type from a non-conformant backend. Minimal repro: mock client.responses.{parse,create} to return a MagicMock whose .parse() returns a plain string, then call client.get_response(); the AttributeError escapes unhandled because line 736 (_parse_response_from_openai call) is outside the try/except at lines 728-735.

• Failing test: python/packages/openai/tests/openai/test_repro_6235.py
• Files examined: python/packages/openai/agent_framework_openai/_chat_client.py, python/packages/openai/tests/openai/test_openai_chat_client.py, python/packages/openai/pyproject.toml
• Tests run: python/packages/openai/tests/openai/test_repro_6235.py::test_parse_returns_string_instead_of_response_object, python/packages/openai/tests/openai/test_repro_6235.py::test_parse_returns_string_on_create_path
• Reported version: 1.6.0
• Current version: 1.7.0