Python: [Bug]: AzureAIProjectAgentProvider Foundry agents ignore / don’t accept response_format at runtime (agent.run)

[ekinsenler]

Description

When using AzureAIProjectAgentProvider with Azure AI Foundry hosted agents, I can set response_format when creating the agent, but I cannot override response_format per-run. agent.run(..., response_format=...) is ignored. This differs from other Agent Framework flows where per-call structured output is possible.

Code Sample

import asyncio, os
from typing import List
from pydantic import BaseModel
from azure.ai.projects.aio import AIProjectClient
from azure.identity import InteractiveBrowserCredential
from azure.core.exceptions import ResourceNotFoundError
from agent_framework.azure import AzureAIProjectAgentProvider

class SchemaA(BaseModel):
    title: str
    summary: str

class SchemaB(BaseModel):
    topic: str
    bullets: List[str]

async def main():
    cred = InteractiveBrowserCredential()
    async with AIProjectClient(
        endpoint=os.environ["AZURE_AI_FOUNDRY_PROJECT_ENDPOINT"],
        credential=cred,
    ) as pc:
        provider = AzureAIProjectAgentProvider(project_client=pc)
        name = "repro-runtime-response-format"
        try:
            agent = await provider.get_agent(name=name)
        except ResourceNotFoundError:
            agent = await provider.create_agent(
                name=name,
                model="gpt-5.2",
                description="Repro: runtime response_format override",
                instructions="Return JSON matching the requested schema.",
                response_format=SchemaA,  # works
            )

        prompt = "Summarize: observability for LLM apps. Add key points."
        r1 = await agent.run(input=prompt)  # follows SchemaA
        r2 = await agent.run(input=prompt, response_format=SchemaB)  # expected SchemaB
        print("SchemaA:", r1)
        print("SchemaB override:", r2)

if __name__ == "__main__":
    asyncio.run(main())

Error Messages / Stack Traces

Package Versions

agent-framework: 1.0.0b260114

Python Version

Python 3.13

Additional Context

No response

[moonbox3]

This should be fixed as of #3192.

[moonbox3]

Based on the new way to handle options, one would pass in the response_format like:

r2 = await agent.run(input=prompt, options={"response_format": SchemaB})

[ekinsenler]

Update / finding: I tested your suggestion to pass run-time schema via options — “Based on the new way to handle options, one would pass in the response_format like: r2 = await agent.run(input=prompt, options={"response_format": SchemaB})” — using the repro below. The run-time response_format is still ignored for Foundry agents created via AzureAIProjectAgentProvider. The agent is created with response_format=SchemaA and r1 correctly conforms to SchemaA. However, r2 continues to conform to SchemaA even when attempting to override to SchemaB (both via response_format=SchemaB and via options={"response_format": SchemaB}), instead of returning {topic, bullets} as expected.

prompt = "Summarize: observability for LLM apps. Add key points."
r1 = await agent.run(input=prompt)  # follows SchemaA (title, summary)
r2 = await agent.run(input=prompt, response_format=SchemaB)  # ignored
r3 = await agent.run(input=prompt, options={"response_format": SchemaB})  # also ignored
print("SchemaA:", r1)
print("SchemaB override:", r2)
print("SchemaB override (options):", r3)

This strongly suggests the provider/agent wrapper is not applying or forwarding the run-time response_format override on this code path.