Python: Filter conversation_id when passing kwargs to agent as tool

[dmytrostruk]

Motivation and Context

Resolves: #3101

There was a change added recently that enabled kwargs propagation within as_tool() logic for user-defined runtime context (e.g. api_token, session_id etc):

agent-framework/python/packages/core/agent_framework/_agents.py

Lines 468 to 478 in 9e3b2fa

	async def agent_wrapper(**kwargs: Any) -> str:
	"""Wrapper function that calls the agent."""
	# Extract the input from kwargs using the specified arg_name
	input_text = kwargs.get(arg_name, "")
	# Forward all kwargs except the arg_name to support runtime context propagation
	forwarded_kwargs = {k: v for k, v in kwargs.items() if k != arg_name}
	if stream_callback is None:
	# Use non-streaming mode
	return (await self.run(input_text, **forwarded_kwargs)).text

When a parent agent delegates to a nested agent via as_tool(), the parent's conversation_id is being forwarded. The conversation_id is set in kwargs during function invocation loop:

agent-framework/python/packages/core/agent_framework/_tools.py

Line 1743 in 9e3b2fa

def _update_conversation_id(kwargs: dict[str, Any], conversation_id: str | None) -> None:

The nested agent accepts conversation_id and attempts to perform API calls on the parent's conversation, which is in a "waiting for tool outputs" state, resulting in:

Error code: 400 - {'error': {'message': 'No tool output found for function call call_xxx'}}

This PR excludes conversation_id from forwarded kwargs in as_tool() method, allowing nested agent to run a new conversation.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• All unit tests pass, and I have added new tests where possible
• Is this a breaking change? If yes, add "[BREAKING]" prefix to the title of the PR.

[markwallace-microsoft]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/core/agent_framework
_agents.py	305	55	81%	51, 90, 96, 99, 102, 404–406, 452, 506, 649, 818, 821–823, 940–943, 945, 948–950, 1051, 1092, 1094, 1103–1108, 1114, 1116, 1126–1127, 1134, 1136–1137, 1145–1149, 1157–1158, 1160, 1165, 1167, 1201, 1246–1247, 1249, 1251, 1262
TOTAL	17429	2663	84%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
3149	213 💤	0 ❌	0 🔥	1m 6s ⏱️

[Copilot]

Pull request overview

This PR fixes a bug where conversation_id was being forwarded from parent agents to nested agents when using the agent-as-tool pattern, causing "No tool output found" API errors. The fix excludes conversation_id from kwargs propagation in the as_tool() method, allowing nested agents to run their own conversations independently.

Changes:

• Modified as_tool() in _agents.py to filter out conversation_id from forwarded kwargs
• Added a test case to verify conversation_id is not forwarded to sub-agents
• Added two new sample files demonstrating the agent-as-tool pattern for OpenAI and Azure AI

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
python/packages/core/agent_framework/_agents.py	Updated kwargs filtering to exclude both arg_name and conversation_id from forwarded kwargs
python/packages/core/tests/core/test_as_tool_kwargs_propagation.py	Added test case verifying conversation_id exclusion while confirming other kwargs are still forwarded
python/samples/getting_started/agents/openai/openai_responses_client_with_agent_as_tool.py	New sample demonstrating agent-as-tool pattern with OpenAI Responses Client
python/samples/getting_started/agents/openai/README.md	Added entry for the new OpenAI agent-as-tool sample
python/samples/getting_started/agents/azure_ai/azure_ai_with_agent_as_tool.py	New sample demonstrating agent-as-tool pattern with Azure AI
python/samples/getting_started/agents/azure_ai/README.md	Added entry for the new Azure AI agent-as-tool sample