Python: fix(core): conversation_id not updated in options during tool invocation loop

[moonbox3]

Motivation and Context

When using the OpenAI Responses API with tool calls, the framework would fail with:

No tool call found for function call output with call_id <call_id>

This occurred in checkpoint/resume scenarios with handoff workflows, but could affect any multi-turn tool invocation.

In _tools.py, both function_invocation_wrapper and streaming_function_invocation_wrapper update kwargs with the new conversation_id after each API response, but the options dict (which is passed to subsequent API calls) was not being updated.

The flow:

1. API call returns tool call + new response_id
2. Framework invokes tool locally
3. Next API call needs previous_response_id = new response_id to link the tool result
4. Bug: options["conversation_id"] still had the OLD value
5. API rejects: "that call_id doesn't exist in the response chain"

The fix:

Update options["conversation_id"] alongside kwargs when a new conversation_id is received:

 if response.conversation_id is not None:
     _update_conversation_id(kwargs, response.conversation_id)
     # Also update options dict so subsequent API calls use the new conversation_id
     if options is not None:
         options["conversation_id"] = response.conversation_id
     prepped_messages = []

Description

• Fix conversation_id not updated in options during tool invocation loop

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.

[Copilot]

Pull request overview

Fixes a tool-invocation loop bug where a newly returned conversation_id (from hosted/server-managed conversations such as OpenAI Responses) was applied to kwargs but not propagated into the options dict used for subsequent API calls, which could break multi-turn tool execution in checkpoint/resume workflows.

Changes:

• Update options["conversation_id"] whenever a response returns a new conversation_id (non-streaming + streaming tool loops).
• Add a regression test that verifies the updated conversation_id is used on the next iteration for both non-streaming and streaming flows.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
python/packages/core/agent_framework/_tools.py	Ensures conversation_id updates are applied to options between tool-loop iterations so subsequent API calls remain on the correct server-side response chain.
python/packages/core/tests/core/test_function_invocation_logic.py	Adds coverage proving the next API call receives the updated conversation_id after a tool-call response (streaming and non-streaming).

[markwallace-microsoft]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/core/agent_framework
_tools.py	771	81	89%	226, 272, 323, 325, 353, 523, 558–559, 661, 663, 683, 701, 715, 727, 732, 734, 741, 774, 845–847, 888, 910–938, 973, 981, 1222, 1559, 1646–1650, 1672–1673, 1789, 1791, 1856, 1948, 1954, 1998–1999, 2012–2013, 2056, 2140, 2220–2222, 2271, 2328–2329, 2336–2337
TOTAL	16366	1918	88%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
3990	221 💤	0 ❌	0 🔥	1m 13s ⏱️

[eavanvalkenburg]

Included in #3379