Python: fix(ag-ui): clean confirm_changes approval payloads in snapshot

[Henry Su (hsusul)]

Motivation & Context

In AG-UI, when human-in-the-loop tool execution is approved via the confirm_changes flow (e.g., CopilotKit useHumanInTheLoop), the approval response payload ({"accepted": true}) remains in snapshot_messages because _clean_resolved_approvals_from_snapshot looks up tool results in result_by_call_id using snap_msg.toolCallId. For confirm_changes synthetic tool calls, toolCallId is the confirm_id, which does not match the original tool call ID in result_by_call_id.

As a result, the raw approval JSON payload persists in snapshot_messages and the original tool call result is omitted. On subsequent turns, clients that reconcile UI state from MESSAGES_SNAPSHOT (such as CopilotKit) see an un-resulted tool call and revert the completed tool chip back to "Running" / "in progress".

Description & Review Guide

• What are the major changes?
  • Updated _clean_resolved_approvals_from_snapshot in agent_framework_ag_ui/_agent_run.py so that when toolCallId does not directly match result_by_call_id (e.g. confirm_changes synthetic approval tool calls), it still cleans the {"accepted": ...} approval payload and replaces it with the executed tool result (or confirmation summary).
  • Added unit test suite in tests/ag_ui/test_confirm_changes_snapshot.py verifying that confirm_changes approval and rejection payloads are cleaned up and replaced with actual tool results in snapshot messages.
• What is the impact of these changes?
  • Tool chips in AG-UI clients (e.g. CopilotKit) remain in the completed state after human approval rather than reverting to "Running" on subsequent turns.
• What do you want reviewers to focus on?
  • The fallback logic in _clean_resolved_approvals_from_snapshot for matching synthetic approval tool calls when tool_call_id is not present directly in result_by_call_id.

Related Issue

Fixes #6828

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR is linked to an issue and there is no other open PR for this issue (see Related Issue above).
• This is not a breaking change.

[Copilot]

Pull request overview

This PR fixes AG-UI’s confirm_changes human-in-the-loop flow so that end-of-turn MESSAGES_SNAPSHOT no longer retains the raw approval payload ({"accepted": ...}) and instead preserves the executed tool result, preventing clients (e.g., CopilotKit) from reverting completed tool chips back to “Running” on subsequent turns.

Changes:

• Updated _clean_resolved_approvals_from_snapshot to handle confirm_changes synthetic tool call IDs when replacing approval payloads in snapshot tool messages.
• Added new unit tests to validate snapshot cleaning behavior for confirm_changes approval and rejection cases.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
python/packages/ag-ui/agent_framework_ag_ui/_agent_run.py	Adjusts snapshot post-processing to replace approval JSON payloads with tool results, including a fallback path for confirm_changes.
python/packages/ag-ui/tests/ag_ui/test_confirm_changes_snapshot.py	Adds tests ensuring confirm_changes approval/rejection payloads are cleaned and replaced appropriately in snapshot messages.

[Eduard van Valkenburg (eavanvalkenburg)]

Henry Su (@hsusul), thank you for the confirm_changes snapshot cleanup work. The consolidated draft PR #7345 incorporates the fix while correlating the synthetic confirmation back to its original function_call_id so unrelated results are never joined. Closing this PR in favor of the consolidated work.