Python: (ag-ui): fix approval payloads being re-processed on subsequent conversation turns

[moonbox3]

Motivation and Context

Fixes a bug where approval payloads in ag-ui were being re-executed on subsequent conversation turns, causing duplicate tool invocations and stale approvals to be re-processed.

When a user approves a tool call in ag-ui:

1. The approval payload ({"accepted": true}) is processed and the tool executes
2. The snapshot sent back to CopilotKit still contains the raw approval payload
3. On the next turn, CopilotKit re-sends the full history including the approval
4. The adapter re-detects the approval and re-executes the tool

This caused tools to run multiple times and approval history to become corrupted.

Description

• Closes Python: [AG-UI]Tool execution repeats after Human-in-the-loop approval #4075

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/ag-ui/agent_framework_ag_ui
_agent_run.py	434	86	80%	148–155, 194–195, 202, 299, 311, 315, 317–318, 334, 361–362, 417–419, 431–433, 537, 542–543, 545, 649, 651–652, 705, 720–721, 728, 777–779, 796, 802, 810, 812, 820, 850–856, 859–862, 864–873, 876, 884–887, 894, 897–898, 903, 909–911, 915, 920–924, 950–952
TOTAL	21837	2767	87%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
4562	244 💤	0 ❌	0 🔥	1m 19s ⏱️

[Copilot]

Pull request overview

Fixes AG-UI human-in-the-loop approval payloads being re-processed on subsequent turns by cleaning approval JSON payloads out of the outgoing snapshot and replacing them with the executed tool result, preventing duplicate tool invocations when CopilotKit replays history.

Changes:

• Add _clean_resolved_approvals_from_snapshot to rewrite tool-role snapshot messages containing approval payloads into the corresponding tool result.
• Invoke snapshot cleaning after _resolve_approval_responses in run_agent_stream as a defense-in-depth measure.
• Add regression tests covering snapshot cleaning, “no-op” behavior when no approvals exist, prevention of stale approval reprocessing, and ensuring fresh approvals still get processed.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
python/packages/ag-ui/agent_framework_ag_ui/_agent_run.py	Adds snapshot-cleaning logic and calls it after approval resolution to prevent stale approvals from being replayed.
python/packages/ag-ui/tests/ag_ui/test_message_hygiene.py	Adds tests validating snapshot cleaning and that cleaned snapshots don’t re-trigger approvals on later turns.
python/packages/ag-ui/tests/ag_ui/test_message_adapters.py	Adds a test ensuring a fresh approval still produces a function_approval_response.