Python: Fix AG-UI approval resume at the protocol boundary

[moonbox3]

Motivation & Context

AG-UI approval resumes can replay a terminal function result before the approval response synthesized from the resume payload. Connector-specific filtering would scatter AG-UI protocol semantics across providers and can allow stale approval controls to be replayed. The fix belongs at the AG-UI protocol boundary so every connector receives the appropriate transcript.

Description & Review Guide

• What are the major changes? Validate approval responses against the server-owned pending registry, correlate request and interrupt aliases by call occurrence, reject stale, unknown, or cross-call responses, and remove only completed local approval controls before static execution and provider handoff. Hosted approvals retain trusted server metadata, expose an accepted-only decision schema, and remain provider-bound; deferred approvals remain available to in-run middleware.
• What is the impact of these changes? AG-UI approval resumes no longer re-execute completed local tools or send local approval controls to a raw chat client. Hosted and deferred tool flows keep their existing paths. There are no Ollama-specific changes; Ollama works through the corrected AG-UI transcript.
• What do you want reviewers to focus on? Occurrence and alias matching, the local-versus-hosted boundary, and the endpoint regressions that observe the raw provider input.

Related Issue

Fixes #7451

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. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and the title prefix in sync automatically.

[Copilot]

Pull request overview

Fixes an incompatibility between AG-UI’s synthesized approval-resume user messages and the Ollama connector by dropping control-only approval-resume messages during Ollama message formatting, and adds a regression test to ensure the rest of the transcript (user → assistant tool call → tool result) still reaches Ollama.

Changes:

• Skip role="user" messages that contain only function_approval_response content when formatting user messages for Ollama.
• Add a regression test validating the approval-resume message is omitted from the outgoing Ollama messages payload.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
python/packages/ollama/agent_framework_ollama/_chat_client.py	Drops control-only approval-resume user messages by returning an empty formatted message list.
python/packages/ollama/tests/test_ollama_chat_client.py	Adds a regression test asserting the approval-resume message is not forwarded to Ollama.

[github-actions]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/ag-ui/agent_framework_ag_ui
_agent_run.py	1319	112	91%	139, 199, 218, 265, 283, 299–300, 422, 475–476, 483, 592, 596, 598, 615, 642–643, 761, 773, 787, 791, 795, 798, 803, 808, 817, 820, 827–833, 866, 878, 889, 892, 928, 982–986, 1039, 1054, 1057, 1059, 1085, 1111–1113, 1173, 1175, 1177, 1180–1184, 1198, 1206–1211, 1220–1221, 1270–1273, 1284, 1292, 1324, 1339, 1365, 1377, 1407, 1411, 1414, 1416, 1457–1459, 1520, 1583, 1589–1590, 1829, 1837, 1867, 1871, 1916, 2118, 2246, 2306, 2323, 2361, 2467, 2495, 2503, 2505, 2508, 2514, 2569, 2572, 2582–2583, 2590, 2637
_run_common.py	601	41	93%	76, 151–152, 154, 156, 159, 165, 167, 181, 188, 212–213, 226, 245, 281, 325, 337, 339, 341, 344–348, 526, 806–807, 1131–1132, 1137, 1139–1141, 1150, 1158, 1170, 1172–1175, 1238
TOTAL	44657	4103	90%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
9130	34 💤	0 ❌	0 🔥	2m 30s ⏱️

[github-actions]

Automated Code Review

Reviewers: 5 | Confidence: 54% | Result: All clear

Reviewed: Correctness, Security Reliability, Test Coverage, Failure Modes, Design Approach

---

Automated review by moonbox3's agents

[github-actions]

Automated Code Review

Reviewers: 4 | Confidence: 35% | Result: All clear

Reviewed: Correctness, Security Reliability, Failure Modes, Design Approach

---

Automated review by moonbox3's agents