Python: Execute every approved tool in a parallel AG-UI HITL batch

[Ahmed Hassan (AAH20)]

Motivation & Context

AG-UI HITL resume currently executes only the first approved tool in a parallel always_require batch. CopilotKit and other AG-UI clients synthesize one user message per interrupt, so the second (and later) approval-control messages look like a new user turn. History sanitization then injects a synthetic "Tool execution skipped - user provided follow-up message" result for leftover gated siblings, and ID-less dedup can collapse distinct approval-control messages that hash the same str(Content).

This fixes the silently dropped calls so a complete resume for a parallel gated batch executes every accepted tool once. Rejected siblings still do not execute. Real follow-up user text still skips leftover pending calls. Queued one-at-a-time ToolApprovalMiddleware behavior and never-require mixed-batch siblings are unchanged.

Description & Review Guide

• What are the major changes?
  • _sanitize_tool_history no longer treats consecutive approval-control user messages as follow-up that abandons leftover gated call_ids.
  • _deduplicate_messages keys ID-less approval-control messages by (approval_id, call_id) instead of hashing str(Content).
  • Spec matrix, package protocol notes, unit tests, and an AG-UI endpoint regression cover the parallel gated batch.
• What is the impact of these changes?
  • A complete AG-UI resume with two (or more) accepted interrupts from one assistant tool batch now executes every approved local tool and persists both terminal results.
  • Mixed accept/reject consecutive approval-controls reach _resolve_approval_responses without skip injection.
  • Follow-up user text after a partial approval still injects skip for leftover siblings.
• What do you want reviewers to focus on?
  • The sanitizer exception is limited to approval-control-only user messages (no follow-up text). Automated triage originally pointed at _local_approval_content_ids_to_remove; calling _resolve_approval_responses directly with unsanitized responses already executes both calls, so this change stays in adapters rather than the occurrence matcher.
  • Confirm queued ToolApprovalMiddleware one-at-a-time surfacing and never-require sibling release still pass.

There is no other open PR for this issue.

This area is marked as function-calling-loop core-team ownership in python/AGENTS.md. The spec matrix row and extra AG-UI validation commands required by that document are included in the change.

Related Issue

Fixes #7569

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 title prefix in sync automatically.

If this parallel HITL drop is showing up in a live agent deployment, A2Z SOC’s Instant Audit ($499) and consultation are available for production agent-action gating reviews.

Made with Cursor

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[github-actions]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/ag-ui/agent_framework_ag_ui
_message_adapters.py	656	52	92%	93–94, 101, 107, 233–234, 243–245, 248, 251–255, 257–262, 265, 274–276, 279–282, 318, 490, 611–613, 670, 673, 675, 678, 681, 697, 714, 736, 836, 852–853, 924, 946, 1016, 1051–1052, 1120, 1163
TOTAL	46853	4364	90%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
9473	36 💤	0 ❌	0 🔥	2m 11s ⏱️

[Evan Mattson (moonbox3)]

Thanks for working on this. Please also have a look at the failing CI/CD checks.

[Evan Mattson (moonbox3)]

Looks like CI/CD is okay now after a merge from main.

[Evan Mattson (moonbox3)]

/review

[Evan Mattson (moonbox3)]

Ahmed Hassan (@AAH20) there's no need to update the 004-python-function-call-loop.md doc. We can revert the change, please.

[Evan Mattson (moonbox3)]

Please re-open when ready to address the feedback / remaining file conflict.