Python: Fix stateless replay of reasoning-paired tool calls by moonbox3 · Pull Request #7233 · microsoft/agent-framework

moonbox3 · 2026-07-21T03:28:18Z

Motivation & Context

Reasoning-model responses can bind a reasoning item to a tool-call item. During stateless replay, Agent Framework strips response-scoped reasoning items, but the client-side MCP path continued to serialize the paired function_call and function_call_output. Foundry rejects that orphaned call with HTTP 400 in multi-agent workflows. The earlier fix in #6907 covered hosted mcp_call items but not client-side MCP tools represented as function calls.

Description & Review Guide

What are the major changes? Extend stateless Responses replay sanitization to identify both hosted MCP and client-side function calls paired with reasoning. Completed historical function-call groups are removed atomically with their results, while active function loops remain intact. When middleware terminates a function loop before a follow-up assistant response, core records a durable marker on every result in the terminated batch and the OpenAI serializer uses it to remove the reasoning-paired calls and results atomically. Add deterministic end-to-end regression coverage through WorkflowBuilder, FoundryAgent, AgentExecutor, Responses SDK serialization, and Foundry hosting, including the middleware-terminated path.
What is the impact of these changes? Multi-turn and multi-agent reasoning workflows no longer send orphaned function_call or mcp_call items when service-side storage is unavailable. Existing non-reasoning tool history and active client-side function loops retain their current behavior.
What do you want reviewers to focus on? Please focus on the completed-function-loop detection in _tool_call_ids_paired_with_reasoning, the batch-wide middleware-termination marker, and the public workflow regression that distinguishes active tool execution from historical cross-agent replay.

Related Issue

Fixes #6074

No other open PR exists for this issue. This completes the client-side function-call path that remained after #6907.

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.

github-actions · 2026-07-21T03:31:53Z

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/core/agent_framework
_types.py	1232	97	92%	61, 70–71, 125, 134, 153, 155, 159, 163, 165, 167, 169, 187, 191, 217, 239, 244, 249, 253, 283, 709–710, 891–892, 1350, 1425, 1460, 1480, 1490, 1548, 1682–1684, 1969–1974, 1999, 2054, 2059, 2069, 2077, 2084–2088, 2106, 2179, 2192, 2197, 2310, 2333, 2594, 2618, 2717, 2898–2899, 3001, 3230, 3283, 3302, 3341, 3352, 3354–3358, 3360, 3363–3371, 3381, 3470, 3607, 3612, 3617, 3622, 3626, 3712–3714, 3743, 3831–3835
packages/foundry/agent_framework_foundry
_agent.py	261	41	84%	119, 122, 244–245, 249–251, 256–259, 358, 466–467, 484, 491, 494, 505–517, 521–522, 720–721, 724, 750, 760, 782, 870, 875, 879
packages/foundry_hosting/agent_framework_foundry_hosting
_responses.py	856	95	88%	186–189, 276, 281, 305, 360–361, 375, 378–379, 427–428, 438, 483, 515–517, 520–521, 532, 575, 589, 664, 667, 673, 675, 747, 834–835, 953, 1183, 1196, 1687–1689, 1691, 1738–1739, 1741–1742, 1744–1745, 1747–1748, 1753, 1762, 1765–1767, 1769, 1779, 1783, 1796, 1887–1888, 1903–1904, 1906, 1911–1915, 1917, 1924–1925, 1927–1928, 1934, 1936–1940, 1947, 1953, 1975, 1981, 1987, 1989, 1991–1994, 2002, 2004, 2065–2067, 2081–2082, 2085–2086
packages/openai/agent_framework_openai
_chat_client.py	1369	117	91%	313, 326, 676–680, 688–691, 697–701, 751–758, 760–762, 769–771, 829, 837, 860, 1077, 1136, 1138, 1140, 1142, 1208, 1222, 1302, 1312, 1317, 1360, 1482–1483, 1498, 1763, 1870, 1875–1876, 1959, 1969, 1996, 2002, 2012, 2018, 2023, 2029, 2034–2035, 2115, 2159, 2162–2165, 2179, 2189–2190, 2202, 2244, 2309, 2326, 2329, 2356–2358, 2397, 2414, 2417, 2479, 2486, 2523–2524, 2559, 2597–2598, 2616–2617, 2789–2790, 2808, 2894–2902, 3080, 3095, 3145, 3155–3157, 3184–3186, 3196–3197, 3203, 3218, 3351–3352
TOTAL	45262	5070	88%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
9255	33 💤	0 ❌	0 🔥	2m 10s ⏱️

github-actions

Automated Code Review

Reviewers: 5 | Confidence: 80% | Result: All clear

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

Automated review by moonbox3's agents

Copilot

Pull request overview

This PR fixes stateless replay for reasoning-model Responses by ensuring tool-call history that was originally paired with server-scoped reasoning items is not replayed “orphaned” (which Foundry rejects with HTTP 400), covering both hosted MCP calls and client-side function calls across workflow/multi-agent boundaries.

Changes:

Extend stateless replay sanitization to identify tool-call IDs paired with text_reasoning, including completed client-side function_call loops (call + result) and hosted MCP calls.
Update OpenAI chat client serialization behavior/tests so completed reasoning-paired function-call groups are removed when service-side storage is unavailable, while active tool loops remain unchanged.
Add deterministic end-to-end regression tests covering workflow cross-agent replay (FoundryAgent/WorkflowBuilder) and hosted server replay (ResponsesHostServer).

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
python/packages/openai/agent_framework_openai/_chat_client.py	Drops reasoning-paired tool-call items atomically during stateless replay, including completed function-call loops.
python/packages/openai/tests/openai/test_openai_chat_client.py	Updates unit coverage to assert completed reasoning-paired function-call history is stripped without storage, while active loops remain.
python/packages/foundry/tests/foundry/test_foundry_agent.py	Adds deterministic workflow regression to ensure cross-agent replay doesn’t send orphaned `function_call` items.
python/packages/foundry_hosting/tests/test_responses_int.py	Adds deterministic hosted-server regression ensuring second turn doesn’t replay `mcp_call` without its paired reasoning item.

github-actions

Automated Code Review

Reviewers: 5 | Confidence: 87% | Result: All clear

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

Automated review by moonbox3's agents

TaoChenOSU · 2026-07-21T20:11:55Z

Why did we choose to drop the function call and result pair instead of replaying the reasonings?

Key decisions: - Request encrypted reasoning on client-managed Responses calls while preserving caller include values. - Store encrypted payloads in Content.protected_data and reconstruct one provider reasoning item per reasoning id. - Replay active and completed function call/result groups; retain continuation-owned history behavior and the existing orphan-safe MCP path. Files changed: - python/packages/openai/agent_framework_openai/_chat_client.py - python/packages/openai/tests/openai/test_openai_chat_client.py Next iteration: - Extend encrypted reasoning preservation to streaming and framework serialization boundaries. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Key decisions: - Capture encrypted reasoning from terminal streamed output items in Content.protected_data. - Preserve summary and private reasoning as distinct framework contents while reconstructing one provider reasoning item per id. - Prove replay after Message JSON and workflow checkpoint round trips, including encrypted-only and completed function groups. Files changed: - python/packages/core/agent_framework/_types.py - python/packages/openai/agent_framework_openai/_chat_client.py - python/packages/openai/tests/openai/test_openai_chat_client.py Next iteration: - Extend lossless stateless reasoning replay to hosted MCP call/output groups. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Key decisions: - Preserve hosted MCP call/output groups in client-managed history instead of deleting them when reasoning cannot be reconstructed. - Keep call/result coalescing and orphan-result exclusion intact, while retaining continuation-owned duplicate avoidance. - Cover completed, active, and multi-call reasoning groups plus the public outgoing request boundary. Files changed: - python/packages/openai/agent_framework_openai/_chat_client.py - python/packages/openai/tests/openai/test_openai_chat_client.py Next iteration: - Preserve middleware-terminated and parallel function groups atomically. - Add preflight rejection for non-replayable reasoning groups in the dedicated validation slice. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Key decisions: - Return ordinary function results when middleware terminates a loop, removing the provider-specific durable marker. - Preserve every parallel call and available sibling result as one encrypted reasoning group in stateless replay. - Prove successful and policy-blocked batches through the public two-agent Foundry workflow and outgoing HTTP boundary. Files changed: - python/packages/core/agent_framework/_tools.py - python/packages/core/tests/core/test_function_invocation_logic.py - python/packages/openai/tests/openai/test_openai_chat_client.py - python/packages/foundry/tests/foundry/test_foundry_agent.py Next iteration: - Add preflight rejection for non-replayable and partially compacted reasoning groups. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Key decisions: - Validate client-managed reasoning groups after compaction and report every affected reasoning and call identifier before transport. - Permit service-owned continuation and fully excluded atomic groups while rejecting partial compaction projections. - Surface encrypted-reasoning capability failures without lossy retries. Files changed: - python/packages/openai/agent_framework_openai/_chat_client.py - python/packages/openai/tests/openai/test_openai_chat_client.py Next iteration: - Run the resource-specific Foundry proof and finish PR microsoft#7233; that live proof remains intentionally local and requires the configured developer resource. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

moonbox3 · 2026-07-22T04:45:17Z

Why did we choose to drop the function call and result pair instead of replaying the reasonings?

I fixed this in latest commits.

… into issue-6074-mcp-reasoning-latest

Python: Fix reasoning-paired client tool replay

c4a8b3d

Copilot AI review requested due to automatic review settings July 21, 2026 03:28

giles17 added the python Usage: [Issues, PRs], Target: Python label Jul 21, 2026

Copilot started reviewing on behalf of moonbox3 July 21, 2026 03:28 View session

github-actions Bot reviewed Jul 21, 2026

View reviewed changes

Copilot AI reviewed Jul 21, 2026

View reviewed changes

moonbox3 marked this pull request as ready for review July 21, 2026 03:44

github-actions Bot reviewed Jul 21, 2026

View reviewed changes

Python: Handle middleware-terminated reasoning tool loops

16e0732

moonbox3 enabled auto-merge July 21, 2026 06:57

eavanvalkenburg reviewed Jul 21, 2026

View reviewed changes

Comment thread python/packages/openai/agent_framework_openai/_chat_client.py Outdated

moonbox3 and others added 5 commits July 22, 2026 08:07

Python: Preserve reasoning metadata in Foundry hosting

818e783

eavanvalkenburg reviewed Jul 22, 2026

View reviewed changes

Comment thread python/packages/openai/agent_framework_openai/_chat_client.py Outdated

Comment thread python/packages/openai/agent_framework_openai/_chat_client.py Outdated

Python: Avoid duplicating reasoning text metadata

6e52ad1

moonbox3 had a problem deploying to github-app-auth July 22, 2026 12:11 — with GitHub Actions Failure

Merge branch 'main' into issue-6074-mcp-reasoning-latest

53bf024

moonbox3 temporarily deployed to github-app-auth July 22, 2026 12:11 — with GitHub Actions Inactive

eavanvalkenburg approved these changes Jul 22, 2026

View reviewed changes

TaoChenOSU approved these changes Jul 22, 2026

View reviewed changes

Comment thread python/packages/core/agent_framework/_types.py

moonbox3 added this pull request to the merge queue Jul 22, 2026

github-merge-queue Bot removed this pull request from the merge queue due to failed status checks Jul 22, 2026

Python: Gate encrypted reasoning for Foundry agents

4ec0cbc

Merge remote-tracking branch 'origin/issue-6074-mcp-reasoning-latest'…

e84a3de

… into issue-6074-mcp-reasoning-latest

moonbox3 temporarily deployed to github-app-auth July 22, 2026 23:00 — with GitHub Actions Inactive

Python: Type stateless reasoning integration test

df10b26

moonbox3 temporarily deployed to github-app-auth July 22, 2026 23:19 — with GitHub Actions Inactive

Python: Narrow Foundry mock call arguments

344404f

moonbox3 temporarily deployed to github-app-auth July 22, 2026 23:28 — with GitHub Actions Inactive

moonbox3 enabled auto-merge July 22, 2026 23:44

alliscode approved these changes Jul 22, 2026

View reviewed changes

moonbox3 added this pull request to the merge queue Jul 22, 2026

Merged via the queue into microsoft:main with commit a2927c1 Jul 22, 2026
37 checks passed

moonbox3 mentioned this pull request Jul 23, 2026

Python: [Bug]: FoundryAgent inherits unrelated OPENAI_CHAT_MODEL for agent-reference requests #7272

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Python: Fix stateless replay of reasoning-paired tool calls#7233

Python: Fix stateless replay of reasoning-paired tool calls#7233
moonbox3 merged 14 commits into
microsoft:mainfrom
moonbox3:issue-6074-mcp-reasoning-latest

moonbox3 commented Jul 21, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Jul 21, 2026 •

edited

Loading

Uh oh!

github-actions Bot left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

github-actions Bot left a comment

Uh oh!

Uh oh!

TaoChenOSU commented Jul 21, 2026

Uh oh!

moonbox3 commented Jul 22, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

Uh oh!

Conversation

moonbox3 commented Jul 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Motivation & Context

Description & Review Guide

Related Issue

Contribution Checklist

Uh oh!

github-actions Bot commented Jul 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Python Unit Test Overview

Uh oh!

github-actions Bot left a comment

Choose a reason for hiding this comment

Automated Code Review

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

github-actions Bot left a comment

Choose a reason for hiding this comment

Automated Code Review

Uh oh!

Uh oh!

TaoChenOSU commented Jul 21, 2026

Uh oh!

moonbox3 commented Jul 22, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

moonbox3 commented Jul 21, 2026 •

edited

Loading

github-actions Bot commented Jul 21, 2026 •

edited

Loading