Python: align Foundry hosting response conversion

[Eduard van Valkenburg (eavanvalkenburg)]

Motivation & Context

Foundry hosting converted some structured arguments and results with Python string representations, which can produce invalid JSON on the wire and erased falsey function results. MCP result serialization also differed depending on whether a matching call was still active, while supported shell execution limits were dropped during response conversion. These inconsistencies make replayed items and hosted response events less faithful to Agent Framework Content values.

Description & Review Guide

• What are the major changes? Use one JSON-safe argument/result serializer, preserve falsey results, align correlated and uncorrelated MCP output serialization, retain supported shell timeout/output limits, remove an unreachable duplicate result branch, and delegate shared output-item parsing to the input-item converter.
• What is the impact of these changes? Structured values remain valid JSON, 0 and False survive conversion, MCP outputs are consistent, shell settings round-trip through Agent Server 2.1 models, and the duplicated parser logic is reduced without changing public APIs or dependencies.
• What do you want reviewers to focus on? Please verify the serializer semantics and the behavior-neutral delegation between the generated input and output item shapes.

Related Issue

N/A — focused follow-up cleanup.

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.

[Copilot]

Pull request overview

Aligns Foundry hosting response conversion with JSON-safe serialization and preserves shell execution metadata.

Changes:

• Unifies argument/result serialization and preserves falsey values.
• Consolidates input/output item conversion logic.
• Adds regression tests for serialization, MCP output, and shell limits.

Reviewed changes

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

File	Description
_responses.py	Updates response serialization and item conversion.
test_responses.py	Adds conversion and round-trip coverage.

---

💡 Add a code-review agent skill for context-aware, tailored reviews. Learn more in the docs.

[github-actions]

MAF Automated Review — Iteration 1

Result: Findings reported
Scope: full PR (1 commit(s)): e5c2c5fe9824
Model: gpt-5.6-sol

Overview

The PR successfully centralizes JSON-safe conversion, preserves falsey function results and shell limits, and adds focused parity tests for correlated and uncorrelated MCP output. The shared conversion path also preserves approval storage and the specialized OutputItem cases. Two MCP output boundaries remain: arbitrary values permitted by the public Any contract can now fail serialization, and uncorrelated mappings with a text field can lose accompanying structured data.

Reviewed the supplied pull-request change set across correctness, security/reliability, architecture, and failure behavior.
2 verified findings remained after source verification (2 medium) across 1 file. Details are attached to the affected lines below.

Affected areas: python/packages/foundry_hosting/agent_framework_foundry_hosting/_responses.py

[github-actions]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/foundry_hosting/agent_framework_foundry_hosting
_responses.py	778	100	87%	116–118, 234, 299–300, 314, 317–318, 385, 394, 471, 503, 518–521, 585, 635, 697, 709, 725–726, 731–734, 738–740, 745, 753, 756, 767, 812, 822, 833–835, 849–851, 854, 888–891, 896, 899, 919–920, 1017, 1088–1091, 1095–1097, 1102–1106, 1111, 1118, 1133–1139, 1149, 1155, 1159, 1183, 1192–1193, 1197, 1695–1696, 1700, 1745, 1747, 1749, 1751, 1755, 1763, 1766–1770, 1772, 1841, 1843, 1854–1855, 1862
TOTAL	48348	4484	90%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
9870	36 💤	0 ❌	0 🔥	2m 47s ⏱️

[github-actions]

MAF Automated Review — Iteration 2

Result: No findings
Scope: 22 net-new commit(s): dc7714295fff, 52a376453994, 34d1ac70b2fc, 178f365dc69e, 44c8cf6a2012, 6ed829cf713e, f7e78c689bb9, 341d0ec35bbc, 947d933f2385, de02975ee606, 6d532cf77e26, edfe115ea06b, 1aca2a95e3b9, 4c0bff8b77d5, a617392cc79b, 5996105a1bf2, d2a934d53530, 018dea570b84, 11d628d60ff2, 6a0773ba2180, 1a40758bfa86, a49a59772cf6
Model: gpt-5.6-sol

Overview

The change centralizes JSON-safe conversion, preserves falsey values and structured MCP output, retains shell limits, and delegates shared output parsing to the input converter. Focused tests cover unsupported leaf values, structured mappings, replay metadata, MCP correlation, and streaming lifecycle behavior. No new Critical, High, or Medium issue remains after reconciling the implementation with those guards and the supplied prior feedback.

Reviewed the supplied incremental change set across correctness, security/reliability, architecture, and failure behavior.
No publishable findings remained after source verification for this scope.