Python: FoundryEvals always emits arguments field for tool calls

[Sadok Barbouche (cr-sbarbouche)]

Motivation & Context

AgentEvalConverter.convert_message only set the arguments field on a tool_call content item when the parsed arguments were truthy, so tool calls with no model-supplied arguments were uploaded without an arguments field at all. Foundry's tool-aware evaluators (task_adherence, tool_output_utilization, tool_call_accuracy) require every tool_call item to have an arguments field, even if empty, and fail with FAILED_EXECUTION otherwise. This contributes to the FoundryEvals evaluation scenario for agents with zero-argument tools.

Description & Review Guide

• What are the major changes? python/packages/core/agent_framework/_evaluation.py: tc["arguments"] is now always set, defaulting to {} only when the parsed arguments are None (an explicit is not None check, not a truthiness check, so valid-but-falsy arguments like 0/""/False are preserved).
• What is the impact of these changes? Zero-argument tool calls now serialize with "arguments": {} instead of omitting the key, matching what Foundry's evaluators require. No behavior change for tool calls that already had arguments.
• What do you want reviewers to focus on? Whether defaulting to {} (vs. omitting the field) is the correct shape for all Foundry evaluators, and whether the is not None check should also apply to the existing {"_raw_arguments": "[unparseable]"} fallback path.

Related Issue

Fixes #7714

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.

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds support for tool calls that have no arguments by ensuring the evaluation conversion always emits an arguments field (defaulting to {}) and covering the behavior with a new test.

Changes:

• Add a test for zero-argument tool calls (arguments=None) expecting {}.
• Update message-to-eval conversion to always include arguments, defaulting to {} when missing.

Reviewed changes

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

File	Description
python/packages/foundry/tests/test_foundry_evals.py	Adds regression test to validate zero-argument tool calls serialize with empty arguments.
python/packages/core/agent_framework/_evaluation.py	Ensures tool calls always include an arguments key, defaulting to {} when absent.

---

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

[Evan Mattson (moonbox3)]

Please use our pull request template, and per contributing guidelines: respond to and resolve all open PR comments.

[Evan Mattson (moonbox3)]

/review

[github-actions]

MAF Automated Review — Iteration 1

Result: No findings
Scope: full PR (2 commit(s)): 042676c982e3, 103185927c44
Model: gpt-5.6-sol

Overview

The change unconditionally emits the Foundry-required arguments field while preserving all non-None argument values, including falsy values, and defaulting only missing arguments to {}. The focused regression test covers the zero-argument case, while existing tests protect JSON-string parsing and mapping arguments. No publishable Critical, High, or Medium defect remains after reconciliation.

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

[github-actions]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/core/agent_framework
_evaluation.py	721	75	89%	164, 172, 488, 490, 587, 599, 791, 794, 872–874, 879, 916–919, 975–976, 979, 985–987, 991, 1024–1026, 1082, 1118, 1130–1132, 1137, 1161–1166, 1259, 1337–1338, 1340–1344, 1350, 1389, 1739, 1741, 1749, 1759, 1763, 1808, 1826–1827, 1905, 1907, 1913, 1921, 1936, 1974, 1980–1984, 2016, 2039, 2047–2048, 2050, 2075–2076, 2081
TOTAL	46816	4360	90%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
9466	36 💤	0 ❌	0 🔥	2m 31s ⏱️