Python: parse structured response value from final message

[Sheldon (liuzemei)]

Motivation and Context

Fixes #6366.

When an agent response contains intermediate messages and a final structured JSON payload, AgentResponse.value currently parses self.text, which concatenates all message text. That can join intermediate tool/context JSON with the final response JSON and cause ValidationError / JSONDecodeError from trailing characters.

Description

• Parse structured ChatResponse.value and AgentResponse.value from the last non-empty message text instead of the public concatenated .text value.
• Keep .text behavior unchanged for callers that want the full concatenated response text.
• Add regression coverage for Pydantic response formats and JSON schema mapping response formats.

Testing

uv run --group dev pytest packages/core/tests/core/test_types.py -q -k 'final_message or empty_text or invalid_schema or chat_response_with_format or mapping_response_format'\nuv run pytest packages/core/tests/core/test_types.py -q\nuv run pytest packages/core/tests/core/test_middleware.py -q\nuv run pytest packages/core/tests/workflow/test_workflow.py -q -k 'AgentResponse or response or output'\nuv run ruff format --check packages/core/agent_framework/_types.py packages/core/tests/core/test_types.py\nuv run ruff check packages/core/agent_framework/_types.py packages/core/tests/core/test_types.py\ngit diff --check\n```\n\n### Contribution Checklist\n\n- [x] The code builds clean without any errors or warnings\n- [x] The PR follows the Contribution Guidelines\n- [x] All unit tests pass, and I have added new tests where possible\n- [ ] **Is this a breaking change?** If yes, add "[BREAKING]" prefix to the title of the PR.

[Copilot]

Pull request overview

Note

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

This PR updates structured-output parsing for ChatResponse.value and AgentResponse.value to parse only the final non-empty message (ignoring earlier messages), and adds regression tests for multi-message responses.

Changes:

• Add _last_non_empty_message_text() helper to select the final non-empty message text.
• Update ChatResponse.value / AgentResponse.value to parse structured output from the final message only.
• Add tests asserting intermediate messages are ignored when parsing structured output.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
python/packages/core/agent_framework/_types.py	Parse structured output from the last non-empty message instead of concatenated response text.
python/packages/core/tests/core/test_types.py	Add tests validating structured parsing ignores intermediate messages for both response types.

[Sheldon (liuzemei)]

@microsoft-github-policy-service agree

[github-actions]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/core/agent_framework
_types.py	1226	97	92%	61, 70–71, 125, 130, 149, 151, 155, 159, 161, 163, 165, 183, 187, 213, 235, 240, 245, 249, 279, 705–706, 887–888, 1346, 1421, 1456, 1476, 1486, 1538, 1672–1674, 1959–1964, 1989, 2044, 2049, 2059, 2067, 2074–2078, 2096, 2169, 2182, 2187, 2300, 2323, 2584, 2608, 2707, 2888–2889, 2991, 3220, 3273, 3292, 3331, 3342, 3344–3348, 3350, 3353–3361, 3371, 3460, 3597, 3602, 3607, 3612, 3616, 3702–3704, 3733, 3821–3825
TOTAL	44044	5270	88%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
8801	33 💤	0 ❌	0 🔥	2m 17s ⏱️

[Eduard van Valkenburg (eavanvalkenburg)]

and some tests are failing, please have a look at those as well Sheldon (@liuzemei)