[Saquib Jawed (saquibjawedbit)]

Motivation & Context

A tool parameter annotated as Annotated[int, "How many items", Field(ge=1, le=10)] loses everything after the string description. _parse_annotation wrapped the trailing metadata in a single tuple, which Pydantic treats as one opaque metadata object and ignores. The constraints were missing from the JSON schema sent to the model, and out-of-range arguments were accepted when the tool ran. The same Field works when it is the only metadata, and the function's docstring already says the remaining annotations are returned as-is.

Description & Review Guide

• What are the major changes? _parse_annotation unpacks the metadata after the description into the new Annotated instead of wrapping it in a tuple. It uses the tuple-subscript form so it stays valid on Python 3.10. Two regression tests cover the parsed Annotated metadata and the end-to-end @tool schema and invocation validation.
• What is the impact of these changes? Field(...) constraints and other Pydantic metadata placed after a string description now appear in parameters() and are enforced on invoke. Parameters with only a description are unchanged.
• What do you want reviewers to focus on? The one-line change in _parse_annotation. If the trailing Field sets its own description, it now takes precedence over the leading string, following Pydantic's normal metadata merge order. Previously that Field was dropped entirely.

Related Issue

Fixes #8643

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]

Note

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

Copilot review overview

Review effort: Lite
Findings: 1 · 1

Open (2)

• This strict dict equality is brittle: JSON schema generation can legitimately add (or rename) keys… · New
• Matching on an exact validation message fragment is prone to breaking across Pydantic… · New

What changed in this PR

Fixes loss of Pydantic Field constraints when a tool parameter uses Annotated[...] with a leading string description by preserving metadata items after the description.

Changes:

• Update _parse_annotation to expand trailing Annotated metadata items instead of wrapping them in a single tuple (so Pydantic can interpret them).
• Add a unit test verifying _parse_annotation preserves extra metadata after a description string.
• Add an end-to-end tool test ensuring JSON schema includes min/max constraints and invoke enforces validation.

File	Description
python/​packages/​core/​agent_framework/​_tools.py	Fixes Annotated reconstruction so metadata after a string description remains individually visible to Pydantic.
python/​packages/​core/​tests/​core/​test_tools.py	Adds regression tests for preserved metadata and for schema/validation behavior via @tool.

---

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

[Saquib Jawed (saquibjawedbit)]

@microsoft-github-policy-service agree

[Eduard van Valkenburg (eavanvalkenburg)]

The issue is not a issue, closing this PR.