`FunctionTool.strict_json_schema` is missing when using Chat Completions API

[mhjuan]

Please read this first

• Have you read the docs?Agents SDK docs → yes
• Have you searched for related issues? Others may have faced similar issues. → yes

Describe the bug

FunctionTool.strict_json_schema is not passed to the ChatCompletion API, because it is not converted in chatcmpl_converter.py::Converter.tool_to_openai()

    @classmethod
    def tool_to_openai(cls, tool: Tool) -> ChatCompletionToolParam:
        if isinstance(tool, FunctionTool):
            return {
                "type": "function",
                "function": {
                    "name": tool.name,
                    "description": tool.description or "",
                    "parameters": tool.params_json_schema,
                },
            }

        raise UserError(
            f"Hosted tools are not supported with the ChatCompletions API. Got tool type: "
            f"{type(tool)}, tool: {tool}"
        )

Due to this, the tool JSON schema returned by the API does not consistently follow the format specified in the request.

Debug information

• Agents SDK version: v0.6.3
• Python version: Python 3.12

Repro steps

import json

from agents import function_tool
from agents.models.chatcmpl_converter import Converter


@function_tool(strict_mode=True)
def dummy_tool(x: int) -> int:
    """Return x unchanged for testing strict_mode propagation."""
    return x


# Convert the tool as the OpenAIChatCompletionsModel does
openai_tool = Converter.tool_to_openai(dummy_tool)
print("Tool as sent to ChatCompletion (from OpenAIChatCompletionsModel):")
print(json.dumps(openai_tool, indent=2))

Expected behavior

Since strict_mode=True is set on dummy_tool, the printed function schema should include:

"strict": true

Actual behavior

{
  "type": "function",
  "function": {
    "name": "dummy_tool",
    "description": "Return x unchanged for testing strict_mode propagation.",
    "parameters": {
      "properties": {
        "x": {
          "title": "X",
          "type": "integer"
        }
      },
      "required": [
        "x"
      ],
      "title": "dummy_tool_args",
      "type": "object",
      "additionalProperties": false
    }
  }
}

Suggested Fix

Add "strict": tool.strict_json_schema to the function object, similar to how it is handled in openai_responses.py

I’m happy to open a PR with this change. Just filing the issue first to avoid duplicate work.