[Bug] Incorrect required field in params_json_schema causes tool misuse

[gdisk]

Description

When generating params_json_schema for tool definitions, the current implementation incorrectly includes all parameters in the required array, regardless of whether they are truly mandatory.

This behavior may lead to unexpected or erroneous tool-calling behavior in downstream models. For example, the model may be forced to supply values for optional parameters, and in practice it often fills them with fallback values such as empty strings (""), zero (0), or literal "null" strings. This can degrade tool performance and cause hard-to-debug failures in production.

Expected Behavior

Only parameters that are explicitly marked as required should appear in the required field of the generated JSON Schema. Optional parameters should remain absent from the required array and should not be forced by the schema.

[seratch]

Thanks for reporting this. I traced this through the function tool schema generation path, and the current behavior is intentional for strict function-calling schemas: OpenAI strict mode requires every property to be listed in required, with optional fields represented by allowing null as a type option.

If you need traditional JSON Schema optional-key behavior, please create the tool with @function_tool(strict_mode=False) or use FunctionTool(..., strict_json_schema=False). That opts out of strict mode and preserves optional parameters with defaults as optional schema fields.

I am going to close this as expected behavior rather than an SDK bug. If you have a concrete reproduction where an optional nullable field generated by strict mode is not accepted or is invoked incorrectly despite the schema allowing null, please share that narrower example and we can revisit it.