function_schema() raises opaque ValueError for params named model_dump / model_validate / etc.

[devteamaegis]

Calling function_schema() (or using @function_tool) on a function whose parameter is named model_dump, model_dump_json, model_validate, model_validate_json, or model_validate_strings raises an unhandled ValueError deep inside Pydantic rather than a clear SDK error.

Trigger

from agents.function_schema import function_schema

def search_tool(model_validate: str, query: str) -> str:
    return f"search: {query}"

schema = function_schema(search_tool)

Traceback

File ".../src/agents/function_schema.py", line 407, in function_schema
    dynamic_model = create_model(f"{func_name}_args", __base__=BaseModel, **fields)
  ...
  File ".../pydantic/_internal/_fields.py", line 97, in _check_protected_namespaces
    raise ValueError(...)
ValueError: Field 'model_validate' conflicts with member <bound method BaseModel.model_validate ...>
  of protected namespace 'model_validate'.

Pydantic's create_model() refuses any field whose name matches a built-in BaseModel method in a protected namespace. The error message gives no hint that the issue is the parameter name or how to fix it. These five names are distinct from model_config / model_fields / model_computed_fields (covered by PR #3548, which surfaces a different TypeError).

The fix is to detect these names before calling create_model() and raise a UserError with an actionable message.

[norika1207-lab]

One practical way to make this easier to evaluate is to turn the report into a small contract test around the runtime boundary, not only the user-facing symptom.

For function_schema opaque ValueError for params named model_dump/model_validate, I would check three things: first, the failing input is preserved as a minimal fixture; second, the SDK emits the same observable event/result shape before and after the fix except for the intended change; third, the trace or serialized session contains enough metadata to explain what happened after a resume or retry.

The part that matters for agent systems is durability: if a tool call, memory write, handoff, or structured-output step fails, downstream code should not have to infer state from natural-language output. A focused regression test around that boundary would make the issue much easier to close with confidence.