Issue with output_type and Custom LLM Providers in openai-agents-python

[hlahany]

Describe the bug

When using the output_type parameter in the Agent class with a custom LLM provider, a validation error occurs with the returned JSON. This leads to the following error from Pydantic when trying to parse the result:

for TypeAdapter(MapperOutput); 1 validation error for MapperOutput
  Invalid JSON: expected value at line 1 column 1 [type=json_invalid, input_value='\```json\n{\n    "mapping...erage classes."\n}\n```', input_type=str]

Debug information

• Agents SDK version: v0.2.3
• Python version: v3.12.0

Repro steps

class MapperOutput(BaseModel):
    mapping_table: str = Field(..., description="The mapping table")
    additional_notes: str = Field(..., description="Any additional notes for the mapping.")

mapper_agent = Agent(
    name="mapper_agent",
    instructions=mapper_agent_instructions,
    output_type=MapperOutput,
    model=OpenAIChatCompletionsModel(model=model, openai_client=client)
)

Expected behavior

The output_type mechanism should be robust to code block markers and successfully parse the JSON, or should ensure that the model output is stripped of such markers before attempting to parse.