Agent session error in OpenAI Agents SDK when handling tool output

[mehdinathani]

Please read this first

• Have you read the docs? Yes — Agents SDK docs
• Have you searched for related issues? Yes — I could not find a similar open issue.

---

Describe the bug

When running an agent tool in the openai-agents-python SDK, the process unexpectedly terminates with an openai.error.InvalidRequestError despite valid inputs.
The issue appears to occur intermittently and may be linked to tool input/output schema validation not being enforced consistently.

Error message:

openai.error.InvalidRequestError: Invalid tool output: expected object matching schema, got null

---

Debug information

• Agents SDK version: v0.0.3
• Python version: 3.10.14
• OS: Windows 10 Pro 64-bit

---

Repro steps

Run the following minimal script:

from openai import Agent, Tool

# Simple echo tool
class EchoTool(Tool):
    name = "echo"
    description = "Echoes the input back"
    input_schema = {"type": "object", "properties": {"message": {"type": "string"}}}
    output_schema = {"type": "object", "properties": {"reply": {"type": "string"}}}

    def call(self, input):
        return {"reply": input["message"]}

# Create agent with tool
agent = Agent(
    name="TestAgent",
    tools=[EchoTool()],
)

# Trigger tool
response = agent.run("echo", {"message": "Hello World"})
print(response)

---

Expected behavior

The script should print:

{"reply": "Hello World"}

Instead, the run intermittently fails with the InvalidRequestError shown above.

---

[seratch]

Thanks for writing in, but it seems your code snippet does not work. When you want to run an agent with tools + structured result, you can do something like this:

import asyncio
from typing_extensions import TypedDict
from agents import Agent, Runner, function_tool

class EchoResult(TypedDict):
    reply: str

@function_tool
def echo(message: str) -> EchoResult:
    return EchoResult(reply="You said: " + message)

agent = Agent(
    name="TestAgent",
    instructions="You are a helpful assistant that can echo messages using the tools.",
    tools=[echo],
)

async def main():
    response = await Runner.run(agent, "Hello World")
    # print(response.raw_responses)
    print(response.final_output)

if __name__ == "__main__":
    asyncio.run(main())
```

Hope this helps. Whenever you have a more specific question, please feel free to open a new question issue for it.