Handoff Agent and Tool Call Not Triggering Reliably in Multi-Agent Setup

[adhishthite]

Please read this first

• Have you read the docs?Agents SDK docs : YES
• Have you searched for related issues? Others may have had similar requests : YES

Question

I am trying to do hand-offs in my production code, but I am facing a peculiar issue. I am simplifying the code here to show what I am getting.

Here's my code:

import asyncio
import random
from agents import Agent, Runner, function_tool
from agents.extensions.handoff_prompt import RECOMMENDED_PROMPT_PREFIX
from openai.types.responses import ResponseTextDeltaEvent
from pydantic import BaseModel

@function_tool
def how_many_jokes() -> int:
    return random.randint(1, 5)

class JokeFormat(BaseModel):
    joke: list[str]
    language: str

async def main():
    format_agent = Agent(
        name="Format Agent",
        instructions=f"""{RECOMMENDED_PROMPT_PREFIX}
        You will format the received joke as a Markdown. You will only return the formatted jokes.""",
        model="gpt-4.1",
        handoff_description="You are a joke formatter. You will format the jokes in Markdown.",
    )

    english_agent = Agent(
        name="English Joker",
        instructions=f"""{RECOMMENDED_PROMPT_PREFIX}
        First call the `how_many_jokes` tool. Then, tell those many jokes in English. The jokes should be very short!
        Once you have the jokes, you will pass them to the Format Agent
        """,
        tools=[how_many_jokes],
        model="gpt-4.1",
        handoff_description="You are a joke teller. You will tell jokes ONLY in English.",
        output_type=JokeFormat,
        handoffs=[format_agent]
    )

    marathi_agent = Agent(
        name="Marathi Joker",
        instructions=f"""{RECOMMENDED_PROMPT_PREFIX}
        First call the `how_many_jokes` tool. Then, tell those many jokes in Marathi. The jokes should be very short!
        Once you have the jokes, you will pass them to the Format Agent
        """,
        tools=[how_many_jokes],
        model="gpt-4.1",
        handoff_description="You are a joke teller. You will tell jokes ONLY in Marathi.",
        output_type=JokeFormat,
        handoffs=[format_agent]
    )

    agent = Agent(
        name="Triage Agent",
        instructions="You determine which agent to use based on the user joke language preference.",
        handoffs=[english_agent, marathi_agent],
        model="gpt-4.1",
    )


    result = Runner.run_streamed(
        agent,
        input="Hello, tell me a joke in English.",
    )
    print("=== Run starting ===")

    async for event in result.stream_events():
        # We'll ignore the raw responses event deltas
        
        if event.type == "raw_response_event":
            # print(f"** -- Type of event data: {type(event.data)}")
            if isinstance(event.data, ResponseTextDeltaEvent):
                print(event.data.delta, end="", flush=True)
            continue
        
        # When the agent updates, print that
        elif event.type == "agent_updated_stream_event":
            print(f"\nAgent updated: {event.new_agent.name}")
            continue
        
        # When items are generated, print them
        elif event.type == "run_item_stream_event":
            if event.item.type == "tool_call_item":
                print("-- Tool was called")
            elif event.item.type == "tool_call_output_item":
                print(f"-- Tool output: {event.item.output}")
            elif event.item.type == "message_output_item":
                pass
            
        else:
            print(f"** -- Skipped Event Item type: {event.item.type}")


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

For this, consistently, I am getting the following output:

=== Run starting ===

Agent updated: Triage Agent

Agent updated: English Joker
{"joke":["Why did the scarecrow win an award? Because he was outstanding in his field!"],"language":"English"}%          

Problem

As you see, neither the tool is called, nor the format_agent is called. How do I ensure (not enforce, but ensure) that tools and the handoffs will work as expected? I have tried changing the prompts a little bit, the max I have gotten is that the how_many_jokes is called - but the format_agent is never called.

Example:

=== Run starting ===

Agent updated: Triage Agent

Agent updated: English Joker
-- Tool was called
-- Tool output: 2

{"joke":["Why did the scarecrow win an award? Because he was outstanding in his field!","Why don’t skeletons fight each other? They don’t have the guts."],"language":"English"}% 

@rm-openai Could you help me with please? I am running the latest version of the SDK.

[PedroStival]

you should use this formater as tool

tools=[
spanish_agent.as_tool(
tool_name="translate_to_spanish",
tool_description="Translate the user's message to Spanish",
),
french_agent.as_tool(
tool_name="translate_to_french",
tool_description="Translate the user's message to French",
),
],

like this