Python: Bug: Handoff functions with custom arguments

[hjerbii]

Describe the bug
When using Handoff orchestration and defining the handoffs, we specify the source/target agents and the handoff rule but not the extra argument(s) or information that we want source agent to transfer to target agent.

I went through the code and if I am not mistaken, handoffs are converted to tools with empty parameters (here):

Expected behavior
It would be nice if we can decide what parameters to transfer to target agent(s), or is it already possible to do so?. In my case, sometimes the source agent should act as a "supervisor" for the target agent and give it very specific task to accomplish to not get lost.

Platform

• Language: Python
• Source: main branch of repository
• AI model: OpenAI:GPT-4o
• OS: Windows

Thanks in advance!

[SrijitaRoy913]

Any update on this?

[TaoChenOSU]

Hi,

The Handoff orchestration isn't currently designed for handling this kind of scenario. The agents in handoff orchestration don't get inputs, instead they simply continue the conversation left off by the previous agent.

The scenario you described sounded more like an agent-as-a-tool scenario. Feel free to let us know if you have any questions about agent as a tool.

[TaoChenOSU]

Closing. Reopen if necessary.