input list for agents when using as tools

[MingxuanZhangPurdue]

Please read this first

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

Describe the feature

What is the feature you're requesting? How would it work? Please provide examples and details if possible.

When using agents as tools, the current SDK does not support to extract the function calls trace for agents used as tools, however, it is important for some applications to view, investigate the agents behavior when used as tools

[seratch]

Thanks for sharing this feedback. Just to confirm, the issue you mentioned is the lack of function call tracing data inside an agent when the agent is used as a tool of a different agent? I wanted to check if I understand correctly because the issue title looks a bit different.

[MingxuanZhangPurdue]

Thanks for sharing this feedback. Just to confirm, the issue you mentioned is the lack of function call tracing data inside an agent when the agent is used as a tool of a different agent? I wanted to check if I understand correctly because the issue title looks a bit different.

Hi!!!

Thanks for the fast reply!!! I really appreciate it. Sorry for the confusing title.

Basically, right now, to the best of my knowledge, there is no easy/obvious way to access the function calls or the input history for agents when they are used as tools inside another agent, I am wondering if the SDK can have an additional feature to get this data for agents when used as tools.

I understand, this data might be accessible from the trace? But I really hope to directly access it from the SDK itself.

Thanks!!!!!!!!!!!

[seratch]

Thanks for clarifying! The tool only accepts defined parameters, and the absence of input items when using an agent as a tool is by design. Exposing the internal state of the caller agent (like conversation history or other data) would act like a global object, which makes complex use cases harder to manage. Because of that, we don’t plan to change this behavior.

I know this may not be the perfect answer for you, but I’d encourage designing your tool so it can take in enough information from the main agent. If you need a summary of the ongoing conversation, you can add a parameter for that in the function tool and ask the LLM to generate a useful summary when calling it. If you’d like to go with this approach, you can set up a function tool that internally calls Runner: https://openai.github.io/openai-agents-python/tools/#customizing-tool-agents

Hope this helps clarify things and it was helpful to you.

[github-actions]

This issue is stale because it has been open for 7 days with no activity.

[github-actions]

This issue was closed because it has been inactive for 3 days since being marked as stale.