The Final Agentoutput.raw Converts Tool's Raw_output Into A String

easy fix

ah ok, thank you for fixing it!

Thanks for catching that 👍

I ran into a similar issue when the return_direct tool raises an exception. AgentWorkflow._call_tool converts the exception to a string here. This propagates up to the AgentOutput here, but from there I don't see an easy way to tell that the AgentOutput.raw was actually an error instead of a string.
Maybe the if-condition should be filtering out Tool outputs that are errors, or maybe the ToolOutput shouldn't convert the exception to string, or maybe both?

I guess since it's a return_direct tool, it's ok that the final AgentOutput contains an error, it just needs to have a way to tell that it's an error.

@JG someone else brought this up somwhere, I was thinking of adding an argument to tools like raise_errors=True (or a global option on the workflow itself) -- unsure of which is best, and haven't had time to take a stab

As a quick solution, can we change raw_output=str(e) to raw_output=e here?

Yea that could work 🤔

If I can't tell the difference between a regular string response from the LLM and an error message in the StopEvent, then I'd either have to stop using StopEvent or stop using return_direct it seems like. Just don't want error messages leaking out to end users.