Hi all, What do you guys think about a dedicated H...
# questions
l
Hi all, What do you guys think about a dedicated Human In Loop escalation tool in software-agent-sdk? In current implementation, Finish tool conflates two responsibilities: 1. Ask query to the user 2. Complete the task When I tried to build an orchestration layer above the Agent abstraction, got stuck with this responsibility mix up. I did a work around by adding special instructions to produce identifiable artifacts. This is not a scalable solution. So why don't we add a new tool?
m
If I understand your point: you would like to add a new tool ask_a_question right?
l
yes.. exactly..
Copy code
TOOL_DESCRIPTION = """Signals the completion of the current task or conversation.

Use this tool when:
- You have successfully completed the user's requested task
- You cannot proceed further due to technical limitations or missing information

The message should include:
- A clear summary of actions taken and their results
- Any next steps for the user
- Explanation if you're unable to complete the task
- Any follow-up questions if more information is needed
"""
This is the current description of FINISH tool. According to this, agent is supposed to generate a FINISH action when it needs to ask_a_question to the user.
m
Are you using the SDK? If so, you could implement a custom
AskQuestion
tool to provide to your agent. (This would be a quick and easy solution to unblock you right away.) Otherwise, this is something we have also been discussing here: https://github.com/OpenHands/software-agent-sdk/issues/2697 I'm interested to know what other people think about the
finish
tool. Should we narrow its responsibility, or is it already equivalent to an
AskQuestion
tool? One experiment you could try is creating a custom
AskQuestion
tool, giving it to the agent, and seeing if it uses it properly to ask questions proactively. This would help us see if it makes sense to have a separate tool, or if the
finish
tool is already enough.
l
Yes.. I'm using SDK. I will implement the AskQuestion tool and try to validate the concerns you mentioned in https://github.com/OpenHands/software-agent-sdk/issues/2697.