[Bug]: The LLM doesn't understand the Finish tool very well

[enyst]

Is there an existing issue for the same bug?

• I have checked the existing issues.

Describe the bug and reproduction steps

Looking at trajectories, specially on evals, but also on those in the resolver I think, I seem to see too often that when the LLM is happy with the job, it sends a MessageAction, not a FinishAction. We insert then a fake user message to remind it to can call the Finish tool if it's done, and then it calls the Finish tool. That's one more iteration, often.

Maybe we can explain better or more clear that it's supposed to call Finish when it thinks the task is done.

OpenHands Installation

Docker command in README

OpenHands Version

No response

Operating System

None

Logs, Errors, Screenshots, and Additional Context

No response