Have the agent intelligently watch console output to terminate when stuck

Feature Requests
1

Sometimes the terminal wont reach a natural termination point and stay alive, but will have encountered an issue. I have to manually cancel it, then it proceeds to understand what went wrong and fix it. Which is amazing.

But if the agent could periodically view the output and realize when it’s stuck and choose to terminate it automatically (even if this is an optional feature / setting), that would speed up development and complete the loop in an awesome way.

So many trial and error iterations could run without my manual interaction.

A very simple version of the feature could just be to terminate if no new output happens for a while.

2

Isnt a timeout already integrated? perhaps this should be a bug report. But overall i agree, sometimes the timeout isn’t fast enough as it waits for more.

3

Yeah in my case, I never see any timeout. If there is one, it’s not mentioned in the settings. Not clear that this is a bug.

4

I used recently console output and it stopped watching by itself. However i have also noticed that it sometimes didnt stop. So i think this needs at least clarified how it is processed and how it decides when to stop.

1 Like
5

It happened to me here too.
I’m on version 0.47.1