I’ve been dealing with this issue for a while now. I tried all sorts of things with my terminal environment to fix it. What I’ve determined is that when the agent is running a terminal command in “non-interactive” mode, all output is hidden from the user and only a cancel button and loading spinner present.
I know that there is actually output occurring because I can click “Pop out terminal” and I will see it’s been super busy. The problem with this is that when you pop out the terminal the agent loses it’s ability to see it and interact with it, the terminal goes into the background process.
This drives me nuts because I can’t follow along with the results. There are times where the agent can hang/become blocked due to an unclear ending of the terminal action. Without the ability to see the terminal output there is no way to know whether a long running terminal action is stuck or just processing. If you pop out, the agent will move on as if the command completed, to which you need to waste a fast request just to tell it to stop and wait.
This has bugged me for months now so I felt I should post. I tried to edit my terminal environment after realizing the agent’s environment mirrored mine (it diverged at first so I didn’t realize it was possible), and still wasn’t able to have a meaningful effect on non-interactive terminal mode.
The agent’s ability to be a super human terminal user is insanely powerful and absolutely life changing. Fixing this would improve it a whole lot for me.