We need a HALT instruction #450
Description
What version of Code is running?
0.5.15
Which model were you using?
gpt-5
What platform is your computer?
Debian 12 on WSL2
What steps can reproduce the bug?
Sweet mother of god, I just watched a horror movie.
I have two big project repos mounted in a docker container, a new backend and a new frontend. They were developed separately to specs and heavily mocked. The time has come to remove the mocks and wire the frontend to the real backend data sources.
I went through detailed prompting and have code and it's sub agents happily beavering away for probably 60-90 minutes, meticulously writing new functions and peer tests, testing them in isolation and committing them. It was working through building out the entire API interface layer on the frontend side (unified interface to the backend, so all components have a single touch point and single source of truth). It all looked GREAT, much better than attempts to get parallel work going previously.
Then it reached a point where it ran the larger test suite and I happen to be in another window doing something.
I come back and holy shit, it is literally reversing all of it's work because other tests in the suite are now failing, since the mocks have been removed and we have real code in place. Like literally rapidly editing all the files and taking out an hour of dozens of agents work, "so the tests will pass".
My issue isn't that this happening, and part of the fault surely lies with me.
My issue is that there is no emergency stop button. I pounded the escape key a dozen times, and I got all kinds of messages that things were stopped, but more and more changes kept happening. We need a proper HALT THIS SECOND emergency stop. Was the issue just buffering? Why does so much work happen after you hit ESC to kill auto drive? It isn't quite clear what the chain of events should be there.
What is the expected behavior?
No response
What do you see instead?
No response
Additional information
No response