In the latest version 0.46.8, whenever the agent executes a command in the terminal, it gets stuck until manually canceled.
I have tried using Bash, Zsh and Fish, but the issue does not seem to be related to the shell. The command exits successfully, but the agent remains stuck, waiting for the terminal tool to finish.
Same issue, wasted a few 10s of prompts due to this hanging, many more considering the lost discussions because it cannot execute. @ericzakariasson@danperks could we get some response
I think the pattern I’ve noticed is that the first time the agent calls the terminal in a conversation, it works fine. Every subsequent terminal use reuses the first terminal, and that leads to this bug.
So to avoid this, I basically always press “pop out terminal” after the first one finishes to ensure that subsequent calls create a fresh one each time → no more bug.
Still very annoying (especially when it does multiple commands in succession and you’re basically trying to snipe the pop out button in like 2 seconds) but I’m also noticing the same bug with other AI IDEs FWIW, so maybe it’s an OS-wide or model issue?
Tried a few other things. When i have some description in the settings it somehow asks me permissions everytime but if I just have allow list like below, it works fine:
another things is p10k is definitely messing with cursor. I have completely remove p10k configuration and now it does not get stuck(like cancel with lading indicator). It just exits immendiately after command is run and controil is back to chat
Just tried it one more time with bash and zsh in 0.46.11, no p10k, but it still gets stuck until I manually cancel the terminal command, rolled back to 0.45 again