I’m having this problem as well, working on my laptop with an older version has meant development didn’t get stalled this last week or two though, this issue is critical
I am paying you a lot of money, Cursor. Might need some sort of refund as this whole past week has been beset with problems - this isn’t what I upgraded for.
I also have all the commands related to my tests stopped running, such as npm test and the like, as if the Cursor terminal stopped working correctly, clearly a massive problem.
The only way this continues to work is if I select a really OLD conversation from the archive and have that open a terminal - this then reverts to old behaviour but i then have to keep using an older conversation and watch the “context” - but this can get you going again. Go into the agent conversation archive… select an old one - where it WAS working.. restart that conversation and have it do something simple… it works again… not sure for how long… but this seems to be working for now
Got the same bugs on version 2.1.50 (WSL).
It goes beyond the model failing to use the terminal (exit code 0, no output); it also breaks the auto-apply feature, forcing manual approval everytime.
It seems to happen specifically in workspaces with a heavy chat history/large database.
To fix the problem I deleted the project’s folder located in %APPDATA%\Cursor\User\workspaceStorage. It looks like the database gets corrupted or something. But this solution clears the chat history which is not satisfying.
Don’t know yet if the bug can happen again after cleaning the data.
3 Likes
Yup - this is the fix. If anyone else has this problem, do this 100%. It is a corruption issue indeed.
Thanks for the fix. I am curious about how you troubleshoot this?
Just chiming in, what helped me is to switch to Cursor Nightlies - currently on version 2.2.8.
I had temporarily downgraded to 2.1.36 which was working for me.
Today I noticed 2.2.9 was released. I installed it and things are working for me! Hope it solves it for the rest of you too.
I’m working on multiple servers (WSL local and remote SSH) and noticed that remotely I didn’t have this bug. Then I tested locally with an empty directory asking the AI to create a simple file in a repo and it was working without any problem.
Finally, given this information, a discussion with Gemini brought up that the chat history is the big difference between the projetcs and it told me to try this fix.
Thanks for this… worried that there seem to be different behaviors with a lot on WSL
This topic was automatically closed 22 days after the last reply. New replies are no longer allowed.