I’m getting two frequent issues with the chat part of the IDE
Tasks, especially longer running ones, get cancelled with no input from me. The agent thinks I did it and reports “the user cancelled the task”. I saw “^C” in the output for a simple dir command
$ dir /s /b _build*.exe
^CTerminate batch job (Y/N)? dir /s /b _build*.exe
Terminate batch job (Y/N)?
Sporadically, any task, simple or complex, will just hang and not run. Restarting Cursor tends to get it working again, but it is disruptive to have to keep restarting.
Steps to Reproduce
Use Cursor normally on Windows.
Expected Behavior
Task run immediately and don’t get cancelled without input from me.
Operating System
Windows 10/11
Current Cursor Version (Menu → About Cursor → Copy)
Thanks for the quick response! I’m still getting random ‘q’ characters. The other issues tended to happen after working with the agent for a bit, time will tell if it fixed those issues.
when you say “new chat”, do you mean literally start an entirely new chat? I continued the same chat after restarting all instances of Cursor I had running after making the settings change. I verified after bringing one back up that the setting stuck.
I did notice that there are more terminals listed in the add context menu that I expect. I have 2 chat windows open and 1 personal terminal, but there are 5 listed in the menu
Another bit of info that might be related to commands hanging - I had one command hang in the AI terminal, but it also hung in mine when I ran the same command manually. I laid out some changes that I thought could help in the chat and pushed the message to force stop the hung command. The AI made some local changes, then tried to build. The build command hung before generating any output. Maybe forcing the hung command to abort put something on the back end in a bad state?
Has there been any progress on finding the root cause of this? Commands are still regularly cancelled or hang in the AI terminal. I have better luck when asking it to run the tasks in the background, which spawns a visible terminal and I can report when they are done, but that is inefficient, and requires piping the output to a file which the AI then tries to use commands to read and often hang or get cancelled as well. These issues are very disruptive to my daily workflow.
