? Where should we push the ‘bugfix/remove-feedback-memory-type’ branch? [Use arrows to move, type to filter]
org/repo
Create a fork of org/repo
Skip pushing the branch
Cancel
Here’s an example where cursor sits and waits until timing out. Not quite a bug but it would be great if it could recognise that the terminal has become interactive and respond to it properly.
Do the regular terminal commands work for you? In my case all commands get stuck until I manually cancel them: [BUG] Terminal tool gets stuck after exit code
It’s 90% of commands. Didn’t see a pattern. What I did to fix this is just delete all my allowlist/blocklist settings and leave them empty except for block delete. Now it’s working fine. The settings seem to conflict with each other, so I think the way it’s designed needs a rethink, but it’s working fine for me now. Provided it doesn’t delete anything outside my project it can’t really do much harm anyway, and it hasn’t so far.
Using yolo mode with whitelisted command with great success here. Just instruct the LLM to pipe the result of the command into cat, this will bypass the interactive mode that gets you stuck.
I got mixed up with two separate issues - other commands were getting stuck but that was due to user approval, entirely separate issue.
@Starman are you instructing the LLM within a rule, or a specific chat message prompt?
I made this a general rule in .cursorrules. With Claude 3.7 demonstrating enhanced free-agency, you cannot predict which command line utility the LLM will use. So my recommendation is to make it general.