I can’t seem to get the Agent to run with “Auto-run” selected without needing to use the “Run Command” button with 3.5 Sonnet
I’ve noticed it as well. It’s a bit misleading that it’s auto-run enabled but does not automatically run the commands. From my understanding, the auto-run prompt should be enough to instruct the agent, but I’ve found that it ignores it most of the time. I add any command I’m fine with it running into the allow list, and that does the trick for me.
Add the commands that won’t run into the allow list, just the first part, such as cd and that should fix it.
But, be careful when adding commands to your allow list; agents sometimes stack commands. For instance, I don’t have cd as an approved command because in one of the projects I’m working on, the agent tries to run the startup bash script itself each time, but it always cds into the directory first. So I leave it out. It halts the auto-run, and I do it myself, then reply canceling the command with whatever is needed.
Thanks for that.
What do you have in the Instructions line?
I tried adding a nonsense command “xsfg” as a Not Allowed item. Adding the allowed commands one by one would take a while.
| mikes-bowden
March 13 |
- | - |
I’ve noticed it as well. It’s a bit misleading that it’s auto-run enabled but does not automatically run the commands. From my understanding, the auto-run prompt should be enough to instruct the agent, but I’ve found that it ignores it most of the time. I add any command I’m fine with it running into the allow list, and that does the trick for me.
Add the commands that won’t run into the allow list, just the first part, such as cd and that should fix it.
But, be careful when adding commands to your allow list; agents sometimes stack commands. For instance, I don’t have cd as an approved command because in one of the projects I’m working on, the agent tries to run the startup bash script itself each time, but it always cds into the directory first. So I leave it out. It halts the auto-run, and I do it myself, then reply canceling the command with whatever is needed.
This is all I have in the instructions. It sometimes works, but primarily, the allowed commands work the best, at least for me.
Automatically execute safe, non-destructive commands, including compilation commands, build commands, git operations (except force pushes or hard resets), package manager operations (npm/pip/yarn install, update), file navigation, search and replace operations, model context protocol tools, test runners, and debugging commands.
1 Like
Any cursor doc for this?
This is gold! Thank you!
March 13 |
- | - |
This is all I have in the instructions. It sometimes works, but primarily, the allowed commands work the best, at least for me.
Automatically execute safe, non-destructive commands, including compilation commands, build commands, git operations (except force pushes or hard resets), package manager operations (npm/pip/yarn install, update), file navigation, search and replace operations, model context protocol tools, test runners, and debugging commands.
1 Like
Not that I can find. It has the old for the Yolo mode. I believe I read in a post here somewhere that they are working on updating the docs.
Thanks. I simply stack them as I think of something to add and let AI consisify it for me.
Did you notice a change over last two minor versions? 0.46 and 0.47?
The YOLO prompt needed to be empty in order to make the YOLO allowed commands work.
I dont put anything into that field, but use allow and disallow lists.
remove the auto-run prompt. that breaks everything. allowlist works without it only (cursor team will look into that afaik)
also: sudo alone is too dangerous!
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.