I just updated Cursor this morning. I saw that the composer now has an “agent” mode. I enabled it and I’ve noticed that the agent has now stopped applying any of its changes to my code. It almost seems like it has a different source of truth, but it seems to believe that it’s actually applied the changes.
Is there any specific information I can share to help the Cursor team debug this? I also enabled logging in my project, and I’m happy to give you more information to help you debug this. I’m really excited to try out agent mode in Cursor!
Thanks for replying! Also, I’ve generally just set it to non-privacy mode. I’d like to support you however I can, and I can understand that this mode is likely very helpful for constructing good internal evals. After I started using Cursor, I started experiencing a similar sort of joy with programming as when I first learned to program. And I’m really excited to see where you’ll take all of this!! I hope that the Cursor team will be respectful of my private repo data!
Here’s the request ID for the last composer: 32d6ebfa-ed9d-4e08-8d65-7f791510b425.
Here’s a request ID for the last apply: ffdbdda3-6bd5-4ad8-ae3a-633b8764f7f7. Note that I did an “undo” of this apply, and rejected all the changes because I couldn’t understand them.
I think I have a bit more clarity on the bug. It seems that the agent mode isn’t producing a diff. It seems that within the agent mode’s conversation, it’s correctly tracking the state of the files? But it looks like there’s a bug in actually visually demonstrating and applying those diffs to the codebase?