Some Bugs with Cursor

Feedback
1

I am currently using the Nightly, Web, and Stable versions simultaneously.

Web Version:

The web version recently updated its UI, allowing you to directly read modified files from the sidebar, which is great.

Bug Fixes

  1. The model selection list missed the “Auto” option, and the multi-model rules are simple. I can’t use a single model for multiple rounds.
  2. Gray hints like Plan and Update cannot be clicked to view what the model is thinking.
  3. Cannot change the title.

Feature Requests

  1. Dragging a file into the input box does nothing now.
  2. No voice input, @, or / buttons now.
  3. When there are to-do items, I hope to see the progress in the main list. The current effect makes it difficult for me to observe which instances are close to completion and which have just started.
  4. I hope to support prompt configuration for instance titles, PR titles, and descriptions, so I can quickly filter out the instances I want to follow in the main list.
  5. Hope to support previewing on the web just like z.ai. Maybe you need to set up a preview service in AWS and forward the port? It would be even better if it could integrate automatically with my Vercel or Cloudflare Worker, since both already do branch previews in Github. This way, adding a “Preview” button allows me to access the link with one click.
  6. Lacks token consumption statistics.
  7. Lacks autocomplete for the input box.

Stable Version:

After updating Agent Layout, it is indeed cleaner in some cases. Multi-model also saves me some retry time.
There’s a part of the workflow I’m not clear about: when I’ve selected one instance that I want, do I have to create a PR? Is there no way to merge it locally into my working branch?

Bug Fixes

  1. After using multi-model, I can’t open multiple Reviews for side-by-side comparison. When I click into another instance, the Review page automatically switches to the Review for that instance.
  2. Instances repeatedly try to use git staus, then wonder why there are no changes. Then it reads the file in the terminal, tells me it modified the file, but the file still has the old content. Several instances wasted a lot of time on this.
  3. After using multi-model, a bunch of modified files are opened, but they are all unsaved. How is this designed? This is very disruptive and may be caused by Bug 2? But I can’t waste time saving them one by one.
  4. After entering text, the voice button is incorrectly hidden.

Feature Requests

  1. The instance list is very simple and much uglier than the web version, with less information. It only shows the title, [num] Models, [time ago], and nothing more.
  2. Hope to have split-screen previews with multi-model. This way, I can quickly preview the results of each instance and start screening. Otherwise, I have to apply and preview them one by one.
  3. View instances of multiple projects at the same time in the Agent Sidebar.

Nightly Version

Bug Fixes

  1. After clicking “Try Agent Layout”, there is no button to switch back. This is very confusing. I ultimately had to rely on a shortcut key to resolve it.

Thanks for reading.

@danperks @andrewh