Agents Editor -> Browser Tool: 2 Issues

Support Bug Reports
, , ,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

I really like the Agents window (vs. the Editor window) and have been attempting to familiarize myself with it over the past month or two. I do have 2 issues I’ve captured directly related to the “Browser” windows when using this version of the Cursor app, though.

  1. “Design” button. Often after making a change, especially if i specifically selected an HTML object/element on the page, Cursor will implement the change. It does select the correct element, so that’s good, but then the “Design” button gets grayed out, even after Cursor has finished its implementation. I’ve found a work-around for this which is to simply hit F12 twice to open/close the Dev Tools panel. This will “renable” the Design button. It’s not a huge issue because the work-around is fast, but this seems like it could be a simple User Experience fix.

  2. Browser memory consumption is rough. I like using the Design tool as mentioned in #1 above, but using Cursor’s built-in browser is EXTREMELY memory-inefficient. I’ve got 32 GB of RAM, and it’s pressed to its limits when using the built-in browser. I did a quick comparison between my Cursor memory usage vs. Microsoft Edge memory usage, and while a particular tab in Edge might use ~400 MB of memory, the browser window in Cursor has ballooned up to as high as 12 GB of memory for that same window. It probably would have continued to grow over time as well. I suspect this is related to some kind of memory leak, and I saw a few forum posts related to memory leaks that didn’t seem to be quite related to this (they were mostly about Mac, and I’m on Windows). My repo IS on Windows, though, and I am using Docker to serve the application locally–so it’s possible there’s something related to those in here. Regardless, I wanted to provide these 2 bugs/feedbacks.

Bonus #3 unrelated to Browser Tool: Cursor deleted my local database data recently. This is my fault because I basically use the app in Yolo mode, and fortunately the deletion only impacted the database entries and not my desktop folders or (as I read some other horror stories about) the entire hard drive. Either way, it did inspire me to set up a backup policy on both my machine as well as my local dev setup. Now when I run my “ship-changes” skill for this repo, it automatically backs up the local database with a 14-file retention that FIFOs. And I also set up hard disk backups to have an easy restore point if my hard drives should unfortunately get wiped. No bug here–just thought I’d mention it as something that Cursor might want to consider any time someone sets up a workspace–instructing users to ensure they have some sort of “backup and recovery” plan.

As always, loving Cursor. Can’t wait to see what comes in the future. It’s been a tremendously enjoyable experience so far, so keep up the good work. $200/month max plan subscriber here for around half a year and no intentions of stopping any time soon. It’s cost me pennies on the dollar and improved my development speed 10x easily.

Steps to Reproduce

See description above

Expected Behavior

See description above

Operating System

Windows 10/11

Version Information

A shame there’s no option to do this in the Agents window. Here’s the paste form Help → About in the Editor window:

Version: 3.9.16 (system setup)
VS Code Extension API: 1.105.1
Commit: 042b3c1a4c53f2c3808067f519fbfc67b72cad80
Date: 2026-06-27T06:41:01.941Z
Layout: editor
Build Type: Stable
Release Track: Default
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.256
OS: Windows_NT x64 10.0.26200

For AI issues: which model did you use?

Composer 2.5 mostly, but I have an “agent team” skill that I will throw at problems I’m unable to resolve in 2 turns. It uses Composer 2.5 as a “moderator”/“lead” and then asks 4 sub-models to analyze the same problem (Opus 4.8, Composer 2.5, Gemini 3.1 Pro, and GPT 5.5), then each sub-agent analysis gets passed back to the moderator for consolidation and consensus.

Does this stop you from using Cursor

No - Cursor works, but with this issue

6

Hey! Thanks for the incredibly thorough writeup and the kind words. :slight_smile:

  1. “Design” button graying out after an edit. This is a known issue and we already have a bug open for it, so it’s on our radar to fix.

  2. Browser memory usage. Were you seeing this in the Agents window or the Editor window? It matters because the two manage browser memory differently. In the Agents window, each tab is a full embedded browser view with its own renderer process, and we intentionally keep a warm pool of those views alive. The Editor window doesn’t keep that pool (it frees a view when you close the tab), so if you saw the 12 GB there, that points more toward a leak than expected behavior. Either way it’s useful data!

  3. Database deletion / backups. Glad the damage was contained! One thing worth trying if you haven’t: Auto-review run mode. It’s the default now on new installs, and instead of full YOLO, it lets the agent run while also running a classifier against commands, surfacing spicier ones for approval. Worth a look!

8
  1. Ok great!
  2. I’ve been trying to exclusively use “Agents” window these days, especially to leverage the Design feature; but, with the memory leak issues i mentioned above, I’ve reverted to simply having an open browser on another screen (I use Microsoft Edge fully updated on latest patch typically). If I need to make changes, I just take a screen (Greenshot Editor is my tool of choice), and then I paste the screenshot. It uses a few more tokens, but generally I get the desired results I expect. Would still prefer to use the Agents window exclusively, though, for the added benefits. I suspect future enhancements will come to the Agents window as well that I’ll want to leverage.
  3. Ok I’ll check out Auto-review run mode. From a quick glance, the .cursor/permissions.json file seems to be right up my alley for an extra safeguard while still letting is “mostly run unattended”. In proper agentic workflow, I’ll probably just ask Cursor to quiz me on what I’d like added into that file :slight_smile: