I’m seeing the annoying behavior of the agent text input box losing focus when the agent requires input through the “Run command?” dialog. I always need to re-focus it with ⌘+Y to be able to accept the command with enter. Not sure if this is the same issue as OP has though. I’m running Cursor 1.5.9.
Previously, when Claude responded with a suggested action and the menu with Run / Skip / Reject appeared, pressing Return would activate Run. As of the recent update, the keyboard focus is pulled back to the editor window instead, so pressing Return just inserts a newline into the editor. This breaks the flow, since the natural muscle memory is to just hit Return to confirm.
Impact:
Slows down workflow, forces me to use the mouse or arrow keys to reselect the menu, breaks natural keyboard flow.
Steps to Reproduce
Ask Claude for a code change that triggers the Run / Skip / Reject menu.
When the menu pops up, press Return.
Instead of triggering Run, a newline is entered into the editor.
Expected Behavior
Pressing Return should trigger the default action (Run) (if I had just typed a question into Claude chat and pressed return and not shifted focus elsewhere)
Operating System
MacOS
Current Cursor Version (Menu → About Cursor → Copy)
When the Chat Agent reaches a pause for command approval the prompt window is now instantly disabled, which has the effect of shifting focus to the open text editor window and if user was drafting next prompt in chat prompt window they are now typing away in the code text editor window with breaking effect. This is a critical issue that needs to be rectified immediately. Users typing in chat prompt should never finish a sentence in the code text editor screwing up their code.
Below I said that Cursor still works but with this issue, however, if this is not fixed I will stop using Cursor until it is resolved. I dumped the last IDE I was using for a similar issue, and Cursor was great about that until the last few days when a new release introduced this major problem from unexpected behavior that destroys productivity and injects significant frustration and stress that harshes my mellow.
Steps to Reproduce
With a text file open in editor panel, enter a prompt in chat that causes AI agent to run a command that will pause for approval to proceed. Then keep typing a new prompt for when the prior response is completed. When the AI agent is paused waiting for approval to run the command, the prompt window will become disabled for editing and the focus will change to text editor window where user will be typing in the text editor what they intended to type in the prompt.
Expected Behavior
Do not disable prompt entry when waiting for user approval for AI agent to run command. Allow user to continue editing prompt and do not switch focus to editor window.
Operating System
MacOS
Current Cursor Version (Menu → About Cursor → Copy)
Version 1.5.9
VSCode Version 1.99.3
Commit: de327274300c6f38ec9f4240d11e82c3b0660b20
Date: 2025-08-30T21:02:27.236Z (4 days ago)
Electron: 34.5.8
Chromium: 132.0.6834.210
Node.js: 20.19.1
V8: 13.2.152.41-electron.0
OS: Darwin arm64 24.6.0
Additional Information
User cannot know when this is going to happen and really affects productivity to have to write prompts in a separate editor until the agent response resolves.
When auto-run mode is disabled, the agent pauses for permission as expected. However, instead of keeping focus on the command approval option, focus automatically shifts back to the IDE’s open file. This disrupts workflow, forcing the user to use the mouse to re-select the command option every time.
Steps to Reproduce
Disable auto-run mode in Cursor AI.
Trigger the agent to suggest a command.
Wait for the permission prompt.
Notice that focus jumps from the approval option back to the open file in the IDE.
Expected Behavior
Focus should remain on the “choose command” approval option so the user can quickly approve or reject with keyboard input, without being forced to click back into the prompt.
Screenshots / Screen Recordings
Operating System
MacOS
Current Cursor Version (Menu → About Cursor → Copy)
Is this just me? The little pop up command execution window I find gets in the way endlessly. The UX doesnt work. Either it will capture my enter when Im trying to send messages, or its prompting for a command I dont want it to run and I want to just type my message and have it go away. In short it feels like it’s always in the way and super clunky. Surely there must be a better pattern here? It’s now redirecting focus away from the chat window entirely often when it pops up, so that I have to re-click on the chat window to continue typing that was interrupted mid stream.
If I’m in the middle of typing something already, I think I want it to wait until I send my message before popping up and asking if it can run. If I’m typing a message Im already ahead of the chat and giving it new instructions or next steps.
If I’m not typing anything, then have it take over the chat input box for its request, so its clear and that there is only a single input area/enter action. Right now you have the same workspace with two different branching happening at once, its not a good design pattern.
When I’m waiting for Cursor Chat to finish, it often generates commands. Before, I could just hit Enter to allow the command, but now it automatically focuses the code editor, so pressing Enter just adds a new line in the code.
The same thing happens if I’m typing in the chat while the LLM is still generating. If it happens to create a console command mid-reply, my caret jumps into the code editor, and I end up typing there by accident. It’s super frustrating.
Steps to Reproduce
Open Cursor Chat.
Have it generate a reply.
Wait until it generates a terminal command.
At that point, the caret jumps into the code editor.
Expected Behavior
It should not focus the code editor, and it should be possible to press enter to allow the command.
Operating System
Windows 10/11
Current Cursor Version (Menu → About Cursor → Copy)
When cursor agents suggest to run a command, it suggest action in a drop-down menu: “run”, “skip”, “reject”. Those action have a keyboard shortcut. For instance “enter” will “run”.
But the focus is not the drop-down menu, so pressing “enter” actually create a new line in the currently open text tab.
To make “enter” key work, you need to manually click on the terminal command view in the chat. But it defeat the purpose, you had better directly click on the “run” button.
Steps to Reproduce
Request cursor to run any terminal command with yolo mode disabled.
Expected Behavior
When the terminal tool open the menu, pressing “enter” should run the command.
Operating System
Windows 10/11
Current Cursor Version (Menu → About Cursor → Copy)
Hey all - we think this is fixed in a release going out today (1.6.3 prerelease) or later this week stable. Enter works but focus was incorrectly shifted to the editor.