Stopping the agent when it is paused at a terminal command no longer stops agent!

[jrista (Jon)]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

This has actually been a problem for a few versions here. I don’t know if it first started in 2.2, or 2.1, but it stopped working somewhat recently.

It used to be that if you hit the stop button in the prompt area when the agent was paused at a terminal command awaiting confirmation or rejection (or add to allow list), that the agent would immediately stop, and place the cursor in the prompt editor.

Now, when you hit stop…for one, the behavior seems inconsistent. Sometimes it seems that just hovering over the stop button, results in the agent continuing. Sometimes it takes several clicks to stop, and then the agent continues.

The problem in either case, is that the agent CONTINUES!!! This is not the way this worked in the past. At one point, you guys had an explicit “Reject and Offer Alternative” button in the terminal command instance that allowed you to then instruct the agent what to do instead. I liked and preferred that. You then removed the “Reject/Alt” button, and stated that the stop button effectively did the same thing. Back then, when you hit stop, the agent would immediately stop and you could instruct what you wanted it to do with another prompt.

Now, the agent CONTINUES, it does not stop!! It will continue working and that is more often than not quite problematic. Because when I stop the agent from running a command, it is usually because I want to have it do something else, and NOT just continue with its work. Its problematic as well, because without the command executing, the output the agent expects is not returned, and its next steps are often performed incorrectly, or it performs next steps that don’t make a lot of sense.

The way this worked before was much better: The agent STOPPED, which is exactly what you instructed it to do, by clicking none other than: THE STOP BUTTON.

Steps to Reproduce

At a paused terminal command awaiting accept or skip, hit the stop button in the prompt input.

Expected Behavior

Agent should immediately stop and await further prompts.

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.2.43
VSCode Version: 1.105.1
Commit: 32cfbe848b35d9eb320980195985450f244b3030
Date: 2025-12-19T06:06:44.644Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Darwin arm64 25.0.0

Does this stop you from using Cursor

No - Cursor works, but with this issue

[deanrie]

Hey, thanks for the report.

This is a known issue, and the team is working on a fix. A similar case was discussed here: Agent ignoring stop button and calls to stop in the chat

To help the team prioritize, can you share:

• Request ID (chat context menu > Copy Request ID)
• A screenshot or a short video showing the agent continuing after you press Stop while it is paused on a terminal command

Temporary workarounds:

• Close the terminal manually (trash icon in the terminal panel)
• Restart Cursor if the agent isn’t responding
• As a last resort, close the chat tab and start a new one

Let me know if this is critically blocking your work, and we can raise the priority.

[jrista (Jon)]

FWIW, the agent is not hung or otherwise unresponsive. Additionally, I usually don’t pop out the terminals, so there is no trash icon to close the terminal, its just the streaming commands in the agent chat interface.

The problem is just that, hitting stop, does not actually stop the agent, when it pauses at a command waiting for skip or approve. When the stop button is clicked, the agent should stop, period, regardless of why the stop button was clicked. I actually have not tested if stop will properly stop the agent at other times…I don’t usually stop the agent, although sometimes I’ll interject anotehr prompt before the agent is done with the previous, if it starts to go off track…

[jrista (Jon)]

This is still an issue, and it is really a very severe issue. Clicking stop does NOT actually stop the agentI It will continue to operate, and continue to allow the model to execute additional tool calls, make additional changes to files, etc.

This is very “scary” as when I tell the agent to stop, there is a VERY REAL AND EXPLICIT REQUIREMENT that it STOP! Because I have identified a clear reason I need the agent to NOT CONTINUE whatever it is doing. Usually this is because it is doing something incorrectly…changing files in a way I did not instruct it too, or changing the wrong files, or executing terminal commands wrong, etc.

STOP should mean STOP. Period. This current state, which I think has been persistent for over a month now, mid-December at the very least if not before that, where stop is apparently just a “suggestion” to the agent, seems exceptionally risky, and does not reflect the meaning of the darn word. STOP should mean STOP! Fully, completely, do not cross the line, STOP.