I’ve also noticed that I can no longer select text within these generated code blocks in the chat. It was really useful to be able to just cut and paste specific lines and very much hope this is a bug, not an intentional change?
1 Like
Hey, thanks for the feedback. We’ll keep it in mind.
1 Like
With version 1.6.45 of Cursor, I no longer see the Apply button, forcing me to do copy and paste in “Ask” mode. Is this deliberate? I am working in “Auto” mode.
This has been going on for several versions now - why is it so difficult to have a tool that identifies whether the terminal is stuck / finished?
Or even simply having a button that says “Mark command as successful/unsuccessful” so we can manually unstick things without cancelling it and cursor understands the context?
2 Likes
Yea, I suggested this feature a while back. Vote and comment on it, maybe they will add it.
I found that using these dropdown buttons behaves in the way we want. Still would prefer it keeps the chat panel open when you close the last chat via the tab, since it’s less steps and more intuitive.
Version: 1.6.45 (Universal)
VSCode Version: 1.99.3
Commit: 3ccce8f55d8cca49f6d28b491a844c699b8719a0
Date: 2025-09-22T18:22:38.013Z (1 day ago)
This is absolutely broken - Grok 4 can no longer use Playwright MCP properly - tasks it was previously using well now completely fail
Now that the agent terminal is read only, can we at least have a ‘control-c’ type of button to press because the agent keeps stupidly running commands that never end and thus the terminal never returns when all it needsis a ^C, like it’ll really do dumb things like run tail -f logfile | head -10 and then the log file never gets anything so it just sits there forever. or things like vmstat/nicstat 1 without the duration so they just run forever, etc.. Without the read only terminal i just ^C it and it was fine.
1 Like
Version: 1.6.45 The Terminal has frozen.
The terminal issues are at the most severe they have ever been. Terminal in agent is 100% broken. DOA, dead in the water, just dead. You guys really need to fix this!!!
1 Like
You can usually click on the Skip button but the issue I get is when Cursor runs a package then forgets it started the process which is now hidden (unless I managed to show the terminal first).
Using pkill isn’t always the best way to terminate plus it’s way more cumbersome than Ctrl-C
1 Like
I have noticed for several versions not just latest that sometimes you have to open the Terminal cursor is working and hit Enter once. Just to nudge the prompt a line. I don’t know why, but it consistently works when cursor hangs even though the thing it executed in terminal did finish.
This doesn’t include things like executing commands that have paging (like if you used | more), where you need to help it with space space q , which are eye-roll worthy but their own category of issue separate than the above.
Problem is, I don’t have the icon to open the terminal window. I have no buttons or icons at all. The terminal panel exists, it shows if it was an allowlist command or not, shows the command, but nothing else. I cannot interact with the terminal panels in the agent at all. Its the most severe terminal issue I have encountered yet. Usually I would interact with the terminal by opening the instance, however that was fraught with problems in the past, and with the latest versions of cursor (i.e. 1.6.27 which I rolled back to) the terminal instances are also now readonly, which presents its OWN set of problems.
I have to say, the terminal interactions between agent, user and ide are the single most frustrating thing about Cursor. Terminal is a fundamentally critical part of a modern software engineer’s job, and its the least stable and functional aspect of cursor.
1 Like
Since moving back to 1.6.27, due to the perma-dead terminal issue in >1.6.27, I am still experiencing a lot of terminal hangs, stalls. It used to be that I could open the terminal and hit enter, and get things moving again, or kill the terminal and the agent would continue and try again. With these READ-ONLY terminals now, though, you have to stop the agent, kill the terminal (or, likely, ALL of them, as the agent now seems to somewhat arbitrarily start new ones whenever it sees fit, I’ve had as many as a dozen or so opened at once), then instruct the agent to try again, to which it will often say “SUCCESS” or “MISSION ACCOMPLISHED” and dive right into producing a summary for a half-completed job.
These terminal issues, are really quite severe, and I think the most significant issue Cursor has right now. Terminal is FUNDAMENTAL to modern day software development. It needs to work, it needs to work reliably, and it needs to work all the time, every day, day after day.
2 Likes
The same situation
I have found a couple other hits on this topic but I haven’t seen any solutions yet. the new read only terminal is flat dangerous to people with advanced settings. Or worse, for people who have no clue and just let the agent send it on yolo mode all day. The fact I can’t at least ctl + c in any agent terminal is a solid way to get rm rf’d out of a build.
I have to stop agents from running things daily, npm run dev after every build for example, it will just keep running new dev instances and never checks if one is running. I end up having to grep open ports just to kill them all now because cursor will just run dev every time it builds. the fact that it already will cat your .env, even if you have gitignore and cursorignore is bad enough, but the fact that the agent cannot be interrupted is not cool. even if you click kill terminal it doesn’t always kill the process, in fact it often doesn’t.
I always pop out agent terminals so I can see the entire process that it is running and interupt it at any second. there needs to be a read only toggle, not a forced read only. or, just put it back how it was, it was fine for months.