This issue has been tracked in several places recently, with very vague responses. This week, without any update, foreground commands suddenly started timing out. My team uses various scripts that can run for more than 10 minutes, sometimes 30 minutes or more.
This completely broke our productivity, because our agents could no longer self-correct or run our tests.
As an emergency workaround, I worked on forcing the use of “run in background,” but the agents then overused sleep and tail to check test results. This is far from optimal, and the AI loop becomes very fragile.
I also had to do reverse engineering and prompt engineering to force the agent to use a while loop and wait for an exit code. But suddenly, 24 hours later, without any Cursor update, the format of background terminals changed and once again broke our feedback loop.
Then we noticed that foreground commands no longer seem to time out again (once more, without any update to the Cursor app on our side).
We no longer understand what is happening. Our team uses Ultra plans, and we are faced with software that changes critical behaviors without any warning.
Is it possible to get a detailed explanation? Codex, OpenCode, and Claude Code do not seem to have this type of issue, which makes us consider switching systems, even though we really like Cursor.
All of this is very worrying, and in a company context it is simply not acceptable to have this kind of instability on our side for a 200$/month/user tool
If this is a bug or system issue, I suggest reporting it in the Bug Reports category or emailing [email protected].
Based on my experience, some issues come from the environment. In my case, I had network errors because my company uses a proxy. Sometimes I need to switch Network mode from 2.0 to 1.1 or 1.0 and run diagnostics to confirm the issue. I also use OpenCode in the terminal to check whether the problem is related to Cursor or the local setup.
Aside from that, as long as I use the stable version and avoid beta or nightly builds, I rarely see problems with terminal commands or agent stability. The main exception is when using the Gemini provider, which is a known issue acknowledged by Cursor.
Hi @Naufaldi_Rafif , we don’t have proxy or special env system manipulating requests.
I suspect A/B testing or feature flag controlled behavior here, but not controlled correctly from the cursor team or there provider, on the same cursor version a collaborator have a totally different UI.
Cursor team should help us here to understand there A/B testing and may be an option to opt out. We are not on Beta or Nightly.
My Cursor editor is so buggy that it usually crashes every 15–30 minutes, and the agents can no longer get much of anything finished because they crash every 5–10 minutes. At this point, I have no choice but to cancel the Ultra plan and switch to Claude Code, since work is not getting done.
Cursor stability has never been a strongsuit, but it’s been especially abysmal lately. ‘Rules’ and other features are constantly in broken states without disclosure, tool calls and sub agents freeze and timeout. Needs restarts a couple times a day because of threads that will get bricked. And Cursor support told me I should stay behind a few versions for stability, except the older versions were not stable to begin with, plus cursor pops up the update button right where your mouse might already be so I’ve literally updated on accident before.
I am also cancelling Ultra (I’ve used $500 in on demand usage as well) and will be switching to Antigravity, which has been significantly more stable.
I am in the same boat. Cursor seemed to be working well for me before the new year. Terminal commands ran in the shared terminal space with agents able to both read and write to the terminal, browser integration worked, agents were able to run ssh/scp commands successfully. Now I’ve spent the better part of the week trying to figure out where my functionality went and troubleshooting the issues, generally to no avail. I was highly suggesting this product to my coworkers but after the last 3 weeks the functionality is so spotty I’m spending 80% of my time fighting the tool. What happened???
You describe exactly the issue we’re facing. I don’t know what’s happening inside the Cursor company, but they seem to be following a “move fast and break things” strategy. I hope this is only a temporary phase of their product, because professional development teams need stable tools (stable servers, stable frameworks, stable languages, etc.).
An IDE is a core tool for a development team’s productivity. Every major change is critical and should be shipped with great care. Unfortunately, Cursor is currently struggling to release their product properly. There are too many bugs, frequent breaking changes in developers’ workflows, and some very strange shell behavior.
I really like the Cursor product, but their release process and QA checks seem extremely weak or possibly nonexistent.
I’ve been using Cursor for the last year and a half, and this kind of pattern where the Cursor team ships 5 updates per day every day with lots of instability has been very common as long as I can remember. I stopped using Cursor a few weeks ago, but I came checking back on this forum to see how things are going, and seeing what you describe here, it seems that things have not changed much, unfortunately.
Within the first year, i guess the instability could’ve been justified by being a young app, but after all that time to still have so much instability with the updates is not very professional.
It probably come from the sub-agents timing out when the model’s juggling too much context or hitting server-side glitches. It disrupts everything, right? Maybe try sticking to simpler prompts or restarting the session fresh to avoid the pile-up; could keep things moving until they roll out a fix.