2.0 worktree ear all of my ssd
Not precisely sure whatâs going on, but I used to use Cursorâs agent occasionally to help me navigate file operations on my machine.
This used to work just fine, but now no matter what Auto-Run Mode (âAuto-Run in Sandboxâ or âRun Everything (Unsandboxed)â), it refuses to open a terminal in the Agent chat and write/run a script for me. It instructs me to âSave as your_file.ext and run the command shownâ, but thatâs exactly what I donât want; I want Cursor to be able to run everything in place, nice and concise in one place (the Agent chat).
Not sure what changed here, a bit confused. I even manually went into macOS System Settings and granted Cursor full disk access in case that was the issue, but to no avail.
Yep same thing happened with me. I was working 2 hours ago just fine using my selected claude models. Just now I tried resuming my work and the chat agent panel looked all janked. Like it didnt have the same single line with the check mark after it was done. It looks like the entire code is being shown as its changing and its width is extending beyond the element. Anyway I checked which models mine was using out of curiosity and what in the living f*3k, none of the claude models were even present and all replaced by random models like composer etc. Test your stuff before you launch this is instilling distrust in your so far great product.
The recent changes have negatively impacted my workflow as well to the point where I have downgraded to version 1.7.54.
Specifically, the following problems make version 2.0 difficult to use:
-
Loss of chat position and text selection: When switching from the editor back to the Agents tab, the chat window automatically scrolls to the very end of the conversation. It also removes any text I had selected. This completely disrupts my workflow and forces me to manually navigate back to where I was working.
-
Removed separate chat window: The feature to âpop outâ the chat and agent window, which was available in previous versions, has been removed. This is a significant step backward in usability, especially for users with multiple monitors who need to view the agent and editor simultaneously.
The combination of these issues has led to a much less efficient and more frustrating user experience. Until these features are restored or fixed, I am unable to use the latest version effectively.
For this reason, I have reverted to version 1.7.54, which retains the functionality I need.
2 Likes
Thanks!
1 Like
For some reason since I updated to v2 Iâm not able to put the `@` symbol in the chat screen. This is very weird and super annoying.
1 Like
If you go to Cursor > About Cursor (macOS) or Help > About (Window) then it says the vscode version.
VSCode Version: 1.99.3
I feel like that cursor is becoming an editor for vibe coders instead of an editor for real coders . And I donât like it
7 Likes
auto mode also removed the Haiku 3.5 and this was the best for auto mode.. ⌠NOW Is tottaly âfoolâ IA.. and i need to PUSH each interaction to Solve just one to-do (also is the context is under 33%) .. and i dont have as a model CHat gpt 5 .. but in auto mode just use this. Yesterday i tested disabling componer and worked for a few hours but not. It totally unusable if you have Pay mode but you need to wait until the renew period to âworks fineâ .. this is very dissapointing and maybe also illegal because is unusable and let the user to forcing to do another upgrade or buy more tokens.
1 Like
The same thing happened to me when I installed version 1.6, as the new one completely broke a functionality in my code. The cursor says it didnât make any changes to the code, even though it updated a master branch. Luckily, I had a backup on GitHub.
What ever you guys changed last night (UTC time) has absolutely destroyed productivity due to the Auto Agent becoming absolutely brainless - this was in v1.7.x. Until last night, looking at how the Auto agent responded, there seemed to be a mix of Composer and other responses. It has been AMAZINGLY accurate for the past 7 days or so. Iâm a heavy agent user and have been insanely productive over the period, with probably < 10% of agent instructions needing futher work. Today, EVERY SINGLE OUTPUT is garbage. Itâs literally generating the most brainless code. What did you change!? I erroneously upgraded to 2.x thinking that would help. RIGHT. Another steaming pile of changes to digest - and the agents are still borked. You guys are flying by the seat of your pants!
Hey! We still have the context ring, which shows rules that are active. Weâre also improving this in the next few weeks to show a detailed breakdown of rules/MCPs/prompt etc.
3 Likes
for me the snadbox issue resolve by running these cmd:
exec zsh
unset cursor_snap_ENV_VARS
unset dump_zsh_state
Itâs now clear the Cursor founders are not and never were experienced developers. They Just some young bucks trying to rewrite the rules for software development and democratize coding for the unwashed masses.
I had the day off yesterday and am afraid to even fire up Cursor today.
1 Like
Hey, one of the devs on the Agent Layout feature here!
Weâre working on allowing switching the primary side bar back to the left while in Agent Layout and will have it live in the next day or so.
2 Likes
What happened to the context bar? Or the ability to add active or tabs to context. Our strategy was having everything documented properly, load all necessary context, and clearly define what you want. Adding background context feels unnatural and limited now. Our code base is too large to simply allow the agent to determine only necessary context. Now we are back to having it unclear or not easily seen.
6 Likes
My workflow has been severely knee-capped because I can no longer add my console logsâ terminal window to the context bar, and then have all prompts be able to see those logs.
That particular workflow also allowed me to remove it from the context bar easily, if the AI was unnecessarily reading the logs.
Now I have to @ the terminal every time? I also noticed that @-ing the terminal in 1.x didnât work, it had to be in the context bar. Is there some other concept Iâm supposed to learn to have the same results?
1 Like
This will be very welcome, to see more info about how the context is being used. Repeating from my bug report, I think if there are any kinds of reserves they should be shown as well, so we see how the full available context window is used, rather than having an arbitrary reduction in the available context (which seems to be the case now, at least partly.)
1 Like
@FlavioScimeca That is my concern too. I hope Cursor recognizes that what makes Cursor different from the competition is that it is not designed around vibe coding. Its okay if they want to add a vibe coding view, but I hope they donât keep putting energy into satisfying that audience. Itâs a race to the bottom if they do.
2 Likes
This is a good point. If there is no context bar now, then how do you remove something from context. Based on my tests, it appears that later requests in the same chat chain still have the context from previous queries, as one would always expect.
So we either have to open a new chat with a fresh context if we want to remove a previous file from the context, or explicitly mention that file to be disregarded from context for this query, or explicitly mention the new context completely in each request to give it more weight than the previous queriesâ contexts. I think this is okay, and may have been how it always used to work when there was a context bar, so there is no real change.
The main question is (@andrewh) what happened before when you removed a file from the context bar that was in a previous query but you donât want it in the current query (same chat chain) and how would you perform the same action in v2.0.