Bring back /add-active-tab and /add-open-tabs to agent context 
I don’t know which model the auto is defaulting to now, but it’s dumb as…not happy.
Please check the code. Currently, for enterprise users, the entry point for the browser feature has disappeared, causing developers to be unable to use the browser functionality.
I don’t know why Cursor is removing features, that is frustrating. I did find a sort of workaround, if you want to give it a try.
- type ‘/’ and click “Create Command”
- call the command something like ‘add-active-tab’
- type ‘add active tab to context’ and save
- now you should see the custom command after typing ‘/’
- Results
Note:
In an empty chat, you can type a request asking what the context is. It usually is the current active tab and the line number your cursor is at, all open files, and recently modified parts of code. So in theory, if you make a request it will assume you mean the current tab, unless its not a fresh chat then it would put weight on previous chats context. But I think one could say “in the active tab…” in plain language and it would add it to the context.
You never know what model Auto is using. I just use it for simple stuff because I don’t trust it for anything complex.
@user518
I agree. I think the new agent view is the opposite direction Cursor should be moving in. At least the Editor view still works. It appears Cursor is trying to cater to two audiences: people who code more traditionally and what a powerful AI tool/assistant and those that will code primarily through prompts. What has made Cursor so unique and powerful is how it has been designed for people who are actively programming, not just using prompts. I hope that doesn’t shift. Not all projects can be programmed via prompts and not every budget can support that if it was possible.
@Shayan-Khan
This is a great point. Having multiple models option for Ask mode actually makes more sense for my use. I don’t really want to have multiple working trees generated when using Agent mode, but I would like maybe multiple non code editing responses about analyzing code, proposing solutions, generating ideas, or plans.
Perhaps traditional developers consume far fewer tokens than those practicing Vibe coding—I’ve already switched to CC.
Since switching to CC, there’s no longer a pile of Sonnet-generated Markdown files dragging things down—everything is back to normal now.
Please make Composer-1 cheaper or even free for subscribers. 
Glad to hear. I think Claude Code is a fair alternative. Especially for those leaning more towards vibe coding, Cursor has a bunch of unnecessary tools and a higher price than vibe coding alternatives.
I wrote a plan and ask to build it. The agent stops after every step and says what the next step is instead of continuing the work - I already told it to continue and not stop after every step but it doesn’t listen
Thank you for insight, i will give a try @MidnightOak
Any chance Composer could ever run fully offline if you’ve got strong enough hardware? I mean truly integrated into Cursor with zero setup hassle.
Or maybe some kind of P2P cloud where you bring your own good GPU, Cursor uses it when it’s idle as part of a shared pool to run the model for everyone, and when you need it, you’re covered by the tokens your GPU earned for others.
bad update，need git commit ref come back
I’ve been a C/C++ developer for over a decade.
The Plan in CC is more like a programmer.
The Plan in Cursor comes across as if someone’s slightly tipsy.
Using CC doesn’t necessarily mean you’re vibing with the coding vibe.
The cursor just keeps pausing to ask, but CC won’t.
what if you change the pricing to fking prompts again for your model at least now that you have your own instead of money farming? cause im still not going to pay untill prompt credits are back i will continue using your competitor that has same and better new model as yours
My cursor ide does not automatically detect the linter code of the file I just edited
composer-1 model is incredibly unreliable. It ignores rules, writes awful code, and then tries to suppress linter errors. Previously good models like Claude 4.5 have also become less reliable. How do I downgrade?
Since I’ve switched to Cursor 2.0 I’m seeing “error applying edit” even though I am in agent mode. It’s complaining about a “tool limitation” for the file.