I love the Composer, but it still has a few issues that prevent me from switching for good from the regular chat workflow:
I can’t edit my previous messages to create a new branch of the chat from there (I know it’s more involved as file changes would have to be handled too). The new message history helps a bit, but is not a substitute for branching.
I still can’t ‘#’ folders (I complained about this a while ago). I can ‘@’ folders though not the same ofc.
I need a side-docked composer option. [My reason for that is that I often use the vscode’s ‘Problem’ tab to quickly spot llm mistakes (I’m using ts, and type mistakes are frequent) and be able to screenshot and paste them to be fixed (that’s the best way I found in regular chat too, by the way, as the ‘Fix with AI’ opens a brand new context which is not optimal). This workflow is just too cumbersome with the floating window and impossible while in the Composer Control Panel.]
When I paste an image in the composer area, if it is high res it just takes a lot of space beyond my screen borders, and the composer windows becomes unusable for me.
I hope this helps. I really like the direction the Composer is going!!
it’s a new feature that let’s it make files and do things beyond the scope of one file at a time. Similar to what claude engineer is doing.
Go to Cursor settings → Beta and turn it on
I agree, this is amazing func … needs more UX iteration…
I’ll add another:
It’s unclear what I should use when… ie, if I’m using composer to do stuff, should I continue my chat there? or go back to the Chat?
why not integrate them?
Why is there a # option in Composer, but not Chat?
You can enable it in the Cursor Settings > Beta area.
The interface changed in the most recent Cursor update (0.39.0) in order to add some features users were requesting, such as modified file navigation and composer chat history etc.
I took some screenshots of its basic functionality about a month ago, before the recent interface update.
Yeah. I fully agree. This is a repeating pattern… Cursor throws out new features and at first I’m often puzzled when or how to use them.
One more paragraph in the release notes would be highly appreciated
Also with @ and #. Why is there both? When using # the file is shown in the UI above, when using @ nothing happens. I assume when using # you mark them for edit, when using @ you just add context. But it is still a bit unclear since I cannot see the @ files in the UI I tend to use # anyway…
Thanks for all the feedback, these are very helpful
I am deprecating # for files in favor of either picking files through the file picker (pressing +) or through @-referencing files, and I will iterate on the UX with all these suggestions!
How soon are the codebase index embeddings updated? it seems like it takes a few minutes (or some actions) every time before the Cursor index and “composer” know about your new file changes, not immediately…?
I really love the feature! For now it seems like a pretty balance between too crazy full AI Devin that doesn’t work usually, and boring github copilot when u use AI only for autocomplete.
Regarding file-picker in the composer it doesn’t consider cursorignore and in big projects it becomes a mess.
I love it. The UX is still something to polish.
A simple and naive example: it’s hard to resize the window, and when resized to full vertical height, it sticks like that and it’s hard to change. At least you can close it with [CTRL + I + W]
This thing makes absolutely no sense and I’m still trying to wrap my head around it. I just cannot understand what the devs were thinking when they decided to duplicate the way you can import a file into the LLM context. Also, the plus (+) button seems to lack feature parity with the chat, as it cannot import a Folder or Codebase into the context.
Other things missing:
Allow to drag and drop folders/files from the Explorer to the Composer
An option when you right click folders/files in the Explorer that says Add to Composer context
The reason why I decided to separate the context is so that on the top of the composer bar, you can essentially “pin” context to work on. Say today I’m working on file A, I’d want it pinned instead of constantly having to refer to it. Yet, if file A is very similar to file B, I’d then @-refer file B inside my prompt in one of the instructions within a composer conversation.
I agree that this isn’t the clearest to users, and we’re actively working to make the experience better. Hope this cleared up some of your confusion!
In the latest version, Composer is docked to the site and I can no longer get it back to the weird overlay window. It had its quirks but I kind of got used to it. Any idea how to get back to that window view?
