Long context mode gone in newest update

Bug Reports
49

I think I’ve found a solution:

  1. Install “llacoste2000.llm-copy-files” (search vscode marketplace). Or perhaps there is a better extension. This one copies all regular open files (tabs), but it doesn’t copy diff editors, but neither did Cursor (when using /reference open editors). There is also an option to copy to a new file (which you can then edit before pasting into an LLM).
  2. (Optional) Create and pin a “Custom instructions” file, add anything to there and it will always be copied together with your other files.
  3. Paste into an LLM or assistant of your choice: for example: Cline with your own API, Genie (VSCode extension) with your own API. In both of these you can set unlimited tokens IIRC, ChatGPT, or any API dashboard.
  4. Paste the response from the LLM to either Composer agent (so it can apply the changes), or if it’s Cline it has its own Apply.

This essentially replicates Long Context Mode, the process stays relatively simple (if you can create a shortcut, quickly switch to an LLM and back with shortcuts), gives more flexibility (you can use any custom instructions), and more control - you can see exactly what you paste into an LLM. Apply should also probably work (although I haven’t tried).

I think this is a Win-Win situation. Cursor can focus on advancing their more popular features, and we can still utilize the functionality of Long Context with a few adjustments.

I’m still only beginning to use this alternative, but I can see it being more efficient and less buggy than the Long Context Chat was (where you had to constantly ensure all files get attached, not forget to switch to it from Normal Chat each time, only limited to Openrouter mostly when on your own API key, having to jungle between enabling and disabling “Use your own API toggle”), etc.

Possible improvements:

  • If you can find an extension which also copies Diff editors and works like ‘llacoste2000.llm-copy-files’ otherwise, please post it here.
  • If you find the best way to apply the output from an LLM, would appreciate it if you shared your experience. The Apply feature in Cursor chat was often broken (didn’t detect files correctly for example), so perhaps this will give more options and flexibility.

Update on extensions for copying open files:
I’ve found an extension that copies diff editors as well:
copyai.copyai

However, I had to modify its source code and install from folder, because it copied the diff editor’s left (original) content instead of the content on the right (which is what you usually need), and also I removed the logic to remove comments and new lines. I can’t post the modified code, as I don’t want to be responsible for distribution of extensions I haven’t vetted, but this is the closest thing I could find which after those modifications works perfectly for my needs.

Update on applying the changes:
I used the outlined steps and then used o1 model to create instructions for the Composer Agent. I then went back to Cursor and pasted the instructions I received from o1, and Composer Agent applied them mostly impeccably. This was just a single experiment, but this seems to work better and more efficiently than my previous workflow with the Long Context Chat.

7 Likes