In the Composer window, the user has to select multiple files one by one to build the context to the right amount of files necessary to address a particular request.
Even if the user manages multiple Composer threads with different actions, the selection process can be tedious and repetitive.
Consequences
Managing the Context and files submitted, can often result in repeat work done by Cursor with duplicate files that have been created previously, because the model did not know of relevant files existing–even when trying to do some grep searches for file names that it thinks a function or file should be named (often giving up after not finding what it was looking for or finding something and just reimplementing the solution).
Side Effects
I’ve noticed with a Cursor Composer conversation that was extremely long, I was starting to get frequent crashes. It did not subsume until I started a new Composer conversation–now with a blank history of previous work.
With Context File Sets, users could easily start newer smaller conversations easier, and potentially manage different threads of work (albeit not simultaneously), without full multi-agent functionality.
Have you tried just giving the filepath and name into the chat without attaching the file directly? For me thats usually sufficient or even using a class or function name from the file i want to be edited. So basically the actual content that needs change.
In case many files are attached as context to chat it does indeed pollute quickly the context and may cause it to hallucinate or edit the wrong file.
Something like the workflow which another user created works well for complex changes and reduces the context size plus prevents having to copy paste all the stuff into the prompt each time.
Maybe they can Introduce “Context File Sets”—predefined groups of relevant files users can load instantly. This reduces repetitive selections, minimizes duplicates, and prevents Cursor from reimplementing solutions. For crashes, implement autosave or session-splitting features to handle long conversations and just make them smaller who knows?
