Using @ to insert files into context no longer actually inserts the files into the context.
Sometimes Cursor will read some of the files, 100 lines at a time; other times the files are ignored entirely.
I have to instruct the model to read all the context files and tell it my actual question will come in the next turn, and even that is not foolproof.
Here are some forum threads reporting the same issue.
I use latest version 0.46.11 with Agent and sonnet-3.7-thinking.
Previously, I was confident adding relevant files to context by mentioning them. Now, with partial reads, I suspect that this doesn’t happen anymore: if I mention a file which I think is necessary, Cursor doesn’t add its content to context.
To bypass that, I copy the file content manually, insert it to my chrome’s search tab (to clear it from traces of being uninsertable), copy from there and paste it to chat. This way I can cont…
When I added files to context in composer mode - the content of those files would be passed along and it would be able to read the context
In agent mode, for whatever reason, if I add files to context, it then decides it should search for those same files (as if it can’t see them)
In composer, if I tag 13 files (my routes, controller, several example frontend components) - it sees all of them and follows the pattern
Do the same in agent mode - it uses file search to file the controller, reads…
Title says it all, this is a ridiculous waste of time. I’d rather pay more then wait for whatever type of token-saving exercise this is.
Hello,
For several days now my chat no longer reads the files I give it access to. I’ve tried restarting severla times. I can only make it work by pasting into my apple notes and then repasting into cursor, but this is annoying. Please help.
[image]
I’ve found a workaround. In a fresh chat I do the first prompt using Edit mode to capture entire files by mentioning. Then, if the generation isn’t yet doing 100% what I needed it to do, I switch this chat to Agent mode and continue with entire files in context already.
Yes, there are hacky workarounds, but this is a core feature that is broken, and needs to be fixed.
system
April 12, 2025, 1:05am
4
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.