With the current update, when I ask cursor to produce markdown files it fails to write the entire output to the file and instead outputs pieces of the markdown in composer (but not the file).
In the example below only the top section is written to the file.
It was working in earlier versions for me but it does seem to be a reintroduction of an old bug? (see: Composer writes some file updates in chat instead)
thanks for letting us know! we have a fix that will go out next version
If it helps - it might also be model dependent - I have not had this happen yet with claude - but deepseek, o1, gemini all seem to do this repeatedly. Also for some reason, o1 seems to only create a file if explicity asked even when its implied, it will just share all the output as a thought instead of as a file (never starts attempting to put it into a file.)
Thanks, I think you are correct. I am seeing the same problems with o3 already but not claude.
We are still working on tuning the performance of o3-mini to make sure it behaves like our other models, but this should be solved in a future update soon!
Fixing this requires a setting in the OpenAI API for o3-mini. Simon Willison cover this here …
I hope Cursor will make that change so we can get our markdown back.
