I agree. There should be better official documentation explaining how to leverage this feature best. I have been using Cursor to create a cursorrules.md file based on telling it what I wanted and having it summarize several sessions for areas it struggled with and how we worked to get what I wanted. I then would copy the contents of the markdown file and paste into .cursorrules. After reading about the json option, I asked Cursor to review my rules and suggest improvements including a determination if it understood json better than markdown. It claimed that it did and proposed a JSON version. I am now trying the JSON version of the file and hope to see improvement. However, this is still a shot in the dark, guessing and assuming what will work best. It seems that the team who is developing Cursor should have a clearer idea of what the correct format, structure, and contents should be for optimal results.