The possibility to organize and share .cursor/rules is amazing, 2 key changes would be really awesome.
When a rules file is used by the agent, please allow it to not output the contents to the window (ideally just a small pill or the file could alternatively be added to the list of docs in the composer context.
Request 2 - have a way to designate some files to behave exactly as if it were a base project .cursorrules file or app level config.
I can set up the mdc description finally if using sonnet to almost always choose the file I want to always be included, but it does not seem to impact the full thread and generally is not has fluid in performance as if it were in the root cursorrules file.
The easiest way to see what I mean, I have created a personality file that imparts the persona of a fictional character that also operates with certain working rules. When its in the root, no matter how long the thread gets, it āstays in characterā always.
When instead it is in the rules folder, the convo starts, then reads in the file (output in the conversation confirms this) and will have the persona in the initial response, but throughout the thread it is not as imbued with the persona as if it were in the root.
Thats an awesome idea. Im struggling to separate though the āno matter how long the thread getsā fact from the actual idea.
There is for now in no model no such thing as āno matter how long the thread getsā. The thread has always a limit and no im not talking about context size because AI starts hallucinating way before the context limit of the model if you feed it conflicting information e.g. like debugging different issues during a session.
What you are asking is a .cursorrules file, right? Canāt the actual file just be used for this?
it has a simple .cursorrules file that gives the chat personality of a star trek character - and no matter how much back and forth (within reason) it will stay in character really nicely. will work in quotes, facts, reference events in the star trek universe tied to what we are doing in the momentā¦
now - it will basically understand it and follow the rule in response to the first prompt - but it no where near seems as well ingrained into the whole chat thread.
So my theory without knowing how it works under the hood with cursor, is for some reason the base .cursorrules file could potentially be constantly inserted into every request - whereas when it pulls from the rules folder, seems to be stuck at the beginning of the full threads context history or is processed differently in some significant way.
Hey, this is a great suggestion. I see the clear use case too. Your idea of designating certain files as persistent is technically solid. Iām not sure how this works on the backend but Iām sure thereās a reason for the variation in functionality.
A potential solution could be an option to āpromoteā a rule fileāessentially re-applying it on every request to maintain its influence throughout the thread. I hope we get feedback on this soon.
Here is an example of a feature that is similar to what you are mentioning which Cursor had to remove. You could pin things to a chat and keep it for every prompt. Perhaps a simplified focus persona part like OP mentioned would work if its short enough.
