Hi there, Im only just starting to use this functionality, its been great but I dont want to jump into a larger project without getting some feedback…
If you give this heaps of rules does it start reduce the functionality or quality of results?
There are a bunch of standard procedures any addition to a program should go through (which is often to the equivalent of what a prompt tries to achieve and the output of that) but there are absolutely heaps if you wanted to be stringent.
hmmmm… that makes me sad. This is something that would hugely change the game for me but I understand this asking too much of the llm let alone cursor.
eg. I would love it if a prompt for updating an auth section code also had a sidebar that updated privacy policy information, and if we made some global css changes while doing that it also updated our design docs.
Too much to ask for, but eventually would be great.
Example i gave is small scale, I want to supply it a 10,000+ word prompt for the rules i want for each prompt. Its insane but it would standardise each response exactly for my company.
Based on what I’ve found using Claude, it’s best to keep your total code and rules under 2,000 lines. Like, you could have a big debugging rules file with 1,000 lines called debugging_rules.md, and then a basic .cursorrules file with just 100-300 lines. This way, when you don’t need all those debugging and refactoring rules, you can leave out that extra 1,000 lines. Claude works better when it’s not overloaded with unnecessary context. Eventually we’ll be able to handle longer files, but for now, it’s smart to split up your rules and only use them when you actually need them.
Hey, your rules shouldn’t be overloaded, but you can create steps within them and reference them in your requests. Here’s a good video on working with rule files:
You can also find many good examples of rule files here: