So I noticed that after a few times of asking me essentially “Should I move on to editing the next method?” that Cursor wasn’t following my rules, nor even my prompt which pointed specifically to the name of a rule file, anymore.
I asked it why, and in part it said:
Why Referencing Rules Wasn’t Enough:
- I didn’t maintain strict state between interactions about what phase we’re in
- I didn’t actively validate against the rules at each step
- I treated the rules as reference material rather than hard constraints
I noticed it only loads rules files at the beginning of a chat, but I’m really not sure what to make of this behavior. I guess really there are no ‘rules’ at the AI level, are there. Just info that came in with the rest of the context. And it seems to only look back in its conversation a short distance before not bothering with the older information.