I have tried a few methods, and I’m looking for some pretty basic rules like for example:
= Do not issue cd /project/directory for every single command
= always use x.y.sh to run a build
= never run a build in the background and always pay attention to output.
I will tell you the method I use… I don’t know that it is the best.. but I think its clever..
I ask the agent… The AI knows what it will respond to better than we do..
Prompt: Review the current cursor documentation on the web.. Rules | Cursor Docs .
Prompt: Using the web, identify best practice for .
Prompt: Given your previous findings, generate a cursor ai rule for to apply within the current repository.
Tweak: The new rule until you have it the way you like it. save and set the cursor guides for when each is used.
I know this may sound lazy or over simplified, but honestly, if you don’t know where to start, this method will get you pretty far.
As far as I understand, this is a Cursor IDE crutch, and not an AI initiative.
I don’t have any problems with this.
Now I’ve come to the conclusion that I have a thin wrapper run.ps1 containing a set of flags and help, which I feed into the context along with the prompt. This allows a curious Agent to read only the runner and not clutter the context with 1600 lines of build.ps1.
run.ps1 → build.ps1 → build.[your language] → your app
This ■■■■■■■■ can’t be completely disabled, but GPT-5.x successfully obeys this rule from User Rules.
이분의 말이 저에게는 더 맞다고 느껴집니다.
저도 커서의 규칙을 여러개 설정 해 두고 있는데 커서에게 물어봅니다.
“내가 어떠어떠한 규칙을 정의 하고 싶은데 그것을 너에게 규칙화 하려면 어떻게 해야 할 지 알려줘”
이렇게 물어보면 자세한 방법을 알려줍니다. 아니면 그 방법에 따라서 규칙을 혼자 정의 할 수 도 있습니다!