Then you list all of the things you want it not to do and construct a prompt to audit the codebase for the presence of those things. LLMs are much better at reviewing code than writing it so getting what you want requires focusing more on feedback than creation instructions.