Or if I’m using Cursor or a long-context LLM how would you recommend biasing generation or analyzing the code with this knowledge?
Or if I’m using Cursor or a long-context LLM how would you recommend biasing generation or analyzing the code with this knowledge?