It's very interesting that I can write short, generic, single-sentence instructions which meaningfully improve the performance of a coding agent. Shouldn't Anthropic/OpenAI be writing these instructions in the system prompt for me?(Casey)