This is about a system prompt that gets rid of the conversational style agents use and replaces it with structured cli/tool-like output. Been using this for a few weeks and it's a bit like a relief. Some colleagues use it too and are quite happy with it. Works best with pi coding agent and openai models. Anthropic models are not fully following and still embed the tool-like output in a conversational intro/outro sometimes.