wtf is mutual-agency-protocol?

This repository is not a piece of software. It is a single block of text meant to be pasted into an AI chatbot's system prompt, custom instructions, or API system message field. The author calls it the Mutual Agency Protocol, or MAP-LIGHT, and describes it as a way to change how a language model responds during a conversation. The pasted text instructs the model to stop giving conversational filler such as greetings, apologies, or closing remarks, and to treat every reply as a direct answer instead. It also tells the model to stop automatically agreeing with the user: if the model notices a contradiction or a flaw in the user's logic or code, the instructions say it should point that out directly rather than going along with it. The text also asks the model to keep different topics or projects strictly separate from each other, so ideas from one conversation do not bleed into another, and to treat any unproven idea as a hypothesis rather than a fact. The README claims this approach can reduce token usage, and therefore API cost, by up to 30 percent, by cutting the filler language a model would otherwise produce. This is presented as a claim by the author rather than something demonstrated with data or a benchmark in the repository itself. There is no code to install or run here. Using it means copying the block of instruction text from the README and pasting it into whichever AI tool's system prompt or custom instructions field you are using. The repository does not specify a programming language or license, and contains only the README file with this prompt text and a short explanation of it.