Coordination always entails some form of agreement. For human coordination, agreements are often largely implicit, but that won't cut it for ai agents. They need everything to be as explicit as possible. That includes their agreements as a whole, and also every single tiny aspect of their agreements: every term and condition. But, like humans, agents also need flexibility to construct their agreements to fit the idiosyncratic needs of their scenario. A protocol for agent agreements needs to be robust, explicit, and also highly modular. That's what I built for the @synthesis_md hackathon: an agent agreements protocol and a bunch of modular primitives. Here's how they all fit together.