STANDARDS AND CONVENTIONS

INTEROPERABILITY STANDARDS ACROSS THE ECOSYSTEM Interoperability means the ability of different products to work together. Instructions files are one example, but there are bigger examples. Recall from Chapter 16 that MCP lets a tool built once be used by many different agent products. You met MCP in Chapter 39, and Chapter 45 discussed related protocols that let agents talk to each other, such as the Agent2Agent (A2A) protocol. A protocol is a set of agreed rules for exchanging messages. As of late 2025, MCP and AGENTS.md were both reported to be founding projects of the Agentic AI Foundation, and A2A was reported to be a separate project also hosted by the Linux Foundation. Treat such details as a snapshot, and check the foundations' own websites for the present situation. Why do these standards matter to you as a harness engineer? They lower switching costs. If your project documentation follows a shared convention, you can move from one coding agent product to another without rewriting everything. They reduce lock-in, the problem described in Chapter 49. They also create shared targets for security review, because a widely used standard receives more scrutiny than a private one, although you must still review third-party tools yourself, as taught in Part Nine. A word of caution: a standard being popular does not make it safe or correct. Always read the specification and the implementation you rely on, and never assume that following a standard automatically provides security.