<?xml version="1.0" encoding="UTF-8"?>
<reference anchor="I-D.sunyi-hacp-protocol" target="https://datatracker.ietf.org/doc/html/draft-sunyi-hacp-protocol-00">
   <front>
      <title>HACP: A Capability-Contract Protocol for AI Agents and Edge Hardware</title>
      <author initials="Y." surname="Sun" fullname="Yi Sun">
         <organization>Independent Contributor</organization>
      </author>
      <date month="April" day="30" year="2026" />
      <abstract>
	 <t>   HACP (Hardware Agent Capability Protocol) is a JSON-RPC 2.0
   transport- agnostic protocol that lets a Large Language Model (LLM)
   agent — or any program acting on its behalf — discover, plan,
   execute, observe, and audit operations on physical edge hardware
   (GPIO, I2C, UART, sensors, system telemetry, files) through a single,
   stable, security-checked surface.

   HACP sits below higher-level agent protocols such as the Model
   Context Protocol and above the operating system.  It is designed to
   make heterogeneous edge devices interoperable with diverse AI agent
   implementations without requiring either side to know the details of
   the other.

   This document specifies the HACP wire format, core methods,
   lifecycle, error model, security requirements, and audit semantics.
   Streaming, attestation, and MCP-bridge profiles are sketched as
   optional or preview.

	 </t>
      </abstract>
   </front>
   <seriesInfo name="Internet-Draft" value="draft-sunyi-hacp-protocol-00" />
   
</reference>
