UHP: A Shared Control Plane for Agent Harnesses

Why it surfaced

UHP treats an agent run as a job rather than a model completion, with versioned OpenAPI and JSON Schema, a 52-check conformance suite, and a stated goal of making implementations interchangeable. Its governance and conformance language make this a useful specimen of how the agent-infrastructure layer is trying to establish records, rules, and compatibility.

The Unified Harness Protocol is a draft open standard for starting, monitoring, continuing, cancelling, and retrieving results from complete AI-agent harnesses.