ATTP: Agent Trust Transport Protocol

This document specifies the Agent Trust Transport Protocol (ATTP), a protocol-agnostic framework for trust scoring, cryptographic identity, action-limit enforcement, compliance gating, and tamper-evident audit for autonomous AI agents. ATTP separates trust from identity. Identity protocols answer "who is this agent?" ATTP answers "should this agent be allowed to perform this action, at this magnitude, against this counterparty, right now?" The protocol defines five progressive trust levels (L0-L4), per-agent ECDSA P-256 cryptographic identity, a five-dimension behavioural trust scoring model, action-limit tiers derived from trust scores, real-time compliance gating (sanctions screening, jurisdictional controls), kill switches for instant revocation, anomaly detection for agent behaviour, and a public trust query API. ATTP is transport-agnostic. This document defines the core framework. Protocol-specific bindings map ATTP trust enforcement to individual transports: MCPS provides the binding for the Model Context Protocol (MCP), with additional bindings defined for REST APIs, Google A2A, gRPC, and GraphQL. This specification supersedes draft-sharif-agent-payment-trust-00, broadening the scope from payment transactions to any agent action requiring trust-gated authorisation.