Composable Trust Infrastructure for the Agent Internet: A Federated Architecture for Verifiable AI Agent Identity, Capability Discovery, and Secure Communication

This paper presents a composable trust architecture where six independent, open-source protocols compose into an end-to-end verifiable agent trust stack for autonomous AI agents. The architecture spans capability discovery (agent.json), identity verification (Open Agent Trust Registry), identity resolution (DID Resolution v1.0), encrypted transport (QSP-1), execution attestation (ArkForge), and delegation governance (Agent Passport System). Composition is achieved through three structural properties: a shared key type (Ed25519), a shared hosting convention (well-known files), and unidirectional data flow. The architecture has been validated through a multi-party working group with seven registered issuers, unanimous spec ratifications, and cross-implementation conformance testing across TypeScript, Python, and Swift.