Certificate of Action: A Trust Primitive for Verifiable Autonomous Legal AI Agent Workflows

AI agents increasingly take consequential actions on behalf of principals, yet there is no portable, verifiable artifact that captures why an agent acted, under what authority, and within what policy boundaries. Logs are mutable, traces are opaque, and post-hoc reconstructions break under adversarial conditions. We propose the Certificate of Action (CoA), a cryptographic governance primitive that binds an agent decision to its reasoning context, policy constraints, and authority scope at the moment of execution. The CoA is built on cryptographic primitives already deployed at scale in legally-recognized e-signature infrastructure — X.509 PKI, RFC 3161 timestamping, and ETSI-aligned advanced signature formats — extended to a new evidence class. We describe the CoA's six-layer structure, its relationship to existing evaluation frameworks for legal AI, and its implications for access to justice when individuals must contest or rely on agent-driven decisions. The contribution is a practitioner-grounded proposal that turns governance from a post-hoc audit problem into a verifiable artifact of execution.