Concepts

Agent passports

A signed, publicly verifiable grant naming what an agent may do — and revocable in one signed kill-switch entry.

An Agent Passport is an Ed25519-signed, Merkle-proofed grant that names what an agent may do: its purpose, the tools it may touch, its scope ceiling, its maximum delegation depth, the issuing principal, and an expiry. It is recorded as an entry in your Trust Ledger, so it is tamper-evident and independently verifiable.

Why it matters

Agents have no passwords to rotate and no human to vouch for them at runtime. A passport gives an agent a born-verified identity: a signed statement of authority that anyone can check from the id alone, with no access to your tenant. It is the difference between "we think this agent is allowed to do that" and "here is the signed grant that says so."

Public verification

Every passport has a public verification page at /agent-passport/[id]. Open it and the page re-verifies the Ed25519 signature and the Merkle inclusion proof — a tampered or revoked passport renders as does not verify, never as a green check. An auditor or counterparty can confirm a passport without ever touching your environment.

Revocation as a signed kill-switch

Revoking a passport writes a single signed revocation entry — an instant, provable kill-switch. Revocation status is published as a signed, compressed Token Status List (an IETF draft format) plus an agent CRL, so a verifier can check revocation offline and at O(1) without phoning home, and every set bit maps to a real Merkle-included revocation entry.

Where you see it

Issue, list, and revoke passports from Agent Authority (/agent-authority, badged NEW), which reads GET /api/agent-passport. The same screen also surfaces the MCP supply-chain pin and the public Trust Receipt.