OpenA2A/specs
Lab/ATP

Make trust auditable, so an edit breaks the root

ATP is the trust-protocol face: it issues, distributes and revokes the trust assertions ATX carries, and appends issuance and revocation events to a public Merkle log. It is how the score AIP computes becomes auditable across the ecosystem.

Interactive lab · 4 of 5 in the familystart
the problem

Before the spec: the usual way

The usual way: trust is a private database row. No one outside the authority can audit it, and it can be downgraded or deleted with no trace anyone could detect.

the usual way

trust_db.agents["billing"] = { trustLevel: 3 }

Trust lives in a private row. No one outside the authority can audit it, and it can be changed, or quietly downgraded, with no trace anyone could detect.

What is real here

ATP is 1.1.0-draft (September 2026). The base trust-proof fields shown are the 1.0.0-rc1 ones; the 1.1 series adds the entry type registry and the leaf hash form, listed in the repository CHANGELOG. A revocation is a new log entry, never a deletion. Issuance and revocation events are appended to an RFC 6962 Merkle log; the Registry publishes signed tree heads and inclusion and consistency proofs that anyone can fetch and check; no monitor outside OpenA2A runs today. The Merkle root and inclusion proof here are computed with real SHA-256, so tampering is genuinely detected.

Copy as codeVerify a proof is in the transparency log (illustrative API)
// Illustrative API sketch: @opena2a/atp is not a published package.
// The executable ground truth is the conformance suite (7 byte-pinned
// fixtures, Go + Python verifiers):
// https://github.com/opena2a-standards/atp-conformance
import { getInclusionProof, verifyInclusion } from "@opena2a/atp";

const { proof, root } = await getInclusionProof(logIndex);
// Reconstructs the root from the audit path and compares (RFC 6962).
if (verifyInclusion(entry, proof, root)) trustIsAuditable();
Next in the chain5 of 5: Use a credential the agent never sees→