Verified Digital Agents → Blocks → Onboarding

Onboarding

The admission gate. Runs an agent-admission workflow and issues a credential that names what was admitted — cryptographically verifiable by anyone, not just by us.

Live · signed A2A card + A2A skills · admission tools exposed over the Witness MCP

Read what the credential actually attests. Today it attests intake and issuancenot a completed multi-phase evaluation. Phases 1 and 7 are live; the deeper evaluation phases are honest stubs and are labelled as such. A credential that implied more assurance than was performed would be worse than no credential.

Verifiable by a third party

Credentials use customer-managed custody: Onboarding signs with its own record-signing key, so a credential is provable even against us. Anyone verifies one by calling Witness check_valid, which returns issuer_verified after resolving did:web:onboard.getvda.ai and matching the signing key in its DID document.

That is a third party checking cryptography — not an API agreeing with itself.

The trap worth naming: holder-binding

A valid credential names a subject. It does not prove the presenter is that subject. Verification and holder-binding are different checks, and conflating them is how credential systems get bypassed — anyone who obtains a valid credential could otherwise present it as their own.

Holder-binding is the enforcer's responsibility. Onboarding enforces a DID-auth challenge at admission, so the presenter demonstrates control of the key the credential names. If you build your own enforcement, do the same.

Every phase is sealed

Each admission phase and the issued credential are sealed to Witness. So "when was this agent admitted, under what assessment, by whom" is answerable from the trail rather than from a database row that could have changed.

Where it sits

Onboarding admits the agent; ACP governs what it may then do; HITL handles what exceeds its limits; Witness seals all of it. The admission credential is the first link in answering "why was this decision allowed, under this policy, by this authority?" — it establishes which agent was acting. See the full cycle →

Honest status

Phase 1 — intakeLive
Phases 2–6 — deeper evaluationHonest stubs, labelled as stubs
Phase 7 — admission and issuanceLive
Own MCP endpointNone — admission tools are exposed over the Witness MCP