Get access

Standards for verifiable agent actions

Aegis, by Salanor — the provenance and liability layer for AI agents. Salanor publishes machine-readable contracts so builders, auditors, and regulators can verify agent behavior without trusting a single vendor UI.

APS-1 · version 0.1

Agent Provenance Standard (APS-1)

APS-1 defines a signed JSON event envelope for consequential agent actions: tool calls, model invocations, policy decisions, human approvals, and trace spans. Events are canonicalized (JCS), signed with Ed25519, chained per trace, and anchored for third-party verification.

  • Wire format — stable field names (schema_version, trace_id, action_kind, …) independent of product branding.
  • Signing — digest over canonical JSON; sig_alg + sig_value_b64 on the event.
  • Policy surfacepolicy_decision and optional policy_id for allow / deny / obligation semantics.
  • Interop — TypeScript reference SDK, JSON Schema, and conformance vectors; publishable as a standalone open-source package.

Identity · version 0.1

did:agent

did:agent identifies autonomous software actors in a tenant namespace. Salanor resolves DIDs to signing keys, policy scope, and console metadata. Each organization owns a namespace; each deployed agent receives a stable DID and rotatable keys.

  • Formatdid:agent:<org>:<agent-slug>
  • Binding — APS-1 events reference agent_id and key_id tied to the DID document.
  • Rotation — new Ed25519 keys without changing the logical agent DID.

Specification index

DocumentVersionLicense
APS-1 event envelope0.1 (published)Apache-2.0
did:agent method0.1 (published)Apache-2.0
Transparency / inclusion proofsAppendix (Aegis product)Apache-2.0

Implementer feedback: standards@salanor.com or the contact form (topic: design partner).

← Back to home