Agent context profile: proposed property names for principal, delegation chain and attestation state

We implement the AuthZEN Authorization API in an agent authorization product and have posted an Internet-Draft describing the semantics we carry through it: https://datatracker.ietf.org/doc/draft-mnki-agent-trust-profile/

Our PDP answers for autonomous agents, so subject and context need more than a user id. In practice we pass the agent identity and the principal it represents as subject properties, and delegation depth, effective capabilities, constraints (amount, currency, region), risk tier and verified runtime attestations as context. The decision returns an evidence list rather than a score, because the interesting question is always why.

Proposal: a small agent-context profile for AuthZEN with agreed property names for

• the principal an agent acts for (subject property),
• the delegation chain and its depth,
• effective capabilities and constraints,
• attestation state of the runtime,

so that agent PDPs are interchangeable. Happy to write it up as an implementer's draft if there is interest, and to contribute interop test vectors.

Implementation is open source (https://github.com/MNKIAgentOS/agent-trust, Apache-2.0) with conformance vectors; the hosted endpoint and client paths are at https://mnki.com/docs/integrations for anyone who wants to test against it.

Start by reviewing the proposed Internet-Draft and the AuthZEN API concepts described in the issue; no repository file or entry point is identified. Compare the proposed property areas with the existing conformance vectors and client or hosted integration paths mentioned in the issue. Done would require an agreed profile proposal and interoperability test vectors, rather than a localized code change.