Skip to content
aiyoplanePublic

Latest commit

 

History

4 Commits

Folders and files

Repository files navigation

AEAP — Aiyo Execution Authorization Protocol

An open protocol for authorization decisions at the boundary between autonomous intent and consequential execution.

AEAP defines a Runtime Decision Point that evaluates a proposed action against a policy authored by the party being protected, returns a ternary verdict (ALLOW, ESCALATE, or BLOCK), and emits a cryptographically-signed, portable Execution Receipt that any downstream system can verify offline. The protocol is agnostic to the identity of the acting party, the rail on which value moves (if any), the wallet or signer involved (if any), and the specific category of consequential action being authorized.

AEAP is designed for use at any boundary where an autonomous system — a human-initiated workflow, an AI agent, or a machine-to-machine process — proposes to perform an action with real-world consequences. Payment, DevOps deployment, procurement, data write, inter-agent delegation, and other consequential-action classes are all first-class use cases.


Where AEAP sits

The industry is converging on the recognition that identity, access control, and runtime policy decisions alone are insufficient to govern consequential actions performed by autonomous systems. AEAP addresses a specific layer of that problem.

Runtime authorization — whether an actor may perform a class of action — is well-served by existing identity, delegation, and access-control protocols: OAuth 2.1, OIDC, Okta XAA / ID-JAG, Microsoft Entra, Auth0, the Human Agency Protocol (HAP), Alibaba / Cisco Agent Operation Authorization (AOA), and OpenA2A AAP.

Execution authorization — whether a specific consequential action should execute right now, under the Protected Party's current Policy, against the specific Intent crossing the Composition Boundary — is the layer AEAP addresses.

AEAP's architectural primitives — Intent, Policy, Evidence Signals, Verdict, Execution Receipt, Composition Boundary — define a protocol-level vocabulary for making that authorization decision interoperable and the resulting authorization artifact independently verifiable. The protocol composes with identity, delegation, and runtime-policy systems; it does not replace them.

Independently proposed architectures in the same problem space — including A2SPA (Agent-to-Secure Payload Authorization), the IETF Adaptive Authorization draft's Execution Handle and Finality Sink primitives, SCITT Pre-Execution Permit, DRP, and EMILIA — are evidence that the category is becoming real enough that multiple architects are converging on the same missing layer. AEAP's contribution is a protocol specification whose semantics are intended to be interoperable across such implementations rather than a competing single-vendor mechanism. An AEAP-conformant implementation can consume authorization artifacts from these adjacent efforts as Evidence Signals, and an AEAP Execution Receipt can be consumed as evidence by them.

See §1 (framing and authorship-vs-authority), §2 (five invariants), §9 (Composition Boundary), §10 (deployment-model neutrality), and §11-§13 (HAP, Okta XAA / ID-JAG, and Alibaba / Cisco AOA Interop Profiles) in the specification for the normative treatment.


Current status

AEAP v0.9.2-DRAFT — public draft, published for review and comment ahead of v1.0. The v0.9.2 update is additive: it adds §15 Security Model and Responsibility Boundaries without changing any previously-defined element.

v1.0 is gated on the Appendix A conformance test vector artifact being authored and passed identically by at least two independent reference implementations.


Reference implementations

Two independent reference implementations of the RDP composition pattern demonstrate Core conformance. The two share no vendored code — they are maintained as separate codebases precisely so that passing the same test vectors from both demonstrates AEAP's interoperable implementability.

Node:

Python:

CI/CD:

macOS CLI:

Implementations shared by third parties are welcomed under the same conformance criteria that apply to the reference implementations. See the specification's §1.1 and §1.4 for authorship-vs-authority discussion.


Quick concepts

AEAP's architecture decomposes an authorization decision into named primitives:

  • Intent — a structured representation of exactly one proposed Consequential Action (§3)
  • Policy — declarative rules authored by the Protected Party specifying authorization conditions (§4)
  • Evidence Signal — information the RDP evaluates (value, provenance, attestation) in applying the Policy (§5)
  • Verdict — the ternary output: ALLOW, ESCALATE, or BLOCK (§6)
  • Execution Receipt — a cryptographically-signed, portable attestation of the authorization decision (§7)
  • Composition Boundary — the specific location where the RDP is invoked (§9)

Five invariants hold across every AEAP evaluation (§2.2):

  1. One authorization decision per Consequential Action
  2. The decision occurs strictly between Intent and Execution
  3. Policy is authored by the Protected Party
  4. Authorization decisions MUST be independently verifiable
  5. The RDP fails closed by default

Licensing

AEAP uses a two-license split that respects the different natures of specification prose and executable test content:

See each license file for the full terms.

Reference implementations in the Aiyoplane GitHub organization (npm, PyPI, GitHub Marketplace Action, Homebrew tap) are published under Apache License 2.0 in their own repositories.


About

AEAP was authored and is maintained by Aiyoplane, Inc. The authority of AEAP as a protocol is established by adoption, interoperability demonstrated by Appendix A conformance, and the collective practice of the implementing ecosystem — not by the identity of its author. See §1.1 of the specification for the full authorship-vs-authority statement.

Learn more: aiyoplane.com Trust surface and public artifacts: aiyoplane.com/trust Contact: info@aiyoplane.com


Verify First. Execute Second.