AIDP — Agent Intent & Delegation Protocol

A protocol for explicit, auditable, and constraint-aware agent delegation.

AIDP defines a minimal, transport-agnostic protocol for expressing what an agent is allowed to do, under which constraints, and how failures are observed and handled, when acting on behalf of a user or another agent.

Read the Internet-Draft Problem Statement Example Flow

Problem Statement — Why Agent Delegation Is Currently Unsafe

No explicit declaration of agent intent
No machine-verifiable constraint boundaries
Failures are opaque or silently retried
No shared semantics for delegation across agents
No auditability between autonomous components

What AIDP Introduces

Explicit Intent — structured declaration of what is delegated
Declared Constraints — budget/scope/time/policy limits that are observable
Execution Boundaries (EB) — verify compliance before/after execution
Observable Outcomes — success/failure/violation are explicit

What AIDP Is Not

Not OAuth / authentication / authorization
Not tool invocation schemas or APIs
Not reasoning frameworks
Not model-specific safety techniques

Example Flow

Upstream declares an Intent with explicit constraints
Agent submits intent to an Execution Boundary (EB)
EB returns an Observation (success / failure / violation)
Agent proceeds, retries, or aborts based on the observed outcome

Status & Participation

Specification: Individual Internet-Draft

Feedback: Critical technical review welcome.

Internet-Draft: draft-vandoulas-aidp