Draft Note: AgentIDL — Interface Layer for Constrainable and Auditable Agents

The term “AgentIDL” has appeared in earlier discussions in groups such as
ai-kr and s-agent-comm, where related concerns around agent semantics and
communication were raised. This note consolidates and refines the concept
at an explicit interface and architectural level.

This note proposes an implementation-independent interface layer that
articulates a stable semantic boundary between natural language intent
and executable realization in agent systems.

The document is intentionally architectural and non-normative.

It does not define a framework, protocol, or runtime, and does not
assume any particular agent architecture.

The goal of sharing this draft is to invite discussion around:

- the positioning of an intent-to-execution interface layer,
- its relationship to existing agent frameworks and protocols,
- and the scope boundaries appropriate for such a layer within this CG.

see: https://w3c-cg.github.io/s-agent-comm/notes/agentidl-note

patches are welcome

Received on Monday, 15 December 2025 08:15:12 UTC