Agent Definition Protocol (ADP) + ESP — A Portable Standard for Cross-Framework AI Agents

Feedback
,
1

Hello n8n Community,

I’m working on an open, vendor-neutral standard called **Agent Definition Protocol (ADP)**together with the Execution Semantics Profile (ESP).

The goal is to introduce a portable, framework-agnostic way to define, package, execute, and govern AI agents — doing for agents what OCI containers did for applications and ONNX did for ML models.

What ADP + ESP provide

• A portable agent manifest (runtime backends, tools, flow graphs, evaluation suites, governance metadata)

• An OCI-based packaging format (ADPKG) for registry distribution with signing, SBOM, and provenance

• ESP: framework-neutral execution semantics covering state passing, flow traversal, tool binding, model/prompt resolution, and error handling

• Cross-language SDKs (Python, TypeScript, Go, Rust) with conformance fixtures

Current State

A draft specification is available for public review and technical discussion, including:

• Implementable schemas

• OCI artifact layout

• Conformance scenarios

• Reference SDKs

Repo on Github: GitHub - casabre/agent-definition-protocol: A cross-framework, implementation-agnostic standard for defining, packaging, and deploying AI agents. ADP provides a portable agent manifest, a container build spec (ACS), and an OCI-based agent package format (ADPKG) for reproducible, governable, A2A-compatible agent deployments.

What I’d love feedback on

• Would a standardized agent definition layer help portability across frameworks and platforms?

• Are there core agent concepts you feel are currently missing from the ecosystem?

• Where do you see the biggest interoperability pain today?

• Would a portable agent definition layer be valuable for workflows?

• Where would n8n naturally map: as a runner, orchestrator, or tool host?

• Would standardized agent packages improve reuse inside automations?

If useful, I’m happy to publish concrete examples showing agents defined once and executed across different runtimes or orchestrators.

Looking forward to ecosystem input — the goal is to reduce fragmentation and enable truly portable agents.

Best regards,

Carsten