Agent View Layer
ViewA site emits a text/agent-view companion for each page so AI agents read intent, state, and actions directly — no scraping, no guessing. The view layer for the agent web.
Spec + validator · live at agentviewlayer.org
open standards · the agent web
Open standards for the agent web.
Frontier Infra builds open, portable standards and tools for the agent web — infrastructure that works without a central trust root. AVL makes a site agent-readable, AAR makes an agent's actions accountable, and Claude Layers keeps coding agents disciplined.
The idea
Every funded “agent control plane” governs access and centralizes on a trust root it owns — the business model is being the center. That leaves the open problems unowned: how a site declares itself to an agent, how an agent's actions are proven after they leave any one vendor's boundary, and how a coding agent stays honest.
Frontier Infra builds those as open, producer-owned standards: portable, verifiable with a public key, and designed to compose — not to capture. AVL is the view layer, AAR is the proof layer, and Claude Layers is the discipline layer. Each stands alone; together they're the connective tissue of an agent web that no single company owns.
Projects
Three layers of the agent web — declare, prove, behave. Each stands alone; together they compose.
Agent Control Plane
ProofThe Agent Attestation Record binds an agent's claim to an independently-verified, evidence-committed verdict — checkable with a public key, no central server. The proof layer over the view layer.
AAR v0.02 · spec + zero-dep verifier, public
Claude Layers
DisciplineUniversal coding discipline + role overlays (Forge / Quill / Scout / Warden) + project values + verifiable /goal contracts with Warden sign-off — composed at runtime with clear precedence. A tool, not a website.
v1.0 · install via curl · code only (no site)
Principles
- Open & MIT-licensed. Specs and reference implementations live at github.com/frontier-infra.
- Producer-owned & portable. Verifiable with a public key over
did:web— no central server, no captured trust root. - Composable. Each standard stands alone and composes with the others, and with access-control vendors rather than replacing them.
- Self-demonstrating. These sites are built with AVL to L3 — fetch
/.agentor/agent.txt. We eat our own cooking.
FAQ
What is Frontier Infra?
Frontier Infra is a project that builds open, producer-owned standards and tools for the agent web: AVL (the Agent View Layer), AAR (the Agent Attestation Record / Agent Control Plane), and Claude Layers. Everything is portable, MIT-licensed, and designed to work without a central trust root.
How do AVL, AAR, and Claude Layers fit together?
They are three layers. AVL is the view layer — it makes a website agent-readable. AAR is the proof layer — it makes an agent's actions accountable after the fact. Claude Layers is the discipline layer — it keeps a coding agent honest while it works. Each is independently useful; together they cover declare, prove, and behave.
Why “producer-owned” and “no central server”?
The funded agent-control-plane market centralizes trust on a vendor. Frontier Infra's standards are verifiable with a public key over did:web, so the producer of a site or the signer of a record stays in control and proof survives leaving any one vendor's boundary. The standards compose with access-control vendors rather than replacing them.
Is everything open source?
Yes. The specs, the reference implementations, and the tooling are MIT-licensed on GitHub under github.com/frontier-infra. The sites themselves are built with AVL to L3 — they are their own conformance demos.
Where do I start?
If you publish a website and want agents to read it correctly, start with AVL (agentviewlayer.org). If you run agents and need provable outcomes, start with AAR (agentscontrolplane.org). If you build with Claude Code, install Claude Layers from its repo.