Intent-Layer: A Minimal Protocol for Agent-Readable Websites

Author: Lee Manley
Spec URL: https://intent-web.io
Repo: https://github.com/greenhill32/intent-site
Version: Draft 001

🧭 Purpose

The Intent Layer is a lightweight convention for exposing machine-readable intents on websites. It allows AI agents like ChatGPT, Claude, or Grok to:

• Understand what actions a site supports
• Fetch structured, real-time status
• (Eventually) trigger those actions securely

Inspired by API manifests like package.json or openapi.yaml, this is a minimal format optimized for read-only crawling.

🧱 Structure

agent.json

A flat JSON manifest describing supported actions:

{
  "name": "get_status",
  "method": "GET",
  "endpoint": "/status.json"
}

You can also support multiple intents:

[
  { "name": "get_status", "method": "GET", "endpoint": "/status.json" },
  { "name": "book", "method": "POST", "endpoint": "/api/book" }
]

status.json

Structured data that reflects dynamic or real-time state — such as availability, offers, or links:

{
  "available": true,
  "message": "Currently accepting new projects",
  "booking_url": "https://cal.com/lee"
}

Optional: Schema.org Support

Use semantic HTML tags or embedded JSON-LD to enhance AI understanding. For example:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Person",
  "name": "Lee Manley",
  "url": "https://intent-web.io"
}
</script>

🧠 Real-World Use Case: Creator Agent

Imagine an AI that wants to:

1. Visit your site
2. Read agent.json to learn you support bookings
3. Fetch status.json to check availability
4. Present that to a user

It all works without scraping, parsing, or guesswork.

🌐 Goal

Let the web become agent-actionable by default — with zero backend or complex APIs. If you can host a static file, you can support AI.

This is just the beginning.

For updates, follow @web_intents and visit intent-web.io.