A2WF — Agent-to-Web Framework#

A2WF is an open standard that defines what AI agents can and cannot do on a website via a machine-readable siteai.json file at the site root. Think of it as a legally actionable robots.txt for AI agents.

• Spec: https://a2wf.org/specification/
• GitHub: https://github.com/a2wf/spec
• Schema: https://a2wf.org/schema/core-v1.0.json

Commands#

1. Validate — Check a siteai.json#

Validate a siteai.json file against the A2WF v1.0 spec. The validator is local-only (no network requests).

bash

# Validate a local file
node {baseDir}/scripts/validate.mjs /path/to/siteai.json

# Validate from stdin
echo '{"specVersion":"1.0",...}' | node {baseDir}/scripts/validate.mjs --stdin

To validate a live website: Use web_fetch to download https://example.com/siteai.json, save to a temp file, then validate it.

Output: errors (must fix), warnings (should fix), info (suggestions). Exit 0 = valid, exit 1 = invalid.

If the user pastes raw JSON instead of a file path, save it to a temp file first, then validate.

2. Generate — Create a siteai.json#

Build a spec-compliant siteai.json interactively or from parameters.

bash

node {baseDir}/scripts/generate.mjs \
  --domain "https://example.com" \
  --name "Example Store" \
  --language "en" \
  --category "e-commerce" \
  --jurisdiction "EU"

When used interactively (no flags), ask the user:

1. Domain and site name
2. Category (e-commerce, news, healthcare, banking, restaurant, saas, other)
3. Jurisdiction and applicable laws
4. For each permission group (read/action/data): which operations to allow/deny
5. Rate limits, agent identification requirements, scraping policy

Output is a complete, validated siteai.json ready to deploy at the site root.

3. Audit — Analyze a website's A2WF posture#

bash

node {baseDir}/scripts/validate.mjs https://example.com --audit

Performs:

• Discovery check: Does /siteai.json exist? Correct content-type?
• Schema validation against core v1.0
• Permission completeness: Are critical permissions (scraping, account creation, data access) explicitly set?
• Legal coverage: Are jurisdiction and applicable laws declared?
• Extension check: Are companion extensions present and valid?

Outputs a structured report with a compliance score (0–100).

Quick Reference#

siteai.json structure (v1.0)#

json

{
  "specVersion": "1.0",
  "identity": {
    "domain": "https://example.com",
    "name": "Site Name",
    "inLanguage": "en",
    "jurisdiction": "EU",
    "applicableLaw": ["GDPR", "EU AI Act"]
  },
  "permissions": {
    "read":   { "productCatalog": {"allowed": true, "rateLimit": 60} },
    "action": { "createAccount":  {"allowed": false} },
    "data":   { "customerRecords": {"allowed": false} }
  }
}

Permission groups#

GroupControlsKey examplesreadWhat agents can readproductCatalog, pricing, reviews, faq, contactInfoactionWhat agents can dosearch, addToCart, checkout, createAccount, submitContactFormdataWhat data agents can accesscustomerRecords, orderHistory, paymentInfo, employeeData

Category templates#

Use --category with generate to start from a template: ecommerce, banking, healthcare, news-media, restaurant, saas.

For agent developers#

When browsing a website, check for siteai.json at the domain root before taking actions. Respect the declared permissions — violations are documented and legally actionable under the declared jurisdiction.

See the full Agent Implementer Guide: {baseDir}/references/implementer-guide.md