index-ai-validator

Appearance

Scope ​

This is the single source of truth for what @hardmachinelabs/index-ai-validator does, what it does not do, and what it deliberately is not. Other pages link here instead of repeating it.

The validator is honest by design. It measures a specific, public layer and reports what it found. It does not inflate that into a promise.

What it checks today ​

The current package validates index-ai Level 1 and Level 2a through validateIndexAi() and the index-ai CLI:

  • Level 1 AI Manifest: canonical fetch at /.well-known/index-ai.json, fallback at /index-ai.json (with a warning), JSON content type, JSON parse, schema shape, and an identity.domain host check
  • The manifest access.agent_index declaration
  • Agent Index graph: fetch, JSON content type, JSON parse, schema, nodes array, and rejection of the deprecated pages array
  • Per-node llm_url: structural validation, fetch, and clean endpoint content type (text/markdown or text/plain)
  • Hard HTML leak detection, with tolerated soft inline markup reported as a warning
  • content_chars in exact and max modes, using Unicode NFC code-point counting
  • Conservative security heuristics: secret-shaped values, sensitive variable names, and private infrastructure references in public AI-facing content
  • Shallow discovery hints on the homepage, robots.txt, and /llms.txt

The highest structural level the validator emits is level-2a. See Conformance vs Passed.

What it does not validate ​

The package does not perform:

  • full security audits or vulnerability scanning
  • discovery crawling, sitemap validation, or DNS TXT discovery validation
  • fixture validation
  • Level 2b relations
  • Level 3 MCP

level-2b and level-3 exist as reserved values in the result type, but the current validator does not emit them.

What it is not ​

@hardmachinelabs/index-ai-validator does not:

  • certify compliance
  • guarantee AI traffic
  • guarantee SEO ranking
  • provide legal control over AI agents
  • prove that a site is safe

index-ai is not a formal standard. It is an experimental specification and validator, built in public to explore how websites can expose cleaner, cheaper, and more reliable content surfaces for AI agents.

The HTML report carries the same disclaimer: it is for local or shareable human review, not certification, legal compliance, a traffic guarantee, an SEO ranking guarantee, a security audit, or a vulnerability scan.

Why the honesty matters ​

A validator that overclaims is worse than no validator. The value here is narrow and real: the agent-facing layer of a website becomes testable, with deterministic checks and an exact, auditable size metric. Everything on this page is enforced by the code, not by marketing.