What got in the way
A proposed site-to-agent contract is hard to assess if publishers cannot test discovery, manifests and indexing conventions on real websites. Without a deterministic validator, testing remains speculative and visual inspection fails to capture missing metadata or invalid schemas.
The capability I built
I published a TypeScript validator and CLI that checks whether a website exposes the proposed INDEX-AI layer and reports findings. It verifies endpoint discovery, manifest structure, robots declarations, and content indexing rules without requiring an agent runtime.
Engineering decisions
Expose failures in a repeatable validator rather than rely on visual inspection.
Automated, reproducible checks catch subtle schema drifts and missing fields that human reviewers overlook.
Treat the proposed spec and its reference tooling as separate deliverables.
Decoupling the validation tool from the core RFC ensures third-party implementations can audit against the specification objectively.
Make the result inspectable by developers and site publishers.
Output formats include clear terminal feedback and structured JSON so both human operators and CI pipelines can consume the diagnostic report.
Try the published package
Run the validator directly against any public URL without prior global installation:
npx @hardmachinelabs/index-ai-validator validate <url>Check the latest README and review any operation that touches real credentials or live infrastructure.
Evidence and documentation
Important note on scope: Validates a proposed contract; it does not certify compliance with an industry standard or ensure that a particular AI system will cite a website.