Understand what llms.txt can guide, what it cannot solve, and how to build a durable /agent/* layer around verified site data.
A guide is not a gateway
llms.txt can point an agent toward important pages and explain a site at a high level. It cannot bypass a WAF, repair broken navigation, expose missing product variants, or guarantee that referenced URLs are still valid.
Build the layer from verified facts
An agent-ready layer should be generated from canonical, validated site entities rather than a manually maintained text file.
- A concise llms.txt for orientation and priority paths.
- An agent index that advertises available machine-readable resources.
- Normalized product, service, policy, and organization data.
- An agent sitemap limited to canonical and reachable URLs.
- A capability document for supported actions and authorization requirements.
Keep it synchronized
The layer becomes harmful when it drifts. Catalog webhooks, CMS events, scheduled reconciliation, link validation, and signed revision history turn static files into a dependable operational surface.
Treat every publication as a deployment
Validate schemas and links, scan for secrets and personal data, publish atomically, verify the public endpoint, and retain the previous revision. Machine-readable content deserves the same engineering discipline as application code.
Run the ReachPlane access check to turn the concepts in this guide into evidence from your own website.