Agent files are linked in-page

Shipping an llms.txt, sitemap.md, or AGENTS.md is necessary but not sufficient. Handed a bare URL, an agent crawls the HTML it is given and may never probe for those files on its own. A visible "For agents" link block is what lets an agent discover the files at all. Treat this as a discoverability signal, not a promised token saving: the efficiency win is real only when the agent follows the link, which is not guaranteed on every run.

site Discoverability impl 1.0.0 discovery.in-page-link

How the check decides

The check runs once per site, after every page has been fetched. While each page’s HTML is still parsed, it looks for an in-DOM <a href> that resolves to an agent-discovery file: /llms.txt (or /llms-full.txt), /sitemap.md, /AGENTS.md, or that page’s own .md mirror. It then grades by where such a link lives:

Pass: a top-level page carries an in-page link to at least one agent file. “Top-level” means the homepage (the site’s root URL) or a first-level path such as /docs. An agent landing on either can find your files.
Warn: no top-level page links to an agent file, but a deeper page (for example /docs/agents) does. An agent that lands at the top still can’t find your files, but a crawler eventually will.
Fail: no crawled page links to any agent file. The files, if they exist at all, are undiscoverable in-page.

Only links present in the served HTML count; a link injected by client-side JavaScript does not, because the agents this check is for do not run JS. In single-page mode there is no cross-page view to tell the homepage from a deep page, so the check returns N/A.

How to implement it

Add a small, visible “For agents” block to your homepage (a footer section or header link group works well) that links the agent-discovery files you publish:

Pass

<!-- https://example.com/ (the homepage) -->
<footer>
  <nav aria-label="For agents">
    <a href="/llms.txt">llms.txt</a>
    <a href="/AGENTS.md">AGENTS.md</a>
    <a href="/sitemap.md">sitemap.md</a>
  </nav>
</footer>

The root page links the files, so an agent that lands there can find them.

Fail

<!-- https://example.com/ -->
<!-- /llms.txt and /AGENTS.md are published, but no page links to them -->
<footer>
  <a href="/about">About</a>
  <a href="/contact">Contact</a>
</footer>

The files exist but nothing points to them, so an agent never requests them.