Summary
The scanner currently rewards HTTP-level Markdown availability only: Accept: text/markdown content negotiation and Link: response headers. That excludes the large class of sites on static hosts (GitHub Pages, etc.) that cannot set custom headers or negotiate content — even when they fully expose their content as Markdown via the established static conventions.
Request
Detect and (at least partially) credit the static-host equivalents of these signals:
- Sibling Markdown files — a
page.md (or page.html.md) next to each HTML page (the llmstxt.org per-page convention).
/llms.txt (and /llms-full.txt) at the site root — a Markdown index/dump of the documentation, which is what most AI crawlers and the llms.txt ecosystem actually consume.- HTML
<link rel="alternate" type="text/markdown" href="…md"> in the page <head> — the static-host analogue of the Link: header.
Why
These are the officially recommended workarounds for static hosting (see Vercel's and others' guidance) and are what tools like Claude Code / Cursor use in practice. A site that publishes /llms.txt plus side-by-side .md is materially agent-ready even though it cannot emit the HTTP headers; scoring it as a near-total fail under "Content Accessibility" understates its real readiness and discourages the most accessible, zero-infra path. A distinct sub-score ("Markdown available — static files" vs "via content negotiation") would reflect reality better.
Example
ocsigen.org publishes /llms.txt, per-project llms.txt / llms-full.txt, a sibling .md for every page, and <link rel="alternate"> in each page head — yet it scores as if no Markdown were available, solely because it is hosted on GitHub Pages (which cannot set response headers or negotiate content).
Summary
The scanner currently rewards HTTP-level Markdown availability only:
Accept: text/markdowncontent negotiation andLink:response headers. That excludes the large class of sites on static hosts (GitHub Pages, etc.) that cannot set custom headers or negotiate content — even when they fully expose their content as Markdown via the established static conventions.Request
Detect and (at least partially) credit the static-host equivalents of these signals:
page.md(orpage.html.md) next to each HTML page (the llmstxt.org per-page convention)./llms.txt(and/llms-full.txt) at the site root — a Markdown index/dump of the documentation, which is what most AI crawlers and the llms.txt ecosystem actually consume.<link rel="alternate" type="text/markdown" href="…md">in the page<head>— the static-host analogue of theLink:header.Why
These are the officially recommended workarounds for static hosting (see Vercel's and others' guidance) and are what tools like Claude Code / Cursor use in practice. A site that publishes
/llms.txtplus side-by-side.mdis materially agent-ready even though it cannot emit the HTTP headers; scoring it as a near-total fail under "Content Accessibility" understates its real readiness and discourages the most accessible, zero-infra path. A distinct sub-score ("Markdown available — static files" vs "via content negotiation") would reflect reality better.Example
ocsigen.orgpublishes/llms.txt, per-projectllms.txt/llms-full.txt, a sibling.mdfor every page, and<link rel="alternate">in each page head — yet it scores as if no Markdown were available, solely because it is hosted on GitHub Pages (which cannot set response headers or negotiate content).