middleware.js

// Markdown content negotiation (acceptmarkdown.com).
//
// Agents that send `Accept: text/markdown` get the markdown rendition of a
// page; browsers get the HTML. `Vary: Accept` is set on BOTH variants, without
// which a CDN can hand the cached HTML to an agent (or vice versa) depending on
// which variant happened to land in cache first.
//
// This is the one agent-readiness signal a purely static host cannot provide,
// because it requires code to run on the request.

// Catch-all except requests for files with an extension, so static assets
// (og.png, llms.txt, sitemap.xml, *.md) are served directly and untouched.
export const config = {
  matcher: ["/((?!.*\\.[a-zA-Z0-9]+$).*)"],
};

// Paths that really exist. Anything else is a 404, and an agent asking for
// markdown gets a markdown recovery page instead of an HTML one.
const KNOWN = new Set([
  "/", "/index.html",
  "/about", "/about.html",
  "/contact", "/contact.html",
  "/privacy", "/privacy.html",
]);

const NOT_FOUND_MD = `# 404 — page not found

That path does not exist on this site. Everything that does exist is listed
below, so you can recover without guessing.

## Pages

- [Home](https://defluffer.vercel.app/) — what defluffer is, a demo, and examples
- [About](https://defluffer.vercel.app/about) — what it does, how it is built, who made it
- [Contact](https://defluffer.vercel.app/contact) — support, bug reports, press
- [Privacy](https://defluffer.vercel.app/privacy) — exactly what data is handled

## Machine-readable

- [llms.txt](https://defluffer.vercel.app/llms.txt) — description and when-to-use guidance for agents
- [sitemap.xml](https://defluffer.vercel.app/sitemap.xml) — every indexable URL
- [robots.txt](https://defluffer.vercel.app/robots.txt) — crawl policy (all agents allowed)

## Install

- [Chrome Web Store](https://chromewebstore.google.com/detail/defluffer/ofaajilnnjfcinocgpljpmchpcnbhhln)
- [Firefox Add-ons](https://addons.mozilla.org/en-US/firefox/addon/defluffer/)
- [Source code, MIT](https://github.com/itayzit/defluffer)
`;

const MARKDOWN_FOR = {
  "/": "/index.md",
  "/index.html": "/index.md",
  "/about": "/about.md",
  "/about.html": "/about.md",
  "/contact": "/contact.md",
  "/contact.html": "/contact.md",
  "/privacy": "/privacy.md",
  "/privacy.html": "/privacy.md",
};

// Honour the quality values in Accept rather than a bare substring match, so a
// browser sending `text/markdown;q=0.1` after text/html still gets HTML.
function prefersMarkdown(accept) {
  if (!accept) return false;
  let md = 0;
  let html = 0;
  for (const part of accept.split(",")) {
    const [type, ...params] = part.trim().split(";");
    const q = params.reduce((acc, p) => {
      const m = p.trim().match(/^q=([\d.]+)$/);
      return m ? parseFloat(m[1]) : acc;
    }, 1);
    const t = type.trim().toLowerCase();
    if (t === "text/markdown" || t === "text/x-markdown") md = Math.max(md, q);
    if (t === "text/html" || t === "application/xhtml+xml") html = Math.max(html, q);
  }
  return md > 0 && md >= html;
}

export default function middleware(request) {
  const url = new URL(request.url);
  const target = MARKDOWN_FOR[url.pathname];

  if (target && prefersMarkdown(request.headers.get("accept"))) {
    const rewritten = new URL(target, url);
    return new Response(null, {
      headers: {
        "x-middleware-rewrite": rewritten.toString(),
        Vary: "Accept, Accept-Encoding",
      },
    });
  }

  // Unknown path: still a real 404, but hand agents a markdown recovery page
  // (site map + llms.txt) rather than an HTML one they have to parse.
  if (!KNOWN.has(url.pathname) && prefersMarkdown(request.headers.get("accept"))) {
    return new Response(NOT_FOUND_MD, {
      status: 404,
      headers: {
        "Content-Type": "text/markdown; charset=utf-8",
        Vary: "Accept, Accept-Encoding",
      },
    });
  }

  // Pass through, but still declare that this URL varies on Accept.
  return new Response(null, {
    headers: { "x-middleware-next": "1", Vary: "Accept, Accept-Encoding" },
  });
}