Fix: Markdown content-negotiation

When a user-tier agent (ChatGPT-User, Claude-User, acting live on someone’s behalf, the same tier crawl policy already distinguishes from training and search bots) requests Accept: text/markdown, return a lean markdown body instead of full chrome-heavy HTML. Fewer tokens spent parsing layout, more spent understanding content.

This is one of the agent capability signals shown for reference. It doesn’t affect your Agent Access Grade.

// Pseudocode for any route handler / middleware
export function GET(req) {
  const accept = req.headers.get("accept") || "";

  if (accept.includes("text/markdown")) {
    return new Response(markdownVersionOfThisPage, {
      headers: { "Content-Type": "text/markdown" },
    });
  }

  return renderNormalHtml();
}

Test it

curl -H "Accept: text/markdown" https://yourdomain.com/