Agent markdown
what an agent reads instead of the rendered page
Every page is published a second time as markdown, beside its rendered page, for agents to read.[1] What the rendered page holds is described on Site.
Copies
A page's copy sits in the same folder as its rendered page, as index.md.[1] It holds the page's
title, subtitle, intent and whether it is a draft, then its body and references as written.[1] A link
in a copy to another page points at that page's copy, and a link shown inside code is left as
written.[2] The goals page's copy carries every collected intent.[3] The local server answers a
copy as plain text.[4] A page's Source view links to its markdown file, which opens the copy as
the file itself, so a person or an agent can read the page as pure markdown.[5]
Index
The root of the site holds llms.txt, listing every page's copy with its subtitle, in sidebar
order.[6] Its shape is the one the llms.txt proposal describes: a heading with the site's name, then
sections of links.[7]
Every rendered page names its copy with rel="alternate" type="text/markdown", and the index with
rel="describedby".[8] Those are the link relations the proposal recommends for finding both.[7]
Audiences
A user build carries no copies and no index, because a copy holds the references and red marks a user build withholds.[9]