AI-first metadata

Papyrus can generate llms.txt, llms-full.txt, JSON indexes, and graph data from the same public content set used by posts, tags, RSS, sitemap, and search. That keeps readers, crawlers, and AI tools pointed at the same canonical pages.

llms.txt is the short map: routes, titles, descriptions, and source paths. llms-full.txt is the longer corpus: the important public text from posts and docs, plus code examples that help agents understand how to use the theme.

A consuming site runs the same command against its own content during build.

pnpm papyrus-llms src/content/posts public "$SITE_URL"

Use the deployed site origin for SITE_URL so generated links match RSS, sitemap, robots.txt, canonical URLs, and social metadata.

When the site needs structured data for search or graph views, generate JSON indexes from the same source set, including /ai/graph.json:

pnpm papyrus-ai-indexes src/content/posts public/ai "$SITE_URL" public/demo/site-data.json

Those outputs can power static search records, related-content tools, or external agents without giving them a different content model.

Keep private notes out of public collections. If a post needs a direct route but should not appear in public discovery surfaces, mark it with hidden: true and make sure feeds, sitemap, search, AI indexes, and graph generation all use the same hidden-content rule.