llms.txt & raw markdown endpoints

Red Glare exposes two AI-accessibility endpoints by default:

/llms.txt — index of every non-draft page following the llmstxt.org spec.
/<slug>.md — raw markdown source for every page, served alongside the rendered HTML.

Together they let AI assistants pull your documentation as-authored without screen-scraping HTML.

It's on by default

redGlare({
  llms: true, // default
})

Set llms: false to opt out — the routes aren't injected and the endpoints return 404.

What `/llms.txt` looks like

The llms.txt file is a structured index: site title + description, then a list of pages grouped by sidebar section. Example shape:

# Red Glare

> Astro integration for building USWDS-styled federal documentation sites with Comark.

## Getting Started

- [Installation](https://redglare.dev/getting-started/installation/): Install Red Glare into a new or existing Astro project.
- [Project structure](https://redglare.dev/getting-started/project-structure/): What the files in a Red Glare project do and how they fit together.
- [Quickstart](https://redglare.dev/getting-started/quickstart/): Write your first page, preview it locally, and ship a build.

## Configuration

- [Configuration overview](...): ...
...

AI crawlers (and humans, and feed readers) can hit this one file and discover every page on the site.

What `/<slug>.md` returns

For any rendered page at /some/path/, appending .md returns the raw source:

GET /getting-started/installation/       → HTML
GET /getting-started/installation.md     → raw markdown

This bypasses the rendering pipeline entirely — the response is the .md file as authored. Frontmatter is included. Directives are not expanded — the assistant reads them as ::alert{...} blocks.

Why it matters

AI assistants and search tools increasingly scrape HTML to answer questions about your docs. HTML scraping is lossy — it strips markdown semantics, directives become rendered HTML soup, and code blocks lose their language hints.

By serving the raw markdown alongside the HTML, you give tools a clean, structured copy of your content. The llms.txt index makes discovery trivial. The content is still your primary UI for humans; AI just gets a better view into it.

Excluding pages

Pages with draft: true are excluded from both /llms.txt and their .md endpoint (the .md URL returns 404 for drafts).

Turning off just `/llms.txt` but keeping `.md`

Not currently supported — they're a single feature flag. If you need one but not the other, file an issue on the repo.