feat: add llms.txt, llms-full.txt and per-page Markdown for AI readers

.gitattributes

@@ -0,0 +1,3 @@
+# llms-full.txt is generated from the pages by scripts/build-llms.mjs and mirrors
+# every content change, so GitHub collapses it in diffs. llms.txt stays visible.
+llms-full.txt linguist-generated=true

.mintignore

@@ -0,0 +1,3 @@
+# Build output. A stale export must not be picked up as pages or copied into the next export.
+dist/
+export.zip

CONTRIBUTING.md

@@ -20,6 +20,7 @@ to document one that doesn't work the way you've said.
 
 ```bash
 npm run check
+npm run llms
 ```
 
 This validates that every page in `docs.json` exists and every page on disk is
@@ -40,7 +41,7 @@ reachable from the navigation. CI additionally runs a link check.
 
 1. Create the `.mdx` file with `title` and `description` frontmatter.
 2. Add it to the right group in `docs.json`.
-3. Run `npm run check`.
+3. Run `npm run llms`, then `npm run check`.
 
 ## Security
 

README.md

@@ -12,6 +12,7 @@ Production is a Cloudflare Worker serving the static export as assets
 ```bash
 npm run dev      # mint dev, on http://localhost:3000
 npm run check    # validate docs.json navigation against the files on disk
+npm run llms     # regenerate llms.txt and llms-full.txt after editing a page
 ```
 
 `npm run check` is the fast gate — it catches a page listed in `docs.json` with
@@ -25,9 +26,25 @@ docs.json          navigation, theme, SEO
 introduction.mdx   … the Guide tab, one file per page
 develop/           the Develop tab
 images/  logo/     assets
-scripts/           check-docs.mjs
+scripts/           check-docs.mjs, build-llms.mjs
+llms.txt           AI-readable index of every page (generated, committed)
+llms-full.txt      the whole site as one Markdown file (generated, committed)
 ```
 
+## AI-readable copies
+
+The site follows the [llms.txt](https://llmstxt.org) convention. `llms.txt` at the
+repository root indexes every page with its one-line description, and
+`llms-full.txt` is the whole site as one Markdown file. Both are committed, so
+they read fine straight from GitHub, and both are served from the site root. The
+build also writes every page as plain Markdown beside its HTML, so appending
+`.md` to any page URL returns the Markdown.
+
+`scripts/build-llms.mjs` generates all of it from `docs.json` and each page's
+frontmatter, using only Node built-ins. Run `npm run llms` after editing a page
+and commit the result; `npm run check` fails when the committed copies are out
+of date.
+
 ## Contributing
 
 See [CONTRIBUTING.md](CONTRIBUTING.md). Docs for the app itself live in the

about-these-docs.mdx

@@ -40,6 +40,20 @@ outbound calls, what the helper can do, what the kill switch touches — that's
 not a docs typo, it's a security issue. Please follow
 [Security](/security) and don't open a public issue.
 
+## Reading these docs with an AI assistant
+
+This site follows the [llms.txt](https://llmstxt.org) convention, so you can hand
+it to whichever assistant you use and ask it to walk you through an install or a
+report:
+
+- `/llms.txt` is an index of every page with a one-line summary.
+- `/llms-full.txt` is the whole site in one Markdown file — the thing to paste in
+  when you want the assistant to have everything.
+- Appending `.md` to any page's URL returns that page as plain Markdown.
+
+Both files are generated from these pages, so they carry exactly the same
+caveats as the pages themselves.
+
 ## Conventions
 
 - **privacycommand** is one word, all lowercase, including at the start of a