From a10d1d8162bca209b50a7a71e538f7671e109c3a Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Andrea=20S=C3=A1nchez=20Blanco?=
 <asanchezblanco@stackoverflow.com>
Date: Thu, 21 May 2026 10:49:51 +0200
Subject: [PATCH 1/2] feat(docs): split llms.txt into spec-compliant index and
 full dump

---
 packages/stacks-docs/.eleventy.js |  4 +++
 packages/stacks-docs/llms.11ty.js | 48 +++++++++++++++++++++++++++++++
 2 files changed, 52 insertions(+)
 create mode 100644 packages/stacks-docs/llms.11ty.js

diff --git a/packages/stacks-docs/.eleventy.js b/packages/stacks-docs/.eleventy.js
index d9b9f6981b..acdc51237b 100644
--- a/packages/stacks-docs/.eleventy.js
+++ b/packages/stacks-docs/.eleventy.js
@@ -18,8 +18,12 @@ module.exports = function(eleventyConfig) {
   eleventyConfig.addPlugin(iconPlugin);
   eleventyConfig.addPlugin(headerPlugin);
   eleventyConfig.addPlugin(tipPlugin);
+  // Produces /llms-full.txt — a content dump for tools that want the full
+  // markdown of every page in one file. The spec-compliant index lives at
+  // /llms.txt and is rendered by llms.11ty.js.
   eleventyConfig.addPlugin(llmsTxtPlugin, {
     siteUrl: 'https://v2.stackoverflow.design',
+    outputPath: 'llms-full.txt',
     collections: ['base', 'components', 'develop', 'foundation'],
     additionalMetadata: ['description'],
     normalizeWhitespace: true,
diff --git a/packages/stacks-docs/llms.11ty.js b/packages/stacks-docs/llms.11ty.js
new file mode 100644
index 0000000000..006a5338c4
--- /dev/null
+++ b/packages/stacks-docs/llms.11ty.js
@@ -0,0 +1,48 @@
+// Renders the spec-compliant llms.txt index at /llms.txt
+// (see https://llmstxt.org). The full content dump lives at /llms-full.txt,
+// generated by eleventy-plugin-llms-txt — see .eleventy.js.
+
+const SECTIONS = [
+  { tag: "base", heading: "Base utilities" },
+  { tag: "components", heading: "Components" },
+  { tag: "develop", heading: "Develop" },
+  { tag: "foundation", heading: "Foundation" },
+];
+
+function pageLink(page, siteUrl) {
+  const title = page.data.title;
+  const description = (page.data.description || "")
+    .replace(/<[^>]+>/g, "") // descriptions may contain inline HTML for the rendered page
+    .replace(/\s+/g, " ")
+    .trim();
+  const suffix = description ? `: ${description}` : "";
+  return `- [${title}](${siteUrl}${page.url})${suffix}`;
+}
+
+module.exports = class {
+  data() {
+    return {
+      permalink: "/llms.txt",
+      eleventyExcludeFromCollections: true,
+    };
+  }
+
+  render({ collections, site }) {
+    const sections = SECTIONS.map(({ tag, heading }) => {
+      const pages = (collections[tag] || [])
+        .slice()
+        .sort((a, b) =>
+          (a.data.title || "").localeCompare(b.data.title || "")
+        );
+      const lines = pages.map((p) => pageLink(p, site.url));
+      return `## ${heading}\n\n${lines.join("\n")}`;
+    });
+
+    return `# ${site.title}
+
+> ${site.description}
+
+${sections.join("\n\n")}
+`;
+  }
+};

From 32d39b1aff33ae617871dc84ced391d4b98cb3d8 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Andrea=20S=C3=A1nchez=20Blanco?=
 <asanchezblanco@stackoverflow.com>
Date: Thu, 21 May 2026 11:18:10 +0200
Subject: [PATCH 2/2] fix(docs): make llms.txt HTML stripper resilient to
 unclosed tags

CodeQL js/incomplete-multi-character-sanitization flagged the
single-pass regex strip in llms.11ty.js because an unclosed `<script`
(missing `>`) would leave the pattern in the output. Loop the
replace until the string is stable, per CodeQL's recommended fix.

Output is unchanged for all current inputs (frontmatter descriptions
are author-written and well-formed); this is defense in depth.
---
 packages/stacks-docs/llms.11ty.js | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/packages/stacks-docs/llms.11ty.js b/packages/stacks-docs/llms.11ty.js
index 006a5338c4..4730200001 100644
--- a/packages/stacks-docs/llms.11ty.js
+++ b/packages/stacks-docs/llms.11ty.js
@@ -9,10 +9,22 @@ const SECTIONS = [
   { tag: "foundation", heading: "Foundation" },
 ];
 
+// Descriptions may contain inline HTML for the rendered page; strip it for
+// llms.txt. Loop until stable so unclosed/nested-looking tags can't reintroduce
+// the pattern (CodeQL js/incomplete-multi-character-sanitization).
+function stripHtml(input) {
+  let previous;
+  let output = input;
+  do {
+    previous = output;
+    output = output.replace(/<[^>]+>/g, "");
+  } while (output !== previous);
+  return output;
+}
+
 function pageLink(page, siteUrl) {
   const title = page.data.title;
-  const description = (page.data.description || "")
-    .replace(/<[^>]+>/g, "") // descriptions may contain inline HTML for the rendered page
+  const description = stripHtml(page.data.description || "")
     .replace(/\s+/g, " ")
     .trim();
   const suffix = description ? `: ${description}` : "";