Problem
The llms-markdown integration generates .md files from built HTML pages for LLM consumption. However, pages with interactive Astro islands (demos, tabs, copy button) include <script> and <style> tags in the built HTML that leak through into the markdown output.
This results in chunks of minified JavaScript (Astro island hydration code, Sentry init, etc.) appearing inline in the markdown — making it noisy and unhelpful for LLMs.
Example: https://vjs10-site.netlify.app/docs/framework/react/reference/play-button.md
Root Cause
The integration clones [data-llms-content] elements and strips [data-llms-ignore] descendants, but doesn't remove <script> or <style> tags before passing the HTML to Turndown. Astro islands render as <astro-island> custom elements with inline <script> tags for hydration — Turndown treats the script text content as regular text and includes it in the markdown.
Fix
Strip <script> and <style> tags from the cloned content before Turndown conversion. This removes the hydration scripts while preserving the server-rendered HTML children (including code samples inside tab panels).
PR: #678
Problem
The
llms-markdownintegration generates.mdfiles from built HTML pages for LLM consumption. However, pages with interactive Astro islands (demos, tabs, copy button) include<script>and<style>tags in the built HTML that leak through into the markdown output.This results in chunks of minified JavaScript (Astro island hydration code, Sentry init, etc.) appearing inline in the markdown — making it noisy and unhelpful for LLMs.
Example: https://vjs10-site.netlify.app/docs/framework/react/reference/play-button.md
Root Cause
The integration clones
[data-llms-content]elements and strips[data-llms-ignore]descendants, but doesn't remove<script>or<style>tags before passing the HTML to Turndown. Astro islands render as<astro-island>custom elements with inline<script>tags for hydration — Turndown treats the script text content as regular text and includes it in the markdown.Fix
Strip
<script>and<style>tags from the cloned content before Turndown conversion. This removes the hydration scripts while preserving the server-rendered HTML children (including code samples inside tab panels).PR: #678