docs-from-sitelisted

# docs-from-site — Build docs from a product website The actual work is done by the **`docs-site-crawler`** subagent (Haiku, pinned model). ## Workflow 1. Fetch the sitemap first (`/sitemap.xml`). Fall back to crawling `<a href>` links from the homepage only if the sitemap is missing. 2. Prioritize documentation-relevant paths (`/docs`, `/help`, `/guides`, `/features`) and cap the crawl at ~50 pages. 3. For each page: strip navigation chrome (`<header>`, `<footer>`, `<nav>`, `<aside>`) before converting HTML to Markdown. 4. Extract branding tokens from inline `<style>` or `:root` CSS variables (accent color, background, foreground). Detect the color scheme from background luminance. Look for a theme toggle element to decide between pinning the scheme or setting `system`. 5. Organize output into a structured folder: intro README, getting-started section, per-feature pages, guides, API reference, and FAQ — only if that content was found. 6. Write a `_branding.json` alongside the docs folder. ## Guardrails - Use WebFetch first; escalate to a browser only if `<main>` content is empty. - Hard-exclude auth and commerce paths (`/login`, `/signup`, `/auth`, `/checkout`, `/cart`). - Cap at 50 pages — sites with hundreds of URLs contain mostly blog noise. - Strip navigation chrome before the HTML-to-Markdown pass, not after. - Write in active voice, second person, sentence-case headings. No filler words ("simply", "just", "easily"). Tag every code block with a language. Use relati