ai-design-vault: Skill for Claude Code

SEO audit

Audit the code, and the live URL when one exists. Rank findings by impact and fix them; do not hand back a lecture. Resolve routes, metadata API and the site URL variable from .claude/stack.json.

0. Establish the render model first — it changes everything

framework.renderModel in stack.json:

• spa — the single largest finding on any marketing site. Crawlers get an empty root element; answer engines execute even less JS than Googlebot. Nothing below matters until content is server-rendered or prerendered. Say so first, and propose the framework's prerender/SSG path rather than working around it.
• ssg / ssr / server-components / islands — content is in the HTML. Verify it, do not assume: curl -s <url> | grep -c "<known heading text>", or view-source on the deployed page.
• islands — check that content inside a client island is present in the server-rendered HTML, not mounted after hydration.

1. Indexability — nothing else matters if this is broken

• robots — crawlers allowed, sitemap referenced, no accidental disallow: /, no staging block surviving into production.
• sitemap — every public route is listed. This is the one that rots. Cross-check the actual route files under paths.routes against the sitemap's entries; a route added without a sitemap entry is the most common miss.
• Canonical URL per page; no duplicate content across trailing-slash or parameter variants.
• The site URL variable is set in production (NEXT_PUBLIC_SITE_URL, PUBLIC_SITE_URL, NUXT_PUBLIC_SITE_URL — whatever this stack uses). Unset means canonical and OG tags resolve to localhost for social scrapers.
• No route accidentally opted out of static rendering. Anything that reads headers or cookies at request time — including bot detection — forces dynamic rendering and costs TTFB on every visit.

2. Metadata

Every route produces metadata through the stack's own mechanism (bindings.metadata) — never hand-written <meta> tags scattered in components.