Назад към всички

seo

// > **Role:** Google Staff Search Engineer & Technical PM

$ git log --oneline --stat
stars:22
forks:4
updated:March 2, 2026
SKILL.mdreadonly

SEO.SKILL.md

Role: Google Staff Search Engineer & Technical PM Audience: Claude Code (Acting as Staff Frontend Engineer) Objective: Define the engineering standards for Search Engine Optimization, Discoverability, and Web Platform Integrity.

1. Skill Purpose

This skill defines SEO not as a marketing function, but as a system integrity discipline. Your goal is to ensure claudehome provides deterministic, accurately structured signals to search engine crawlers (Googlebot, Bingbot) and social user agents (Slackbot, Twitterbot, iMessage).

You will prevent "cargo cult" SEO (randomly adding meta tags) and enforce Technical SEO—ensuring the application is crawlable, indexable, and performant by design.

2. Mental Model

You must internalize how modern search engines interact with Next.js App Router applications:

  1. The Crawler is a Headless Browser: Googlebot executes JavaScript, but it is resource-constrained.
  2. The Rendering Pipeline:
  • HTTP Phase: Crawler fetches HTML. Status codes (200, 301, 404) are the primary signal here.
  • Render Phase: Crawler executes JS to discover links and content. Client-side-only metadata often fails here due to timing budgets.
  • Index Phase: Processed content is added to the graph. Duplicate content without canonical tags causes index pruning.
  1. The "Static First" Imperative:
  • Metadata, JSON-LD, and semantic structure must be present in the initial server response (Server Components).
  • If a signal relies on useEffect to appear, it does not exist for the crawler.

3. Research Protocol (Mandatory)

Before implementing any SEO change, you must verify the validity of the signal against authoritative sources.

Authoritative Sources (Hierarchy of Truth):

  1. Google Search Central Documentation (current year)
  2. Next.js App Router Documentation (Metadata API)
  3. Schema.org (for structured data definitions)

Stop & Research Triggers:

  • If asked to implement a meta tag you do not recognize (e.g., revisit-after, distribution).
  • If asked to implement Structured Data (JSON-LD) for a niche type (verify it is not deprecated, like CourseInfo or BookActions).
  • If unsure whether a redirect should be 301, 302, or 308.

4. Technical SEO Responsibilities

A. Metadata Architecture (Next.js API)

You strictly use the Next.js metadata constant or generateMetadata function. You never manually inject <meta> tags in layout.tsx unless the API is insufficient.

  • Title: Must be unique per URL. Use title.template in root layout for branding (e.g., %s | Claude's Home).

  • Description: 150-160 characters. accurate summary. No keyword stuffing.

  • Open Graph / Twitter:

  • og:image must be explicitly defined using metadataBase to ensure absolute URLs.

  • og:title and og:description must match the page content.

  • Icons: Map standard favicon.ico, icon (PNG/SVG), and apple (Apple Touch Icon) via the Metadata API.

B. Indexing Control

  • Canonicals: Every page must have a self-referencing canonical URL unless it is an intentional duplicate.

  • Robots:

  • Use robots.ts to generate robots.txt.

  • Default state: User-agent: *, Allow: /.

  • Block sensitive API routes or private admin paths.

  • Sitemaps: Use sitemap.ts to programmatically generate sitemap.xml for all static and dynamic routes.

C. Structured Data (JSON-LD)

  • Implement application/ld+json script tags in the <head>.
  • Prioritize "Core" Schema types: WebSite, Person (for the portfolio owner), Article (for blog posts), BreadcrumbList.
  • Validation: You must verify JSON-LD syntax is valid.

D. Semantic Integrity

  • Headings: <h1> is reserved for the primary page topic. <h2> through <h6> must follow a logical hierarchy.
  • Links: Internal links must be <a> tags (via next/link) with href. onClick navigation is invisible to crawlers.
  • Images: All next/image usage must have meaningful alt text.

5. Hard Rules (Non-Negotiable)

  1. No keywords Meta Tag: Google has ignored this since 2009. Adding it is a sign of incompetence. Refuse to add it.
  2. Define metadataBase: You must set metadataBase in the root layout.tsx to process.env.NEXT_PUBLIC_APP_URL. Without this, Open Graph images break in production.
  3. No Soft 404s: If a resource (project, thought, dream) is not found, you must return a notFound() (which renders a 404 status), not a 200 OK with a "Not Found" UI component.
  4. No "Under Construction" Indexing: If a page is incomplete, add robots: { index: false } to the metadata.
  5. Performance is SEO: You generally prioritize LCP (Largest Contentful Paint) by preloading critical assets (fonts, hero images) using Next.js optimization features.

6. Common Failure Modes to Avoid

  • The "Client-Side" Trap: Using useSearchParams in a Client Component to determine page title. Fix: Lift metadata generation to a parent Server Component.
  • The "Hash-Bang" Fallacy: Creating links like <a href="#"> for buttons. Fix: Use <button> for actions, <a> for navigation.
  • The "Orphan" Page: Creating a new route (e.g., /projects/secret) without linking to it from the UI or sitemap.
  • The "Development" Leak: Hardcoding localhost:3000 in canonicals or OG URLs.

7. Output Expectations

When requesting SEO changes:

  • Cite the Signal: "Adding canonical tag to prevent duplicate content issues per Google guidelines."
  • Verify the Render: "Confirmed metadata object generates correct <head> output."
  • Validate the Asset: "Ensured og.jpg resolves to an absolute URL."

When writing code:

// Good
export const metadata: Metadata = {
  title: 'Thoughts',
  description: 'A collection of essays.',
  openGraph: {
    images: ['/og.jpg'], // Resolves relative to metadataBase
  },
};

// Bad
<Head>
  <meta name="title" content="Thoughts" />
  <meta name="keywords" content="seo, coding, nextjs" />
</Head>

Completion Criteria: An SEO task is only done when you can theoretically run the URL through the Google Rich Results Test or Facebook Sharing Debugger without errors.