メニュー
SEO validator: meta/OG, Schema.org, hreflang, Core Web Vitals, crawlability. Triggers: SEO, meta tags, Schema.org, hreflang, LCP, INP, CLS, Core Web Vitals, sitemap, crawlability.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/softspark/ai-toolkit --skill seo-validateコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
| name | seo-validate |
| description | SEO validator: meta/OG, Schema.org, hreflang, Core Web Vitals, crawlability. Triggers: SEO, meta tags, Schema.org, hreflang, LCP, INP, CLS, Core Web Vitals, sitemap, crawlability. |
| user-invocable | true |
| effort | medium |
| disable-model-invocation | true |
| context | fork |
| agent | seo-specialist |
| argument-hint | [path] [--scope full|technical|content|performance|geo|rendering|topical] [--severity high|warn|info] [--framework auto|next|nuxt|astro|gatsby|sveltekit|remix|angular|vue|react-spa|vite-spa|cra|static] [--rendering auto|csr|ssr|ssg|isr|hybrid] [--output markdown|json] |
| allowed-tools | Read, Grep, Glob, Bash |
$ARGUMENTS
Scan a codebase for SEO issues using pattern-matching heuristics. Detects W3C/HTML violations, meta tag gaps, structured data problems, hreflang errors, Core Web Vitals risks (LCP/INP/CLS), resource-hint misuse, above-the-fold anti-patterns, GEO gaps (chunk architecture, hedging language, decision frameworks, semantic triples, freshness), topical authority gaps (pillar/cluster structure, orphan pages, cannibalization), SPA/CSR/SSG crawlability problems, technical SEO misconfigurations, and accessibility-for-SEO issues. Read-only — never modifies files.
Standards basis: W3C HTML5 Recommendation, W3C WCAG 2.2, Schema.org vocabulary, IETF RFC 5646 (BCP 47 language tags) for hreflang, web.dev Core Web Vitals thresholds (LCP <2.5s, INP <200ms, CLS <0.1), Google Search Central crawlability guidelines, and emerging GEO (Generative Engine Optimization) practices.
/seo-validate # Scan full project, auto-detect framework
/seo-validate src/ # Scan specific path
/seo-validate --scope rendering # Only SPA/CSR/SSG crawlability checks
/seo-validate --scope performance # Only Core Web Vitals static signals
/seo-validate --scope geo # Only GEO (Generative Engine Optimization)
/seo-validate --scope topical # Only topical authority and cluster architecture
/seo-validate --severity high # Filter to HIGH findings only
/seo-validate --framework next # Force framework (skip auto-detection)
/seo-validate --rendering csr # Force rendering-mode interpretation
/seo-validate --output json # Structured JSON output for CI integration
Scopes:
full (default) — all 10 categoriestechnical — HTML semantics, hreflang, CWV, rendering, technical SEO (categories 1, 4, 5, 7, 8)content — meta/OG, structured data, GEO, a11y-for-SEO (categories 2, 3, 6, 9)performance — only CWV static signals (category 5)geo — only GEO / citability checks (category 6)rendering — only category 7 (SPA/CSR/SSG crawlability) — useful for migration auditstopical — only topical authority and cluster architecture (category 10)Severity filtering: --severity high shows only HIGH, --severity warn shows HIGH+WARN, --severity info shows all. Default: all.
package.jsonGrep/Glob/Read against framework-aware patterns for everything the script does not cover.Run detection before scanning so category patterns can adapt. Detection order:
package.json (if present) and inspect dependencies + devDependencies:| Deps contain | Framework | Default rendering |
|---|---|---|
next | next | hybrid (per-route) |
nuxt | nuxt | ssr |
astro | astro | ssg |
gatsby | gatsby | ssg |
@sveltejs/kit | sveltekit | hybrid |
@remix-run/* | remix | ssr |
@angular/core + @angular/ssr or @nguniversal/* | angular | ssr |
@angular/core alone | angular | csr (flag as SPA) |
vue + nuxt | see nuxt row | — |
vue without nuxt | vue | csr (flag as SPA) |
react + vite without Next/Remix | vite-spa | csr (flag as SPA) |
react-scripts | cra | csr (flag as SPA) |
no package.json OR no framework deps | static | static |
Read config files to refine:
next.config.* — check output: 'export' (forces SSG), images, i18n settings.nuxt.config.* — check ssr: false, generate blocks (SSG export).astro.config.* — check output: 'server'|'static'|'hybrid' and prerender directives.gatsby-config.* — plugin list (gatsby-plugin-react-helmet, gatsby-plugin-sitemap).svelte.config.* — adapter choice (static, node, vercel).vite.config.* + package.json scripts — look for vite-plugin-ssr, vite-plugin-prerender.angular.json — look for SSR builder config.Read entry HTML (public/index.html, index.html, app/layout.tsx, src/app.html, etc.) to confirm whether meaningful content is prerendered or only a mount point (<div id="root"></div>).
Override precedence: --framework and --rendering flags override detection.
Report the detected framework and rendering mode in the Summary table.
python3 ${CLAUDE_SKILL_DIR}/scripts/seo-scanner.py [path] [--output json|text]
Deterministic checks over meta tags, heading order, image alt, JSON-LD, hreflang,
robots.txt, sitemap, Core Web Vitals signals and llms.txt. No framework
awareness — it reads files, not build config. Categories 7 and 10 are not touched.
Read reference/scanner-categories.md in full,
then work every category in --scope with Grep (regex across files) and Read
(config parsing / ordered checks), skipping only what Step 2 already reported.
Patterns are framework-aware — use the framework detected in Step 1 to select the
right rule set.
Categories 7 (Rendering Mode & Crawlability) and 10 (Topical Authority) get no help from the script. Run them in full or say in the report that you did not.
For each finding:
next/image with priority prop" vs. "add <link rel="preload" as="image"> to <head>").definitive for regex matches against known-bad patterns, heuristic for co-occurrence / absence checks.Present findings sorted by severity (HIGH → WARN → INFO), then by file path. State which findings came from the script and which from the manual pass, and name any category you did not complete.
scripts/seo-scanner.py gives a deterministic baseline over 8 of the 10 categories.
The full pattern tables, per-framework rules and standards citations live in
reference/scanner-categories.md.
| # | Category | Covered by the script | Manual pass |
|---|---|---|---|
| 1 | HTML Semantics & W3C | heading order | rest of the category |
| 2 | Meta & Open Graph | title, description, OG/Twitter tags | rest of the category |
| 3 | Structured Data / Schema.org | JSON-LD presence and shape | rest of the category |
| 4 | Hreflang & i18n | hreflang pairs, x-default | rest of the category |
| 5 | Core Web Vitals (static signals) | LCP/CLS/INP signal patterns | rest of the category |
| 6 | GEO (Generative Engine Optimization) | llms.txt presence | rest of the category |
| 7 | Rendering Mode & SPA/CSR/SSG Crawlability | nothing | the whole category |
| 8 | Technical SEO | robots.txt, sitemap | rest of the category |
| 9 | Accessibility for SEO | image alt | rest of the category |
| 10 | Topical Authority & Cluster Architecture | nothing | the whole category |
Categories 7 and 10 have no script coverage at all — they are entirely manual, and skipping them is the most likely way this skill under-reports.
## SEO Validation Report
### Summary
| Metric | Value |
|--------|-------|
| Scope | full / technical / content / performance / geo / rendering / topical |
| Framework detected | next / nuxt / astro / gatsby / sveltekit / remix / angular / vue / react-spa / vite-spa / cra / static |
| Rendering mode | csr / ssr / ssg / isr / hybrid |
| Files scanned | N |
| Public routes found | N |
| Routes with prerendering | N of N |
| Findings: HIGH | N |
| Findings: WARN | N |
| Findings: INFO | N |
### Findings
#### [HIGH] app/layout.tsx:12
Category: HTML Semantics & W3C
Confidence: definitive
Pattern: `<html>` element missing `lang` attribute
W3C Rule: HTML5 §3.2.6
Fix: Add `lang="en"` (or appropriate BCP 47 code) to the `<html>` element.
See: reference/w3c-guidelines.md#lang-attribute
#### [HIGH] components/HomeHero.tsx:24
Category: Core Web Vitals (LCP)
Confidence: definitive
Pattern: Above-the-fold `<img>` with `loading="lazy"`
Rule: LCP anti-pattern — lazy loading the LCP element delays it
Fix: Remove `loading="lazy"`, add `fetchpriority="high"`. For Next.js use `<Image priority />`.
See: reference/core-web-vitals.md#above-the-fold
#### [HIGH] src/App.tsx:1
Category: Rendering Mode & SPA Crawlability
Confidence: definitive
Pattern: CSR-only React app (Vite) with no prerender plugin
Rule: Content-site SPAs without SSR/SSG are invisible to most crawlers
Fix: Add `vite-plugin-ssr` or migrate to Next.js/Remix; OR add `react-snap` for build-time prerender.
See: reference/spa-ssg-patterns.md#react-spa-migration
Confidence values:
definitive — regex match against a known-bad pattern with high precision.heuristic — co-occurrence / absence / ordering / above-the-fold inference — may be false positive.Exit codes (when --output json):
0 — no HIGH findings.1 — one or more HIGH findings.package-lock.json, yarn.lock, pnpm-lock.yaml), vendored directories (node_modules/, vendor/, .git/, dist/, build/, out/, .next/, .nuxt/, .svelte-kit/, public/build/).<noscript> catches only the "no-JS" case, not the "crawler without JS execution" case — don't upgrade a CSR HIGH to WARN just because noscript exists.scripts/seo-scanner.py reads files, not build output. A Next.js metadata export, a useHead() call, or a <svelte:head> block produces correct tags at runtime that the scanner cannot see, so "missing meta description" on a framework project is a claim about the source, not about the page.robots.txt and sitemap.xml are checked at the project root. Frameworks that generate them at build time (next-sitemap, @astrojs/sitemap, gatsby-plugin-sitemap) leave nothing on disk, so absence is not evidence — check the config and the plugin list before reporting it.hreflang="de" is only valid if the German page declares the reverse; a one-file scan sees half the pair and cannot conclude.width/height, unbounded images, render-blocking patterns. Real LCP/CLS/INP come from field data (CrUX, RUM). Reporting a passing CWV category is out of this skill's reach./a11y-validate; category 9 here is deliberately shallow.speakable schema, citation/source markup, AI-extractable content structure, chunk anatomy, 13-week freshness strategy.FAQPage/HowTo/QAPage schema, llms.txt, AI bot robots.txt directives, E-E-A-T signals, automated grep patterns for Category 6.