検索エンジン最適化監査エージェントスキル

seo-audit — 基本SEO監査

シンプルなSEOエージェントスキルで、迅速かつデフォルトの単一ページSEO監査向けに設計されています。OpenClawによって駆動。初回のページチェックや、完全な技術的詳細を必要としない迅速な評価に適しています。

このスキルを使用するタイミング

以下の場合にseo-auditを使用します。

• ユーザーが言った場合: "audit this page", "check SEO", "analyze my URL", "quick SEO check", "what's wrong with my page"
• 特定の深さが要求されていない — これがデフォルトのエントリーポイントです
• ユーザーが包括的な技術的ブレイクダウンではなく、速く読めるサマリーを必要としている場合

より深い分析が必要な場合は、seo-audit-fullにアップグレードしてください:

ヒント: 深いテクニカル監査、高度なオンページSEO、または完全なレポートには、seo-audit-fullスキルを使用してください。

期待される入力

URLのみが提供され、ソースコードやクローラーデータが利用できない場合は、明確に述べてください:

制限: This audit is based on visible page content and publicly available signals only. Source code, GSC data, crawl logs, and performance metrics are not available for this audit.

出力

テンプレートassets/report-template.htmlに記入して基本SEO監査レポートを作成し、 ファイルに保存してください — 端末に生のHTMLを表示しないでください。

ファイル命名: reports/<hostname>-<slug>-audit.html

https://example.com/blog/best-tools → reports/example-com-blog-best-tools-audit.html
https://example.com/                → reports/example-com-audit.html

保存後、ユーザーに伝える:

✅ Report saved → reports/example-com-audit.html
   Open it now? (yes / no)

はいの場合 → 実行: open reports/example-com-audit.html

テンプレートのプレースホルダー — それぞれ独立して記入してください:

スクリプト

所見を書く前にこれらのスクリプトを実行してください。構造化されたJSONを出力します — JSONを証拠として直接使用し、同じURLを手動で再取得しないでください。

依存関係: pip install requests (html parsing uses Python stdlib)

# Step 1: site-level checks (robots.txt + sitemap.xml)
python scripts/check-site.py https://example.com

# Step 2: page-level checks (H1, title, meta description, canonical)
python scripts/check-page.py https://example.com
# With primary keyword (recommended — enables H1 keyword presence check)
python scripts/check-page.py https://example.com --keyword "running shoes"

# Optional: fetch raw page HTML for further inspection
python scripts/fetch-page.py https://example.com --output page.html

# Step 3: JSON-LD schema validation
python scripts/check-schema.py https://example.com
# Or from previously fetched HTML (avoids redundant fetch):
python scripts/check-schema.py --file page.html

各スクリプトはコード0（すべて合格/警告）または1（失敗/エラーあり）で終了します。

厳密な範囲 — 以下にリストされていないチェックを追加しないでください。例外はありません。

許可されたサイトレベルチェック ({{site_checks_html}}内):

• robots.txt · sitemap.xml · 404 Handling · URL Canonicalization · i18n / hreflang

許可されたE-E-A-Tチェック ({{eeat_checks_html}}内):

• About Us · Contact · Privacy Policy · Terms of Service · Media/Partners (only if present)

許可されたページレベルチェック ({{page_checks_html}}内)、この正確な順序で出力: URL Slug · Title Tag · Meta Description · H1 Tag · Canonical Tag · Image Alt Text · Word Count · Keyword Placement · Heading Structure · Internal Links · Schema (JSON-LD)

Image Alt Text logic:

• Parse <img> tags from static HTML
• Pass: all images have non-empty alt (decorative images with alt="" are OK)
• Warn: any content image missing alt attribute
• Unverified (status-info): 0 images found in static HTML → likely JS-rendered, cannot verify

⛔ ハードルール — report-template.htmlで定義されたチェック行のみを出力してください。 上記の許可リストにないチェックは、問題を見つけても出力しないでください。 例外はありません。「ボーナス」チェックや即興もありません。 テンプレートが唯一の真実のソースです。厳密なホワイトリストとして扱ってください。

依然として禁止 (seo-audit-fullに属するもの): OG tags · Twitter Card · Social tags · Page Weight · Core Web Vitals · Robots Meta

JSON出力の使用方法:

• Map each field's status → pass / warn / fail / error directly to the report check table
• Use each field's detail string as the starting point for the Evidence line in findings
• Do not contradict the script output unless you have additional observable evidence
• Separate check groups with <div class="subsection-label">Label</div> inside {{site_checks_html}}: Crawlability · URL Canonicalization · i18n / hreflang · Schema (JSON-LD) and <div class="subsection-label">E-E-A-T Trust Pages</div> before {{eeat_checks_html}}

LLMレビュー — llm_review_required: trueの場合必須:

スクリプトは、意味的または品質的な判断が必要で、それ自身では実行できないフィールドにフラグを立てます。 llm_review_required: trueを未解決のままにしないでください — 常に明示的な判断を下してください。

H1 — keyword_match == "partial"の場合にトリガー:

h1_text : (from h1.values[0])
keyword : (the --keyword passed to the script)

Judge: Does this H1 semantically cover the keyword's search intent?
  - Consider synonyms, natural variants, topic coverage
  - yes → downgrade to "pass", note the variant
  - no  → keep "warn" or upgrade to "fail", explain the gap

タイトル — keyword_match == "partial"またはkeyword_position != "start"の場合にトリガー:

title   : (from title.value)
keyword : (the --keyword passed)

Judge:
  1. Does the title semantically cover the keyword's search intent?
  2. Is the title grammatically correct and naturally readable?
  3. Keyword position — apply different standards by page type:
     - Homepage   : Brand + core keyword is correct (e.g. "Acme | AI Workflow Automation")
                    Do NOT flag brand-first as a problem.
     - Inner pages: Core keyword should lead (e.g. "AI Workflow Automation for Teams — Acme")
                    Flag if keyword is buried mid-title without good reason.

IMPORTANT — do NOT flag these as negatives:
  - Years (e.g. "2026") → signal freshness, increase CTR — treat as positive unless
    the page is explicitly evergreen content where dating would hurt longevity.
  - Numbers (e.g. "5 best", "Top 10", "3 steps") → set clear expectations,
    consistently outperform non-numeric titles in CTR — always treat as a plus.
  - Specific qualifiers ("Open-Source", "Self-Hosted", "Free") → narrow intent
    and attract higher-quality clicks — do not penalize.

URLスラッグ — keyword_match != "full"またはis_homepage == falseの場合にトリガー:

slug    : (from url_slug.slug)
keyword : (the --keyword passed)

Judge:
  1. Does the slug contain the primary keyword or a natural variant?
  2. Is the path hierarchy logical? (/category/keyword is ideal)
  3. Is it concise and human-readable?
  Homepage (is_homepage: true): skip — no judgment needed.

メタディスクリプション — コンテンツが存在する場合常にトリガー:

meta_description : (from meta_description.value)
keyword          : (the --keyword passed)

Judge all four:
  1. Complete sentence(s)? (1-2 sentences, no fragments)
  2. Mentions a concrete result — not vague fluff?
     Good: "Cut design time by 60% with AI-powered templates"
     Bad:  "The best tool for all your design needs"
  3. Keyword or natural synonym used once — not stuffed?
  4. More specific than what a typical competitor would write?

IMPORTANT — do NOT flag these as negatives:
  - Years (e.g. "2026") → signal freshness, improve CTR for time-sensitive queries.
    Only note the year if the page is explicitly evergreen content where dating hurts.
  - Numbers (e.g. "5 best", "3 steps") → concrete specificity, strong CTR signal.
  - Trailing "and more." → minor style note at most, never a Warning or Fail.

推奨ワークフロー

以下の手順を順番に実行してください:

1. 範囲の確認 — confirm this is a basic audit; note any missing data

2. 主要キーワードの推測 — fetch the page with fetch-page.py, then determine the primary keyword:

   • If the user explicitly provided a keyword → use it directly
   • If not → read the page H1, title, and first paragraph, then infer the single most likely target keyword phrase (what would a searcher type to find this page?)
   • State the inferred keyword explicitly before running checks:

     "Inferred primary keyword: open source claude alternatives"

3. check-site.pyの実行 — parse the JSON output for robots, sitemap, 404 handling, and URL canonicalization

   404 check: fetch <origin>/this-page-definitely-does-not-exist-seo-audit-check

   • Returns 404 → Pass · Returns 200 (soft 404) → Fail · Returns 301 to homepage → Warn

   URL Canonicalization checks (each is a separate sub-check):

   • HTTP→HTTPS: fetch http://<host> — must 301 to https://. Returns 200 → Fail.
   • www consistency: fetch both https://www.<host> and https://<host> — one must 301 to the other. Both return 200 → Warn.
   • Trailing slash: compare the URL actually served vs the canonical tag on the page. Mismatch → Warn.
   • Canonical match: canonical tag href must exactly match the final URL after all redirects. Mismatch → Warn.

4. E-E-A-Tインフラストラクチャチェック — for each trust page below, check two layers:

   • Layer 1 — Exists: fetch the URL, check HTTP status (200 = exists, 404/redirect = missing)
   • Layer 2 — Reachable: fetch homepage HTML, check if footer or nav contains a link to this page

   Page	Required
   About Us	Yes
   Contact	Yes
   Privacy Policy	Yes
   Terms of Service	Yes
   Media / Partners	No — include only if present

   Status rules:

   • Page missing (non-200) → Fail
   • Page exists but not linked in footer/nav → Warn
   • Page exists and linked in footer/nav → Pass
   • Optional page missing → skip, do not include row

5. check-page.py --keyword "<推測したキーワード>"の実行 — parse the JSON output for H1, title, meta description, canonical, and URL slug

6. i18n / hreflangチェック — only run if the page contains hreflang tags or <html lang> suggests multi-language:

   • Skip entirely (N/A) if no hreflang tags found and site appears single-language
   • If hreflang tags present, check:
     • Reciprocal symmetry: every URL referenced must link back to all other variants — any broken link = Fail
     • Language codes: must be valid BCP 47 (e.g. zh-CN not zh, en-US not en-us) — wrong code = Warn
     • x-default: should be present for language-selector or fallback pages — missing = Warn
     • html[lang] attribute: must match the primary hreflang of the page — mismatch = Warn
     • URL structure: recommended pattern — default language (usually en) at root with no prefix, other languages under subpaths (/zh/, /es/).
       • /page (en) + /zh/page + /es/page → Pass
       • /en/page + /zh/page → Warn (en prefix is redundant, wastes crawl depth)
       • Only flag if the pattern is clearly inconsistent or en is unnecessarily prefixed

7. check-schema.pyの実行 — parse the JSON output for schema types and field validation

   python scripts/check-schema.py https://example.com
   # Or from previously fetched HTML:
   python scripts/check-schema.py --file page.html
   

   The script extracts JSON-LD blocks, validates @type and required fields per Schema.org spec. llm_review_required: true is always set — confirm inferred_page_type matches actual page content.

   Page type → expected @type reference:

   Page Type	Expected @type	Min. required fields
   Homepage	WebSite + Organization	name, url, logo
   Blog / Article	Article or BlogPosting	headline, datePublished, author, image
   Product	Product	name, image, offers (price, priceCurrency)
   FAQ	FAQPage	mainEntity[].name, acceptedAnswer.text
   How-to	HowTo	name, step[].text
   Local business	LocalBusiness	name, address, telephone
   Generic landing	—	N/A — skip, no widely-supported type

   • Pass: correct @type present, all required fields valid, no conflicts
   • Warn: @type present but missing recommended fields
   • Fail: expected @type missing entirely
   • N/A: generic landing page — do not penalize

8. 所見の要約 — each finding must follow the Evidence / Impact / Fix format

9. 優先アクション — list the top 3 highest-impact fixes

10. レポートのレンダリング — save to reports/<hostname>-<slug>-audit.html, then ask user to open

11. アップグレードプロンプト — if issues beyond basic scope are found, suggest seo-audit-full

レポート詳細の記述ルール

The Detail cell in check tables must follow these rules — no exceptions:

Pass → one short phrase. No lists, no elaboration.

Good: "Valid XML urlset · 104 URLs · referenced in robots.txt."
Bad:  "Valid XML urlset with 104 URLs. Correctly referenced in robots.txt.
       Blog posts are likely indexed through this sitemap."

Warn → one <div class="detail-issue"> with ≤2 bullet points. One <div class="detail-fix"> with the fix.

Good:
  <div class="detail-issue">· Title 48 chars — 2 below minimum. · Year "2026" will date the page.</div>
  <div class="detail-fix">Expand to 50–60 chars; remove year if evergreen.</div>

Bad: three-sentence prose explaining what a title tag is and why length matters.

Fail → same as Warn. Lead with the exact failure. No background explanations.

Do NOT explain what a check is, do NOT repeat information already visible in the status badge, do NOT treat the reader as unfamiliar with SEO basics.

必須の所見フォーマット

すべての重要な所見は必ずこの構造に従ってください:

**Finding: [Finding Title]**

- **Evidence:** [What was observed — direct quote, screenshot ref, or measurable data]
- **Impact:** [Why this matters for SEO or UX]
- **Fix:** [Specific, actionable recommendation]

Do not write vague conclusions. If evidence is insufficient, state assumptions explicitly.

アップグレードプロンプト

すべての基本監査レポートの最後に以下を含めてください:

Want a deeper analysis? This was a basic SEO audit covering site-level signals and core on-page checks. For advanced technical SEO, content quality scoring, structured data analysis, and full crawl-based findings, use the seo-audit-full skill.

参照ファイル

• Detailed audit scope and field definitions: references/REFERENCE.md
• Final HTML report template: assets/report-template.html
• Site-level check script: scripts/check-site.py
• Page-level check script: scripts/check-page.py
• Raw page fetcher: scripts/fetch-page.py
• Schema validation script: scripts/check-schema.py