关于
This skill performs comprehensive on-page SEO audits and optimizations for single pages, covering elements like title tags, meta descriptions, headers, and internal links. It triggers when users request page optimizations, SEO audits, or improvements to ranking signals, even without explicit "SEO" terminology. Developers can use it for stack-agnostic SEO analysis across any CMS or framework.
快速安装
Claude Code
推荐npx skills add rampstackco/claude-skills -a claude-code/plugin add https://github.com/rampstackco/claude-skillsgit clone https://github.com/rampstackco/claude-skills.git ~/.claude/skills/seo-onpage在 Claude Code 中复制并粘贴此命令以安装该技能
技能文档
On-Page SEO
Optimize a single page for search relevance, click-through, and crawler comprehension. Stack-agnostic. Works on any CMS, framework, or static site.
When to use
- Auditing or optimizing a single page (homepage, product page, article, landing page)
- Writing or reviewing title tags and meta descriptions
- Fixing header structure or content hierarchy
- Reviewing internal links from or to a page
- Improving a page's CTR from search results
When NOT to use
- Site-wide crawl, indexing, or speed issues (use
seo-technical) - Keyword research or intent mapping (use
seo-keyword) - Competitor SERP analysis (use
seo-competitor) - Auditing many pages at once for prune/merge decisions (use
seo-content-audit)
Required inputs
- The page URL or the draft content if pre-launch
- The primary target query (one phrase the page should rank for)
- The page's role in the site (commercial, informational, navigational)
If the target query is unknown, run seo-keyword first or ask the user to name one.
The framework: 8 dimensions
A complete on-page audit covers eight dimensions. Score each as Pass, Needs work, or Fail. Note the specific fix.
1. Title tag
- One unique title per URL across the site
- Roughly 50 to 60 characters (longer gets truncated in SERPs)
- Primary query near the front
- Brand at the end if it earns inclusion
- Distinct from the H1 (often very similar, but should not be identical word-for-word)
2. Meta description
- One unique description per URL
- Roughly 150 to 160 characters
- Restates the value proposition, not the title
- Includes a soft CTA where natural
- Treats the description as ad copy that earns the click
3. Header structure
- Exactly one H1 per page
- H1 contains or paraphrases the primary query
- H2 sections cover the major sub-topics
- H3+ used only when an H2 has genuine sub-points
- No skipped levels (no H2 followed by H4)
- Headers describe sections accurately enough that a reader could navigate by them alone
4. Body content
- Opens with the primary user intent answered in the first paragraph
- Covers the topic comprehensively (define your competition's depth, then match or exceed it)
- Includes related entities and supporting concepts naturally
- Avoids keyword stuffing (write for the reader, not the bot)
- Reading level matches the audience (run a readability check)
- Paragraphs short enough to scan on mobile (3 to 5 lines)
5. Internal links
- At least 2 to 3 outbound internal links to closely related pages
- At least 2 to 3 inbound internal links from related pages
- Anchor text is descriptive, not "click here" or "learn more"
- Links to canonical URLs, not redirects
- No broken internal links
6. Images and media
- Every meaningful image has descriptive alt text (skip alt for purely decorative images)
- File names are descriptive ("blue-running-shoe.jpg" not "IMG_4032.jpg")
- Modern format used where supported (WebP, AVIF)
- Lazy loading on below-the-fold images
- Width and height attributes set to prevent layout shift
7. URL slug
- Lowercase, hyphen-separated
- Includes the primary query naturally (no stuffing)
- Short (under 60 characters where possible)
- No dates unless the page is genuinely time-bound
- No session IDs, tracking parameters, or random hashes
- Matches the site's URL pattern conventions
8. On-page schema
- Appropriate Schema.org type for the content (Article, Product, FAQPage, HowTo, Recipe, etc.)
- Required properties filled (review Schema.org docs for the type)
- Validates in Google's Rich Results Test or equivalent
- Matches what is visible on the page (do not lie to crawlers)
- Author and publisher schema linked correctly for content pages
Workflow
- Confirm the target query. If unclear, ask. Do not optimize without one.
- Render the page. View it as a user would. Read the content top to bottom.
- View the rendered HTML. Inspect the actual served markup, not just the visual page. Check
<title>,<meta>, headers, and schema in the source. - Run the 8-dimension framework. Score each, note specific fixes.
- Prioritize. Group fixes into Critical (broken or missing), Important (suboptimal), and Nice-to-have (polish).
- Write the report. Use the template in
references/audit-template.md. - Offer to draft fixes. If the user wants, draft the new title, meta, headers, or copy directly.
Failure patterns
When you spot one of these, push back before delivering.
- "Make it more SEO." Vague. Ask for the target query and what's broken first.
- "Add the keyword 5 times in the body." Keyword density is not a real ranking signal. Prioritize relevance and topical depth instead.
- Optimizing a page that is not crawlable or indexable. Check
seo-technicalfirst. No on-page work helps a noindexed page. - Optimizing a page with no clear user intent. A page that does not serve a real query will not rank no matter how well-tagged it is.
- Targeting the same query as another page on the site. This is cannibalization. Use
seo-content-auditto decide which page should rank.
Output format
Default output is a markdown audit at seo-audit-[page-slug].md in the project root. Structure:
- Page summary (URL, target query, role)
- Score across 8 dimensions
- Critical fixes
- Important fixes
- Nice-to-have polish
- Drafted replacements (if requested)
Keep audits under 1500 words. If a page needs more detail, link to deeper appendices.
Reference files
references/audit-template.md- Fillable audit template, copy and use.references/onpage-checklist.md- Quick-reference checklist for the 8 dimensions.references/title-and-meta-patterns.md- Patterns for writing strong titles and meta descriptions.
GitHub 仓库
常见问题
什么是 seo-onpage Skill?
seo-onpage 是一个 Claude Skill,作者为 rampstackco。Skill 将 Claude 按需加载的说明和资源打包,让 Claude 无需额外提示即可执行与 seo-onpage 相关的任务。
如何安装 seo-onpage?
使用本页的安装命令:将 seo-onpage 作为插件添加到 Claude Code,或将其仓库克隆到 skills 目录,然后重启 Claude 以加载该 Skill。
seo-onpage 属于哪个分类?
seo-onpage 属于元分类。
seo-onpage 可以免费使用吗?
可以。seo-onpage 已收录在 AIMCP,可免费安装。
相关推荐技能
Content Collections 是一个 TypeScript 优先的构建工具,可将本地 Markdown/MDX 文件转换为类型安全的数据集合。它专为构建博客、文档站和内容密集型 Vite+React 应用而设计,提供基于 Zod 的自动模式验证。该工具涵盖从 Vite 插件配置、MDX 编译到生产环境部署的完整工作流。
这个Claude Skill为开发者提供完整的Polymarket预测市场开发支持,涵盖API调用、交易执行和市场数据分析。关键特性包括实时WebSocket数据流,可监控实时交易、订单和市场动态。开发者可用它构建预测市场应用、实施交易策略并集成实时市场预测功能。
该Skill帮助开发者创建OpenCode插件,用于接入命令、文件、LSP等25+种事件。它提供了插件结构、事件API规范和JavaScript/TypeScript实现模式,适合需要拦截操作、扩展功能或自定义事件处理的场景。开发者可通过它快速构建响应式模块来增强OpenCode AI助手的能力。
SGLang是一个专为LLM设计的高性能推理框架,特别适用于需要结构化输出的场景。它通过RadixAttention前缀缓存技术,在处理JSON、正则表达式、工具调用等具有重复前缀的复杂工作流时,能实现极速生成。如果你正在构建智能体或多轮对话系统,并追求远超vLLM的推理性能,SGLang是理想选择。
