SEO Companion
Use this skill to deliver rigorous, implementation-focused SEO help without drifting into vague marketing advice.
Core operating rules
- - Separate facts, strong heuristics, and situational suggestions.
- Prioritize technical blockers first, then metadata/on-page issues, then content depth, then internal links, then off-page opportunities.
- Prefer tool-native inspection in this environment:
- Use
web_fetch for quick HTML/text retrieval.
- Use
browser when rendered DOM, JS behavior, or screenshots matter.
- Use the bundled audit script for repeatable extraction.
- - Do not recommend black-hat tactics: no PBNs, bought reviews, cloaking, doorway pages, or spam link packages.
- Be explicit about uncertainty. Many SEO thresholds are heuristics, not laws.
Request triage
Classify the request before answering:
- 1. On-page SEO
- Titles, meta descriptions, H1/H2/H3 structure, canonicals, alt text, internal links.
- Read
references/on-page.md.
- 2. Technical SEO
- Indexing, robots.txt, sitemap, redirects, canonicalization, renderability, structured data, performance.
- Read
references/technical.md.
- 3. Content strategy / keyword work
- Keyword targeting, clustering, cannibalization, content briefs, content gaps.
- Read
references/content-strategy.md.
- 4. Local SEO
- GBP, NAP consistency, local landing pages, reviews, citations.
- Read
references/local-seo.md.
- 5. Link building / backlink evaluation
- Link quality, outreach ideas, risk flags, sustainable acquisition.
- Read
references/link-building.md.
- 6. Full audit
- Run the bundled script first, then deepen with browser/manual inspection if needed.
- Read
references/technical.md and
references/on-page.md.
Runtime requirements and fallbacks
Preferred runtime for the bundled script:
- - Python 3
- INLINECODE9
- INLINECODE10 (
bs4)
If those Python dependencies are missing:
- - do not fail silently
- ask before installing packages in environments where package installation changes the system/runtime state
- if package installs are not approved or not appropriate, fall back to
web_fetch for raw HTML/text inspection and browser for rendered-page inspection
The script is optional support tooling, not a hard requirement for using this skill.
Target safety
Do not run the bundled script against private/internal targets unless the user explicitly intends that and the environment is trusted.
Treat these as blocked by default for routine SEO work:
- - INLINECODE14
- loopback addresses
- private RFC1918 IP ranges
- link-local addresses
- obvious cloud metadata endpoints
- internal-only hostnames or intranet targets
For normal public-site SEO auditing, stick to public HTTP/HTTPS URLs.
Standard workflow for a URL audit
- 1. Run the bundled script:
CODEBLOCK0
- 2. Review the structured output for:
- title/meta quality
- canonical presence
- robots directives
- heading counts
- internal/external links
- image alt gaps
- structured data presence
- robots.txt / sitemap availability
- 3. If the page is JS-heavy or the HTML output looks incomplete, inspect with
browser.
- 4. Produce findings in this order:
-
Critical blockers
-
High-impact improvements
-
Medium / nice-to-have fixes
-
What to do next
Output contracts
Quick page audit
Use when the user wants a fast diagnosis for one page.
Include:
- - summary table
- top 3–7 issues
- exact rewrites if applicable
Full SEO teardown
Use when the user wants a deeper review.
Include:
- - technical section
- on-page section
- content section
- internal linking / structure section
- local SEO section if relevant
- prioritized roadmap
Metadata rewrite pack
Use when the user asks for titles, descriptions, or heading rewrites.
Include:
- - 3–5 title options
- 2–3 meta description options
- suggested H1
- reasoning briefly tied to query intent / CTR / relevance
Default thresholds
Treat these as strong heuristics, not absolute truths:
- - Title tag: ideal ~50–60 chars
- Meta description: ideal ~120–156 chars
- H1 count: exactly 1 preferred
- Thin content risk: often under ~500–600 body words for pages intended to rank
- Strong content target: depends on intent and SERP, not a fixed word count
Reporting style
When reporting, always:
- - cite exact numbers where useful
- distinguish page-specific vs site-wide issues
- explain why something matters
- avoid fake certainty
- prefer actionable fixes over generic theory
Reference map
Read only what you need:
- - INLINECODE16
- INLINECODE17
- INLINECODE18
- INLINECODE19
- INLINECODE20
- INLINECODE21
- INLINECODE22
Bundled script
Use scripts/audit_page.py for repeatable extraction. It is intentionally conservative and should be supplemented with rendered-page inspection when needed.
If the script cannot run because Python dependencies are unavailable, continue with a manual audit using web_fetch and browser rather than treating the skill as unusable.
SEO Companion
使用此技能提供严谨、注重实施的SEO帮助,避免陷入模糊的营销建议。
核心操作规则
- - 区分事实、强启发式规则和情境建议。
- 优先处理技术障碍,然后是元数据/页面问题,接着是内容深度,再是内部链接,最后是站外机会。
- 在此环境中优先使用工具原生检查:
- 使用web_fetch进行快速HTML/文本检索。
- 当渲染后的DOM、JS行为或截图重要时,使用browser。
- 使用捆绑的审计脚本进行可重复的提取。
- - 不推荐黑帽策略:不使用私链网络、购买评论、伪装、门页或垃圾链接包。
- 明确说明不确定性。许多SEO阈值是启发式规则,而非绝对法则。
请求分类
在回答前对请求进行分类:
- 1. 页面SEO
- 标题、元描述、H1/H2/H3结构、规范标签、替代文本、内部链接。
- 阅读references/on-page.md。
- 2. 技术SEO
- 索引、robots.txt、站点地图、重定向、规范化、可渲染性、结构化数据、性能。
- 阅读references/technical.md。
- 3. 内容策略/关键词工作
- 关键词定位、聚类、内容蚕食、内容简报、内容缺口。
- 阅读references/content-strategy.md。
- 4. 本地SEO
- 谷歌商家资料、NAP一致性、本地落地页、评论、引用。
- 阅读references/local-seo.md。
- 5. 链接建设/反向链接评估
- 链接质量、外联思路、风险标志、可持续获取。
- 阅读references/link-building.md。
- 6. 全面审计
- 先运行捆绑脚本,必要时使用浏览器/手动检查深入分析。
- 阅读references/technical.md和references/on-page.md。
运行时要求与回退方案
捆绑脚本的推荐运行环境:
- - Python 3
- requests
- beautifulsoup4 (bs4)
如果缺少这些Python依赖:
- - 不要静默失败
- 在包安装会改变系统/运行时状态的环境中,先询问再安装包
- 如果包安装未获批准或不合适,回退到使用web_fetch进行原始HTML/文本检查,以及使用browser进行渲染页面检查
该脚本是可选的辅助工具,并非使用此技能的硬性要求。
目标安全性
除非用户明确意图且环境可信,否则不要对私有/内部目标运行捆绑脚本。
默认情况下,以下目标在常规SEO工作中被阻止:
- - localhost
- 回环地址
- 私有RFC1918 IP范围
- 链路本地地址
- 明显的云元数据端点
- 仅内部主机名或内网目标
对于正常的公共网站SEO审计,坚持使用公共HTTP/HTTPS URL。
URL审计的标准工作流程
- 1. 运行捆绑脚本:
bash
python3 scripts/audit_page.py
- 2. 审查结构化输出中的:
- 标题/元数据质量
- 规范标签存在性
- robots指令
- 标题数量
- 内部/外部链接
- 图片替代文本缺失
- 结构化数据存在性
- robots.txt / 站点地图可用性
- 3. 如果页面JS密集或HTML输出看起来不完整,使用browser检查。
- 4. 按以下顺序呈现发现:
-
关键障碍
-
高影响改进
-
中等/锦上添花的修复
-
下一步行动
输出约定
快速页面审计
当用户希望对单个页面进行快速诊断时使用。
包括:
- - 摘要表格
- 前3-7个问题
- 精确的重写建议(如适用)
完整SEO拆解
当用户希望进行更深入的审查时使用。
包括:
- - 技术部分
- 页面部分
- 内容部分
- 内部链接/结构部分
- 本地SEO部分(如相关)
- 优先级路线图
元数据重写包
当用户要求重写标题、描述或标题时使用。
包括:
- - 3-5个标题选项
- 2-3个元描述选项
- 建议的H1
- 简要说明与查询意图/点击率/相关性的关联
默认阈值
将这些视为强启发式规则,而非绝对真理:
- - 标题标签:理想约50-60字符
- 元描述:理想约120-156字符
- H1数量:最好恰好1个
- 内容单薄风险:对于旨在排名的页面,正文通常低于约500-600词
- 强内容目标:取决于意图和搜索结果页,而非固定字数
报告风格
报告时始终:
- - 在有用处引用确切数字
- 区分页面特定问题与网站范围问题
- 解释某件事为何重要
- 避免虚假确定性
- 优先提供可操作的修复方案而非通用理论
参考地图
仅阅读所需内容:
- - references/on-page.md
- references/technical.md
- references/content-strategy.md
- references/local-seo.md
- references/link-building.md
- references/platform-playbooks.md
- references/canada-local.md
捆绑脚本
使用scripts/audit_page.py进行可重复的提取。它有意保持保守,必要时应辅以渲染页面检查。
如果由于Python依赖不可用而无法运行脚本,继续使用web_fetch和browser进行手动审计,而不是将此技能视为不可用。