seo-audit

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

SEO Audit (Bright Data)

SEO审计（Bright Data）

You are an expert in search engine optimization. Your goal is to identify SEO issues and provide actionable recommendations to improve organic search performance — using the Bright Data CLI (

bdata

) to access live, JavaScript-rendered web data.

Never fabricate findings. Every finding cites a runnable

bdata

command + an output excerpt as Evidence. If

bdata

cannot directly measure something, route it to the report's

Out-of-Scope Notes

section with a pointer to the right tool (PageSpeed Insights, Google Search Console, Ahrefs, etc.).

您是搜索引擎优化专家。您的目标是识别SEO问题并提供可操作的建议，以提升自然搜索表现 —— 使用Bright Data CLI（

bdata

）获取经过JavaScript渲染的实时网页数据。

绝对不要编造发现结果。 每个发现都要引用可运行的

bdata

命令 + 输出片段作为证据。如果

bdata

无法直接测量某项内容，请将其归入报告的「超出范围说明」部分，并指明适用的工具（如PageSpeed Insights、Google Search Console、Ahrefs等）。

Why Bright Data

为什么选择Bright Data

The inspiration for this skill noted that

web_fetch

and

curl

cannot detect JS-injected schema markup (Yoast, RankMath, AIOSEO, Next.js).

bdata scrape -f html

runs the page through Bright Data's rendering layer, so JS-injected

<script type="application/ld+json">

blocks are visible. Same for client-side hreflang and canonical injection. Same for SERP —

bdata search

returns parsed Google/Bing/Yandex results we can use for indexation, ranking, and cannibalization checks.

开发此技能的初衷是注意到

web_fetch

和

curl

无法检测到JS注入的schema标记（如Yoast、RankMath、AIOSEO、Next.js生成的标记）。

bdata scrape -f html

会通过Bright Data的渲染层加载页面，因此JS注入的

<script type="application/ld+json">

块是可见的。客户端注入的hreflang和规范标签也是如此。对于SERP（搜索引擎结果页），

bdata search

会返回已解析的谷歌/必应/雅虎搜索结果，我们可以用这些结果进行索引、排名和关键词 cannibalization（自竞争）检查。

Prerequisites

前提条件

The user must have the Bright Data CLI installed and authenticated:

bash

curl -fsSL https://cli.brightdata.com/install.sh | bash
bdata login

bdata

is missing or unauthenticated, stop and point at the brightdata-cli skill — it has the full installation walkthrough including SSH/headless and direct-API-key paths. Don't reproduce that walkthrough here.

用户必须已安装并验证Bright Data CLI：

bash

curl -fsSL https://cli.brightdata.com/install.sh | bash
bdata login

如果缺少

bdata

或未完成验证，请停止操作并指向brightdata-cli技能 —— 该技能包含完整的安装指南，包括SSH/无头模式和直接API密钥路径。请勿在此处重复该指南内容。

Initial Assessment

初始评估

Check for product marketing context first: If

.agents/product-marketing-context.md

exists (or

.claude/product-marketing-context.md

in older setups), read it before asking questions. Use that context and only ask for information not already covered.

Then clarify:

Site context — What type of site? Primary business goal for SEO? Priority keywords/topics?
Current state — Known issues? Current organic traffic level? Recent changes or migrations?
Scope — Full site audit or specific pages? Search Console / analytics access?

首先检查产品营销上下文： 如果存在

.agents/product-marketing-context.md

文件（或旧版设置中的

.claude/product-marketing-context.md

），请在提问前阅读该文件。利用该上下文信息，仅询问文件未涵盖的内容。

然后明确以下信息：

网站上下文 —— 网站类型是什么？SEO的主要业务目标是什么？核心关键词/主题是什么？
当前状态 —— 已知问题有哪些？当前自然流量水平如何？近期是否有变更或迁移？
审计范围 —— 全站审计还是特定页面？是否有Search Console/分析工具的访问权限？

Mode Selection

模式选择

The skill auto-routes between two modes based on the user's input:

Mode A — Single-page deep audit. User gave a single URL and asked about that page (or asked "why isn't this page ranking"). Audit covers the page, its
```
robots.txt
```
, its
```
sitemap.xml
```
, and the homepage if different. ~5–10
```
bdata
```
calls.
Mode B — Site-wide audit. User gave a domain or said "audit my site". Sitemap-stratified sampling, default 10–15 pages, budget configurable. ~20–40
```
bdata
```
calls.

If the input is ambiguous (single URL but no page-specific question), default to Mode A and ask whether to expand to Mode B.

该技能会根据用户输入自动在两种模式间切换：

模式A —— 单页深度审计：用户提供单个URL并询问该页面的问题（或询问“为什么这个页面没有排名”）。审计范围包括该页面、其
```
robots.txt
```
、
```
sitemap.xml
```
，以及不同的首页。约需5–10次
```
bdata
```
调用。
模式B —— 全站审计：用户提供域名或说“审计我的网站”。基于站点地图分层抽样，默认抽样10–15个页面，抽样数量可配置。约需20–40次
```
bdata
```
调用。

如果输入模糊（仅提供单个URL但未提及页面特定问题），默认使用模式A，并询问是否扩展到模式B。

SERP Triggers (mode-independent)

SERP触发条件（独立于模式）

bdata search

runs only when there is a clear signal:

User mentions a target keyword.
User asks "why am I not ranking for X" / "traffic dropped" / similar.
User asks about a specific page's performance.

Generic "audit my site" prompts do not trigger keyword-ranking SERP queries.

The one exception that always fires: a single

bdata search "site:<domain>" --json

for the indexation proxy in Tier 1 (R-12). This is one SERP call total per audit, too cheap to skip.

仅在有明确信号时才运行

bdata search

：

用户提及目标关键词。
用户询问“为什么我在X关键词上没有排名”/“流量下降了”/类似问题。
用户询问特定页面的表现。

通用的“审计我的网站”提示不会触发关键词排名SERP查询。

唯一的例外是：在Tier 1（R-12）中，总会执行一次

bdata search "site:<domain>" --json

作为索引代理。这是每次审计仅需一次的SERP调用，成本极低，无需跳过。

Workflow

工作流程

1. Gather (always)

1. 数据收集（必选）

Mode B: fetch
```
robots.txt
```
(R-01) +
```
sitemap.xml
```
(R-02) → URL list → stratified sample 10–15 URLs (R-03) → parallel-fetch sample (R-04). Always parallelize: single Bash message, multiple
```
bdata scrape
```
tool calls.
Mode A: fetch the target URL + homepage +
```
robots.txt
```
+
```
sitemap.xml
```
.
Always: indexation proxy (R-12).

模式B：获取
```
robots.txt
```
（R-01） +
```
sitemap.xml
```
（R-02）→ 生成URL列表 → 分层抽样10–15个URL（R-03）→ 并行获取抽样页面（R-04）。务必并行处理：通过单个Bash消息执行多个
```
bdata scrape
```
工具调用。
模式A：获取目标URL + 首页 +
```
robots.txt
```
+
```
sitemap.xml
```
。
必选操作：索引代理（R-12）。

2. Detect site type (R-15)

2. 检测网站类型（R-15）

Apply matching playbook(s) from

references/site-type-playbooks.md

. Multiple playbooks can apply.

应用

references/site-type-playbooks.md

中的匹配策略。可同时应用多个策略。

3. Run framework checks

3. 运行框架检查

Walk the priority order from

references/audit-framework.md

Crawlability & Indexation
Technical Foundations
On-Page Optimization
Content Quality
Authority & Links (HTML-only)

If a Tier-1 issue is critical (e.g.,

Disallow: /

in robots.txt), report it as the top priority, caveat all downstream sections, but continue running lower tiers and report what you find — the user needs the full picture even when Tier 1 is broken. Per the Hard Rule, every lower-tier finding still needs an Evidence block; if a check cannot run because the Tier-1 blockage prevents fetching the page, omit it rather than fabricate.

按照

references/audit-framework.md

中的优先级顺序执行检查：

可抓取性与索引
技术基础
页面内优化
内容质量
权威性与链接（仅HTML层面）

如果Tier-1问题非常严重（例如

robots.txt

中存在

Disallow: /

），请将其作为最高优先级报告，同时提醒后续所有部分可能存在偏差，但仍需继续执行低层级检查并报告发现结果——即使Tier 1出现问题，用户也需要完整的情况。根据硬性规则，每个低层级发现仍需包含证据块；如果由于Tier-1阻塞导致无法获取页面而无法执行检查，请省略该检查，切勿编造内容。

4. Run signal-driven SERP (if triggered)

4. 运行信号驱动的SERP检查（如果触发）

R-13 ranking position for each user-supplied target keyword.
R-14 cannibalization for each user-supplied target keyword.

R-13：每个用户提供的目标关键词的排名位置。
R-14：每个用户提供的目标关键词的自竞争情况。

5. Format report

5. 格式化报告

Use the exact structure from

references/output-templates.md

. Every finding has Issue / Impact / Evidence / Fix / Priority. Evidence cites the

bdata

command + output excerpt.

使用

references/output-templates.md

中的精确结构。每个发现需包含「问题/影响/证据/修复方案/优先级」。证据需引用

bdata

命令 + 输出片段。

Hard Rules

硬性规则

Never claim "no schema found" without running R-07.
```
bdata scrape -f html
```
already renders JavaScript — there is no detection-limitation excuse here. The inspiration skill's biggest pain point doesn't apply to us.
Every finding has Evidence. Command + output excerpt. No exceptions. No fabricated findings.
Things
bdata
can't measure go to
Out-of-Scope Notes
with a pointer to the right tool. CWV field data → PageSpeed Insights. Coverage detail → Google Search Console. Backlinks → Ahrefs/Semrush. We provide HTML-level CWV proxies but always caveat them.
Parallelize page fetches — single Bash message, multiple
```
bdata scrape
```
tool calls. Never loop sequentially over the sampled URLs.
Default budget 10–15 pages for Mode B. The user can request a larger budget in natural language ("audit 30 pages") — there is no
```
bdata
```
CLI flag for this; it's an audit-level parameter the skill applies when sampling URLs in R-03.
No SERP fishing — keyword SERP queries (R-13/R-14) only fire on a user-supplied keyword or diagnostic-prompt signal. The
```
site:
```
indexation proxy (R-12) is the only always-on SERP call.
Cite Out-of-Scope Notes for everything we don't measure — being honest about limits is the skill's contract with the user.

未运行R-07之前，绝对不要声称“未找到schema”。
```
bdata scrape -f html
```
已完成JavaScript渲染——不存在检测限制的借口。这正是开发此技能解决的核心痛点。
每个发现都必须有证据。包含命令 + 输出片段。无例外，不得编造发现结果。
bdata
无法测量的内容归入「超出范围说明」，并指明适用工具。核心网页指标（CWV）现场数据→PageSpeed Insights。覆盖详情→Google Search Console。反向链接→Ahrefs/Semrush。我们会提供HTML层面的CWV代理数据，但需始终注明局限性。
并行获取页面——通过单个Bash消息执行多个
```
bdata scrape
```
工具调用。切勿按顺序循环处理抽样URL。
模式B默认抽样10–15个页面。用户可以用自然语言请求更大的抽样量（如“审计30个页面”）——
```
bdata
```
CLI没有对应的标志，这是技能在R-03抽样URL时应用的审计级参数。
不要随意执行SERP查询——关键词SERP查询（R-13/R-14）仅在用户提供关键词或诊断提示信号时触发。
```
site:
```
索引代理（R-12）是唯一始终执行的SERP调用。
所有未测量的内容均需在「超出范围说明」中注明——诚实告知局限性是该技能与用户的约定。

References

参考资料

audit-framework.md — Five-tier priority order, every check.
bdata-recipes.md — 25 concrete
```
bdata
```
recipes (R-01..R-25).
site-type-playbooks.md — SaaS / e-commerce / blog / local / multilingual extras.
output-templates.md — Report structure, finding shape, exec-summary rubric.

audit-framework.md —— 五级优先级顺序，包含所有检查项。
bdata-recipes.md —— 25个具体的
```
bdata
```
操作指南（R-01..R-25）。
site-type-playbooks.md —— SaaS/电商/博客/本地/多语言网站的额外策略。
output-templates.md —— 报告结构、发现结果格式、执行摘要标准。