Community寫作與編輯github.com

seo-page-audit

Audits one web page against one target search query and rewrites what matters. Checks search intent match, title tag, meta description, H1, heading outline, answer-first copy, internal links, structured data and E-E-A-T signals, then returns a prioritized fix list plus a rewritten title, meta description, H1 and outline. Use when the user asks to audit, review, optimize or rewrite a single page for SEO, wants a page to rank for a specific keyword, asks why a page does not rank, or needs a pre-launch SEO check of one URL. Not for full-site technical audits (use website-audit).

seo-page-audit 是什麼?

seo-page-audit is a Claude Code agent skill that audits one web page against one target search query and rewrites what matters. Checks search intent match, title tag, meta description, H1, heading outline, answer-first copy, internal links, structured data and E-E-A-T signals, then returns a prioritized fix list plus a rewritten title, meta description, H1 and outline. Use when the user asks to audit, review, optimize or rewrite a single page for SEO, wants a page to rank for a specific keyword, asks why a page does not rank, or needs a pre-launch SEO check of one URL. Not for full-site technical audits (use website-audit).

相容平台~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/troiestudio/claude-skills/tree/HEAD/skills/seo-page-audit

Installed? Explore more 寫作與編輯 skills: steipete/notion, langchain-ai/langchain, bytedance/podcast-generation · View all 6 →

在你喜歡的 AI 中提問

開啟一個已預先載入此 Agent Skill 的新對話。

說明文件

SEO page audit

One page, one target query, one verdict. The page wins when it answers the query better than the pages already ranking, and when that answer is easy for a search engine or an AI assistant to extract.

Hard rules

  1. One page serves one primary intent. If the page targets several intents, say so and recommend a split before rewriting anything.
  2. Never invent facts, figures, reviews, credentials or author bios. Every claim in the rewrite comes from the page, the user or a source you cite with its date. Mark gaps as [NEEDS INPUT: ...].
  3. Never promise a ranking, a traffic gain or an AI citation. State what the fix changes and why it should help.
  4. Do not touch robots.txt, canonicals, redirects or noindex tags without flagging the change and getting explicit approval.
  5. Measure lengths in characters and pixels honestly. Title and meta length rules are guidance for truncation, not ranking factors.

Inputs to collect

Ask for anything missing, in one message, before starting:

  • URL or page content (HTML, a pasted text, or a file path in the repo).
  • Target query and 2 to 5 close variants. If the user has none, propose candidates from the page content and ask them to pick.
  • Market: country, language, and city if the business is local.
  • Page goal: the one action the visitor should take (call, buy, book, request a quote, sign up).
  • Optional data: Search Console queries and clicks for this URL, current position, competitor URLs.

If you have web access, fetch the page and the current top results yourself. If not, work from what the user provides and say which checks you could not run.

Workflow

Step 1. Classify the intent

Decide the dominant intent of the query: informational, commercial investigation, transactional, local or navigational. Look at what ranks today (format, depth, page type). Write one sentence: "Searchers for [query] want [X], and the ranking pages give them [format]." Then state whether the audited page matches that format. A mismatch is always finding number one.

Step 2. Review the competition

For the top 5 organic results (fetched or provided), note: page type, word range, headings covered, unique elements (calculator, table, pricing, local proof), and what none of them covers well. This gap is where the rewrite earns its place. If no SERP data is available, skip and say so.

Step 3. Run the on-page checklist

Go through references/checklist.md section by section. Score each item OK, FIX or MISSING, and record the evidence (the exact current text, the tag, the count). Do not score from memory of the page. Quote it.

Step 4. Check E-E-A-T signals

Look for first-hand experience (photos of real work, specific cases, process details), identified author or team with role, sources and dates on factual claims, business identity (legal name, address, contact), reviews that can be verified, and an "updated on" date where freshness matters. List what exists and what is missing. Do not fabricate any of it.

Step 5. Check structured data

List the JSON-LD types present and validate them against the visible content. Recommend only types that match the page (see references/schema-snippets.md). Note that Google shows FAQ rich results only for a limited set of authoritative sites, so FAQPage markup is for clarity, not for a guaranteed rich result.

Step 6. Prioritize

Rank every FIX and MISSING item with the priority grid below. Keep the list short enough to act on: at most 12 items, grouped by priority.

PriorityMeaning
P1Blocks ranking or indexing, or breaks intent match. Fix first.
P2Clear relevance or click-through gain, low effort.
P3Improvement worth doing in the next content pass.

Each item also gets an effort tag: S (under 30 minutes), M (half a day), L (needs new content, design or development).

Step 7. Rewrite

Produce, in this order:

  1. Title tag: target query near the start, brand at the end, roughly 50 to 60 characters. Give 2 options with their character count.
  2. Meta description: roughly 140 to 160 characters, the promise, one concrete proof from the page, one action. Give 2 options with counts.
  3. H1: one only, natural language close to the query, not a copy of the title.
  4. Answer block: the 2 to 3 sentences that sit under the H1 and answer the query on their own.
  5. Outline: H2 and H3 structure with one line per section on what it must contain and which source supplies it.
  6. Internal links: 3 to 6 suggested links in and out, with descriptive anchor text and the target URL.
  7. Structured data: the JSON-LD block, filled only with facts that appear on the page.

Output template

# SEO page audit: [URL]
Target query: [query] | Market: [country, language, city] | Date: [YYYY-MM-DD]

## Verdict
[3 sentences: intent match yes or no, the biggest problem, the expected effect of the fixes.]

## Intent and competition
[Intent sentence. Table of the top results: URL, type, what they cover, gap.]

## Findings
| # | Priority | Effort | Area | Current | Fix |
| --- | --- | --- | --- | --- | --- |

## E-E-A-T
Present: ...
Missing: ...

## Rewrite
Title (option A, NN chars): ...
Title (option B, NN chars): ...
Meta (option A, NNN chars): ...
Meta (option B, NNN chars): ...
H1: ...
Answer block: ...
Outline: ...
Internal links: ...
JSON-LD: ...

## Needs input
[List of [NEEDS INPUT] items, each with why it matters.]

## Not checked
[Checks you could not run and why.]

Quality checklist (run before delivering)

  • The intent sentence is written and the format match is judged explicitly.
  • Every finding quotes the current state, not a paraphrase.
  • Character counts are real counts of the proposed strings.
  • The rewrite contains no fact that is absent from the page or the user's input.
  • Only one H1 is proposed, and it differs from the title.
  • Structured data matches visible content and uses valid schema.org types.
  • No promise of ranking, traffic or AI citation anywhere in the output.
  • Unrun checks are listed under "Not checked".

Files

  • references/checklist.md: the full on-page checklist with pass criteria.
  • references/schema-snippets.md: JSON-LD patterns per page type, with rules.
  • examples/example-audit.md: a complete sample audit for a fictional company.

Individual skills in this repo

This repo contains 2 individual skills — each has its own dedicated page.

相關技能