Agent Skill for Claude Code & Codex
SEO your agent can actually reproduce.
Google publishes the criteria. This skill turns them into steps an agent follows — technical SEO, structured data, Core Web Vitals, E-E-A-T, AI search, and full site audits. Markdown only: no API key, no MCP server, nothing to run.
- Markdown only
- English & Japanese
- MIT licensed
Why this one
Three things it does differently.
Zero dependency
Markdown files and nothing else. No API keys to provision, no MCP server to keep alive, no install script. Copy the folder and it works offline.
Strong on edge and static sites
Dedicated reference layers for Astro and Cloudflare Workers/Pages: island hydration against INP and LCP, the blind spots of _redirects and _headers next to Worker code, dynamic sitemaps out of D1 or KV, crawler verification at the edge.
Staleness is tracked, not hoped away
Every reference file carries a last_verified date that CI enforces. Corrected claims trace back to a primary source in research-notes.md, releases follow semver, and a scheduled workflow opens a re-verification issue every month.
Coverage
Eight reference files behind one skill.
SKILL.md routes the request; the reference file that matches gets loaded. Both language versions ship the same eight.
Technical SEO technical-seo.md
robots.txt, sitemaps, canonical, hreflang, index coverage, redirect and status-code handling.
Content SEO content-seo.md
Meta tags, heading structure, internal linking, and E-E-A-T signals that reviewers can actually check.
Structured data structured-data.md
JSON-LD templates: Article, Product, LocalBusiness, BreadcrumbList, VideoObject, Organization, WebSite, Event.
Core Web Vitals core-web-vitals.md
LCP, INP and CLS — the causes, the fixes as code, and the measurement scripts to confirm them.
AI search / GEO / LLMO ai-search.md
AI Overviews and AI Mode, crawler control for GPTBot, ClaudeBot and Google-Extended, citation-ready content, entity clarity, information gain.
Astro SEO astro-seo.md
Client directives against hydration cost, @astrojs/sitemap, ClientRouter, content collections. Verified against Astro 6.
Edge SEO edge-seo.md
Cloudflare Workers and Pages: _redirects, _headers, X-Robots-Tag, HTMLRewriter, D1/KV sitemaps, Pages Functions.
Audit workflow audit-workflow.md
A staged audit process and the report format that comes out of it, so two runs on the same site are comparable.
Usage
Ask in plain language.
The skill description carries the routing. You do not name the file, and you do not name the skill.
- Optimize the meta tags for this page
- Add Article structured data to this blog post
- Run an SEO audit on this site
- How can I improve LCP here?
- Should this site have an llms.txt?
Claude loads seo-mastery, opens the reference file that matches, and answers against the criteria in it — not against a half-remembered blog post.
Install
Two commands.
1Add the marketplace
claude plugin marketplace add kpab/seo-mastery-agent-skills2Install the plugin
claude plugin install seo-mastery@seo-mastery-agent-skills
Installs both the English and Japanese skills, and keeps them updatable through the plugin manager.
Prefer to copy the files?
Clone the repository and copy .claude/skills/seo-mastery into your project. For Codex the path is .agents/skills — it scans that folder in every directory from your working directory up to the repository root. A skill is SKILL.md plus its reference files: copy the whole folder, not just SKILL.md.
Layout
What lands in your repo.
Two skills, English and Japanese, kept in sync by CI. English is the source of truth.
.claude/skills/ ├── seo-mastery/ # English │ ├── SKILL.md # routing + overview │ ├── technical-seo.md │ ├── content-seo.md │ ├── structured-data.md │ ├── core-web-vitals.md │ ├── ai-search.md │ ├── astro-seo.md │ ├── edge-seo.md │ └── audit-workflow.md └── seo-mastery-jp/ # Japanese (same nine files)
Fetching SKILL.md alone leaves the skill incomplete — the reference files are where the criteria live.
Freshness
What is guaranteed, and what is not.
SEO guidance rots. This repository does not claim every line is currently correct; it guarantees the process and the date stamp.
Date-stamped
Every file under .claude/skills carries last_verified in its frontmatter. CI checks it is present, valid, not in the future, and identical between the English and Japanese versions.
Sourced
Claims added or corrected in a verification pass are traceable to a primary source in docs/research-notes.md — the vendor documentation, not a summary of it.
Versioned
Semver read for a knowledge base: MAJOR for reversed guidance or a removed file, MINOR for new knowledge, PATCH for corrections and freshness. The release workflow fails if SKILL.md, both manifests, the CHANGELOG heading and the git tag disagree.
Re-checked on a schedule
A scheduled workflow opens a re-verification issue every month, so a stale claim surfaces as work rather than sitting quietly in a file.
FAQ
Questions worth answering up front.
Does it work with Codex, or only Claude Code?
English or Japanese — which one do I install?
Does it guarantee rankings or AI citations?
Does it need an API key or a server?
Which frameworks are covered?
Put the criteria in the repo.
MIT licensed. Pull requests and issues welcome — the English and Japanese versions have to stay in sync.