The Claude SEO Audit Skill: What It Is and How to Run One

A Claude SEO audit skill is a folder of instructions, scripts, and reference files that teaches Claude to run a structured SEO audit of a website. You install it once, then ask Claude Code to audit a URL, and it works through crawlability, meta tags, structured data, Core Web Vitals, and content quality in a repeatable order instead of improvising. Several open source versions exist today, all MIT licensed, and the good ones pair deterministic scripts with the model's judgment so the factual checks never get hallucinated.

This post explains what an Agent Skill actually is, walks through what an SEO audit skill does step by step, compares the notable open source options, shows how to install one, and is honest about what it catches versus what still needs a human.

What a Claude skill is

Anthropic announced Agent Skills on October 16, 2025. A skill is a directory containing a SKILL.md file with YAML frontmatter, plus optional supporting files: extra markdown guides, executable scripts, and reference material. Anthropic's own description is "folders that include instructions, scripts, and resources that Claude can load when needed."

The frontmatter has two required fields, name and description. The description is what Claude matches your request against when deciding whether to trigger the skill, so it states both what the skill does and when to use it. Skills load in stages, which Anthropic calls progressive disclosure. At startup only the name and description enter the context window, roughly 100 tokens per skill. When your request matches, Claude reads the SKILL.md body, which Anthropic caps in practice at under 5,000 tokens. Bundled scripts never load into context at all; Claude runs them and only their output consumes tokens.

That last part matters for SEO audits specifically. A script that parses robots.txt or measures a title tag length gives the same answer every time. The skill format lets those checks run as code while the model handles the judgment calls, like whether an H1 actually matches search intent.

One constraint worth knowing: skills run in different environments depending on the surface. On the Claude API the sandbox has no network access, so a skill cannot fetch your website there. Claude Code has full network access, which is why every SEO audit skill in circulation targets Claude Code. On claude.ai, custom skills upload as zip files under Settings and are available on Pro, Max, Team, and Enterprise plans, with network access depending on user and admin settings.

What an SEO audit skill does, step by step

The published skills converge on a similar workflow. Using the most detailed public SKILL.md files as the reference, a full audit runs like this:

1. Fetch and render the target. The skill pulls the homepage, and in the more advanced skills captures both raw and rendered HTML so JavaScript-heavy sites are evaluated on what search engines actually see.

2. Classify the site. The skill infers the business type from the homepage, such as SaaS, local service, e-commerce, or publisher, because the checks that matter differ by type. A local business gets NAP consistency and LocalBusiness schema checks; a store gets Product schema checks.

3. Crawl. The site-wide skills follow internal links, respect robots.txt, and cap the crawl. The claude-seo plugin caps at 500 pages with a 30 second timeout per page and a maximum of 3 redirect hops.

4. Run the checks. This is the bulk of the work: robots.txt parsing, sitemap validation, canonical tags, redirect chains, HTTPS and security headers, title and meta description lengths, heading structure, image alt text, JSON-LD structured data validation, internal linking and orphan pages, mobile signals, and Core Web Vitals (LCP, INP, CLS). Most skills also now include AI search readiness: whether AI crawlers are blocked in robots.txt and whether content is structured for extraction into AI answers.

5. Score and prioritize. The output is a weighted health score and a ranked action plan. In claude-seo the weights are technical SEO 22 percent, content quality 23 percent, on-page 20 percent, schema 10 percent, performance 10 percent, AI search readiness 10 percent, and images 5 percent. Reports land as markdown or PDF with fixes ordered critical to low.

Checks that need live Google data, like field Core Web Vitals from CrUX or indexation status from Search Console, only run if you supply API credentials. Without them the skills fall back to lab estimates or skip those sections, and the better ones say so in the report.

The notable open source SEO audit skills

These are the four projects with public code and documented check lists as of August 2026. All are MIT licensed and free.

claude-seo (AgriciDaniel)

The most comprehensive option: an open source plugin with 25 sub-skills and 18 specialist agents covering technical SEO, E-E-A-T content quality, schema, GEO and AI search, local SEO, e-commerce, and international SEO. The /seo audit command crawls up to 500 pages and delegates to as many as 15 specialist subagents in parallel, then aggregates a 0 to 100 health score and a prioritized action plan. Optional integrations add DataForSEO, Ahrefs, Firecrawl, and free Google APIs. The project states it ships no telemetry and works without any paid service.

seo-audit (wonfull888)

A single skill built from Google Search Central guidance and the Ahrefs SEO checklist. It runs 92 checks in full mode across four dimensions: technical SEO (29 checks, 32 percent weight), on-page (27 checks, 29 percent), content quality and E-E-A-T (33 checks, 36 percent), and local SEO (3 checks, 3 percent). Full mode needs a free Google PageSpeed API key, which carries a 25,000 request per day quota; basic mode runs 84 checks with zero setup but skips Core Web Vitals. Installation is a git clone and a copy into your skills folder.

seo-audit-skill (JeffLi1993)

The most starred of the group at 535 stars. It audits single pages rather than whole sites and uses an explicit two-layer architecture: Python scripts handle the deterministic checks (HTTP status, robots.txt parsing per RFC 9309, title lengths, JSON-LD field validation), and the model only makes semantic judgments on fields the scripts flag. The stated rationale is no hallucination on factual checks and no blind spots on semantic ones. It ships basic and full tiers, with the full tier requiring a PageSpeed API key, and installs via npx skills add or as a Claude Code plugin.

seo-claude-code-skills (ccforseo)

A set of 8 free skills where /seo-audit is one command among several, alongside schema generation, keyword clustering, and an AI visibility check. The audit covers 10 categories with more than 80 checks and grades each category A to F. It uses only Claude Code's built-in web fetching and bash, with no API keys required.

How to install one

In Claude Code, skills are just directories. Personal skills live in ~/.claude/skills/ and project skills in .claude/skills/, and Claude discovers them automatically. The simplest install is the wonfull888 pattern:

git clone https://github.com/wonfull888/seo-audit.git && cp -r seo-audit ~/.claude/skills/

Then run /seo-audit https://example.com. The plugin-based projects install through Claude Code's plugin system instead, for example /plugin marketplace add JeffLi1993/seo-audit-skill followed by /plugin install seo-audit-skill.

One caution that comes straight from Anthropic's documentation: only install skills from sources you trust, and audit the bundled scripts before running them. A skill directs Claude to execute code on your machine, so treat installation like installing software. All four projects above have public source you can read first, which is a real advantage over opaque tools.

What it catches versus a human audit

The skills are genuinely strong on the mechanical layer. Robots.txt logic, canonical conflicts, redirect chains, schema field validation, title and description lengths, heading hierarchy, alt text coverage, and orphan pages are deterministic checks, and scripts do them faster and more consistently than a person clicking through Screaming Frog exports. The AI search readiness checks are also ahead of many human audit templates, which have not caught up to checking AI crawler access or content extractability.

The gaps are equally real. Without credentials, the skills have no field performance data, no Search Console indexation truth, and no backlink profile; several of them add these through optional Google, Moz, or DataForSEO integrations, but that is setup work the skill cannot do for you. More fundamentally, no checklist knows your business. Deciding which of forty flagged issues actually moves revenue, whether a page should exist at all, or how the audit connects to a content and link strategy is judgment work. Treat the skill's ranked action plan as a strong first draft of priorities, not a strategy.

A fair summary: an SEO audit skill replaces the data-gathering half of a human audit and does it in minutes for free. It does not replace the interpretation half, and the honest projects say so in their own documentation, with one limiting itself to single pages and others gating their deepest checks behind credentials you must supply.

Where this fits

If you run a finance firm rather than an SEO agency, the interesting thing here is not this one skill. It is the pattern: a folder of instructions and scripts that turns a general model into a specialist that runs a complex, multi-step workflow on command. The same mechanism that audits a website can screen a deal, check a data room, or draft a diligence memo against your firm's checklist. That is the kind of system we build at consultance.ai, custom finance AI like AI deal desk systems and AI for private equity workflows, with the first system live in about six weeks and the client owning the code. If you want to see the skill pattern applied to finance work first, the free prompt and skill library is the place to start.

For the SEO use case itself, install one of the open source skills above. They cost nothing, the code is readable, and a monthly run will catch the mechanical regressions before Google does.

Sources

• Anthropic, Agent Skills documentation (fetched 2026-08-24)
• Anthropic, Introducing Agent Skills announcement (fetched 2026-08-24)
• AgriciDaniel/claude-seo repository and seo-audit SKILL.md (fetched 2026-08-24)
• wonfull888/seo-audit repository (fetched 2026-08-24)
• JeffLi1993/seo-audit-skill repository (fetched 2026-08-24)
• ccforseo/seo-claude-code-skills repository (fetched 2026-08-24)