Jake Labate, SEO Professional
Claude AI

Claude SEO Skills For Repeatable Search Work.

This collection documents Claude Agent Skills built for specific SEO tasks. Each skill packages instructions, inputs, and guardrails so the same job is done consistently every time, instead of relying on one-off prompts.

GitHub repositoryJakeLabate/Claude-SEO-Skills npm packagenpx claude-seo-skills install

Skills In This Collection

How I Build These Skills

  • Each skill targets one clear SEO task with explicit inputs, outputs, and constraints so results stay predictable.
  • Guardrails keep Claude inside known-good SEO practice and discourage fabricated metrics or unsupported claims.
  • Skills are designed to be reused across pages and projects, not rewritten as a fresh prompt every time.

How Every Skill Works: Fetch → Extract → Audit → Report

Every skill follows the same four-stage pipeline. Separating fetching the site from extracting what each audit needs from judging it from explaining it is the single most important design idea in the collection, it keeps results reproducible and ties every finding back to an observed fact.

1 · Fetch

A shared fetch_pages.py crawls the site once and stores the raw HTML and response facts in a single page_cache.json, an audit-agnostic snapshot of what the site actually served.

2 · Extract

A pure extractor reads that cache with --from-cache, parses only the fields this audit needs, and writes a plain, reviewable inventory.json. A full audit fetches once and extracts many times instead of crawling per area.

3 · Audit

A second pure script reads the inventory, applies checks and thresholds, assigns severities, and writes audit_report.json. It makes no network calls, so you can re-run it, tweak thresholds, or diff two runs without re-crawling.

4 · Report

Claude turns the findings into a human-readable report with prioritized fixes, and can export it to a Word .docx or a flat findings.csv for sharing with clients and teams.

  • The shared fetch stage makes the network calls; extract and audit are pure and deterministic, so audits are easy to reason about and cheap to repeat. Crawling once into a shared cache means a full audit doesn't re-crawl the site for every area.
  • The inventory is the source of truth: every finding traces back to an observed fact, never a guess. The reporting step is told explicitly never to invent URLs, prices, or claims.
  • Claude handles judgment that needs context (rewriting titles, prioritizing by traffic) while the scripts handle judgment that should be mechanical (counting characters, comparing hosts, resolving redirects).
  • Two link-graph skills: internal-link-audit and site-architecture-audit, keep their own crawler on purpose, because mapping click depth and resolving linked-but-uncrawled URLs needs more than a flat page cache.

What's Inside Each Skill

Every skill is a self-contained folder built to the same shape, so once you understand one you understand them all.

  • SKILL.md, the instructions Claude reads: when to use the skill, the inputs to collect, a numbered workflow, a checks table with severities, and how to produce the report. Its description is what Claude matches a request against, so it names the problems the skill finds and the trigger phrases.
  • references/audit-checks.md, one entry per check, each with a definition (what triggers it), why it matters (the SEO rationale), and the concrete fix. This file is the spec the auditor implements and the reasoning the report cites.
  • references/report-template.md, the exact Markdown structure of the final report, plus guidance that enforces naming the exact URL and value, giving a concrete replacement, capping long tables, and never inventing facts.
  • scripts/: the extractor and auditor for that audit, the shared fetch_pages.py crawler for the page-cache skills, and the shared md_to_docx.py and findings_to_csv.py converters so every skill can hand back a Word document or a flat CSV of findings.

The Severity Model

Every finding is assigned one of four levels, applied consistently across all skills, so each report leads with what matters most.

Severity Meaning Examples
High Actively harms indexing or ranking, or breaks the page. Fix first. Missing title, sitemap URL 404s, Disallow: /, broken canonical
Medium A real SEO quality problem, but not an emergency. Title too long, page canonicalizes elsewhere, slow LCP proxy
Low Hygiene and polish. Invalid changefreq, generic filename, future lastmod
Info Worth a look, not counted as an issue. Title / H1 divergence, intentional patterns

Principles Behind The Skills

The indexable contract

Every URL you submit to search engines, in a sitemap, an internal link, a canonical, should be a final, canonical, indexable 200 page. Anything that redirects, 404s, is noindexed, is blocked, or canonicalizes elsewhere breaks that contract and gets flagged. Many checks are applications of this one idea.

Standard-library only

The Python scripts depend on nothing but the Python standard library, no pip install, no virtualenv. If you have Python 3.8+, they run. HTML, XML, HTTP, and gzip are all handled with built-in modules.

Shareable reports

The report is produced as Markdown, then md_to_docx.py converts it to a .docx and findings_to_csv.py flattens the findings to a .csv, both with no dependencies. The .docx conversion runs after Claude's judgment, so prioritized fixes and rewritten titles are preserved; the document is the report reformatted, not re-derived from the JSON.

Install The Skills

Every skill lives in the claude-seo-skills npm package. Install all of them into your global Claude skills directory (~/.claude/skills/) with a single command:

npx claude-seo-skills install

  • Install only specific skills: npx claude-seo-skills install sitemap-audit robots-txt-audit
  • Install into the current project instead of the global directory: npx claude-seo-skills install --project
  • List everything available: npx claude-seo-skills list

The audit scripts need Python 3.8+, with no pip install required, every script uses only the Python standard library. You can also copy any skill folder into your Claude skills directory by hand; Claude invokes a skill automatically when a request matches its description. Full instructions are in the repository README below.

Repository README

The README from the Claude-SEO-Skills repository is embedded below, covering the skill architecture, the full list of available skills, and how to install them. To see what the skills produce, browse the sample reports in examples/; to understand the design behind every skill, read docs/CONVENTIONS.md.