Setup guide

Install the technical SEO agent

This guide installs the six modules and gets your first report out. Count ten minutes, eight of them installing tools if you do not have them yet.

In short, read this first

  • The modules are folders. You copy them into ~/.claude/skills/.
  • Claude Code picks them up on the next launch. There is nothing to configure.
  • One module needs two extra tools: Lighthouse and jq.
  • Two modules need a Search Console export.
  • The other three run on a URL, a sitemap or your code.
  • You trigger a module by writing your request, not with a command.
  • Each module returns a report in markdown.
  • No module publishes or edits your live site.

The technical SEO agent, in two lines

Six modules that run inside Claude Code: indexing, Core Web Vitals, structured data, internal linking by structure, internal linking from Search Console, cannibalization.

Each one takes a precise input, runs a fixed pipeline and returns a report. None of them touches your site.

See the six modules in detail

1. What you need

Claude Code installed on your machine.

A terminal open in your project folder, or in any folder if you are auditing a live site.

Node.js, to install Lighthouse.

Search Console access to the site, for two of the six modules.

Without Search Console, four modules out of six still run.

2. Drop in the six modules

Unzip the archive straight into your skills folder.

unzip agent-seo-technique.zip -d ~/.claude/skills/

Each folder holds a SKILL.md file. That file is what Claude Code reads.

Do not rename the folders. The folder name is the module name.

If ~/.claude/skills/ does not exist yet, create it before unzipping.

mkdir -p ~/.claude/skills/

3. Check that Claude Code sees them

List the folder.

ls ~/.claude/skills/

You should see the six names.

indexation-check
seo-core-web-vitals
seo-donnees-structurees
maillage-systeme
maillage-interne-gsc
seo-cannibalisation

Quit Claude Code, then reopen it. It reads skills at startup.

A module missing from this list will never trigger, whatever you write.

4. Install Lighthouse and jq

The Core Web Vitals module calls Lighthouse locally and slices its results with jq.

npm install -g lighthouse
brew install jq

Check that both answer.

lighthouse --version
jq --version

You would rather not install anything globally? Replace lighthouse with npx lighthouse in the commands.

The first run is slower, the result is the same.

The other five modules need neither of these tools.

The two modules that read your Search Console

Internal linking returns the parent, child and grandchild hierarchy, then the list of links to place with their anchor and sentence context.

Cannibalization sorts conflicts into four types and decides: 301 redirect, merge, differentiation, or nothing at all.

See what each module returns

5. Pull a clean Search Console export

Open Search Console, Performance tab.

Pick the period. Three months for cannibalization, six months for internal linking.

Click Export, then CSV.

The archive holds one file per tab. You need Pages and Queries.

Unzip it into your project folder.

Do not re-save the files from Excel. The modules read the original columns.

6. The first request

There is no command to remember. You write what you want.

indexing audit of my-site.com, the URL list is in sitemap.xml

Claude Code recognizes the request and loads the matching module.

The other five trigger the same way.

Core Web Vitals audit of https://my-site.com/sitemap.xml cannibalization audit, the GSC export is in ./gsc-queries.csv internal linking plan from ./gsc-pages.csv internal linking audit of the site, without going through Search Console set up JSON-LD structured data on this project

Start with indexing. As long as a page is blocked, whatever the other modules measure is worthless.

7. Read the report

Each module returns a markdown file.

Critical anomalies first, minor ones after, fixes ranked by priority.

The indexing module separates "not indexed" from "not testable". A page that cannot be tested is not a problem, it is a measurement limit.

The cannibalization module never recommends a 301 redirect without looking at the numbers on both pages.

The only module that writes into your code is the structured data one. Claude Code asks for permission before every write.

8. Common failures

  • Claude cannot find the module.The folder is somewhere other than ~/.claude/skills/, or Claude Code has not been restarted since the copy.
  • lighthouse: command not foundNode is not installed, or the global install failed. Use npx lighthouse.
  • The sitemap comes back empty.The URL points to a sitemap index, not to a sitemap of URLs. Give it the child sitemap.
  • The report says "not testable" everywhere.Google throttles automated requests. Run it again later, or plug in the official Search Console API.
  • The module cannot find the CSV columns.The file was opened then re-saved by a spreadsheet. Start again from the original CSV.
  • The module runs but finds nothing.Check the export period. Three months of data on a new site is not enough to surface a conflict.

You are reading this without the modules

The six modules install in one go, in a single folder, and run locally on your own files.

Your Obsidian vault wired as RAG and a one to one WhatsApp thread come with them.

See the technical SEO agent