Lab component 08

Run the audits, not just read about them

Paste the source of any page and pick a check. These are the actual audit routines from the Claude SEO Skills library, ported to run in your browser. Nothing is uploaded, and you can point the runner at this very page.

Real audit logic Runs entirely in your browser Nothing uploaded

Skill runner · paste source, pick a check

Skills

Load a sample or paste your own source, then run a skill.

These are the same checks that run inside the Claude SEO Skills library, ported to run client side against a parsed document. Nothing is sent to a server, which is also the reason it works on a pasted source rather than a URL: a browser cannot fetch an arbitrary origin without that origin's permission.

Why a skill and not a script

The checks themselves are not hard. Encoding when to run them, what a failure actually means, and how to phrase the fix is the part that takes the time and the part that transfers.

01

Severity has to be earned

A missing alt on a spacer image and a missing H1 are not the same finding. Flat checklists that treat them equally are how audits become unreadable.

02

Findings need a diagnosis

Reporting that a title is 78 characters is data. Reporting that it will truncate at roughly 60 and burying the differentiator is a finding.

03

Every check ships with an eval

Each skill in the library has test fixtures with known defects. If a change to the logic stops catching a fixture, that is a regression, caught before a client sees it.

The full library is at Claude SEO Skills.