SEO Repair — Development skill for Claude Code

seo-repair

[](https://github.com/SurendharVr/seo-repair/actions/workflows/tests.yml) [](LICENSE)

A [Claude Code](https://claude.com/claude-code) skill that treats SEO as a **repair system with a dependency order**, not a checklist.

It crawls your site's source, produces evidence-backed findings in that order, proposes fixes, and re-verifies after they land. It proposes every edit before applying it, and it will not print a number it did not measure.

Crawl → Index → Architecture → Content → On-Page → Performance → Schema → Authority → Monitor

Work done out of order is wasted. Optimising a title on a page Google cannot index changes nothing. Writing schema for a page about to be merged away is thrown out with the page. Building links to either is the most expensive mistake of the three. Three **gates** make that ordering unskippable rather than advisory.

---

Two things that make it different

**It refuses to invent numbers.** Search volume, keyword difficulty, backlink counts and field Core Web Vitals cannot be observed from a codebase. They are never estimated — they become a handoff list naming the exact Search Console or Ahrefs report to open. Every claim in a report is tagged `MEASURED` (observed directly), `INFERRED` (reasoned from observation, stated as such), or `HANDOFF` (not observable here). A local crawl proves a page is *crawlable*; it never proves the page is *indexed*, and the report will not blur those.

The same standard applies to what it writes: no invented author bios, testimonials, ratings or trust claims. A tool that refuses to guess at keyword volume has no business guessing at a person's credentials.

**It checks whether your structured data is honest.** Most validators check that schema parses and has its required properties. This one also checks whether