← ClaudeAtlas

web-seolisted

Read how a site is doing in Google and on-site, decide what is actually wrong, and fix it: index coverage, impressions and queries, the conversion funnel, and the things that break silently — a stuck deploy, a re-enabled crawler block, a page that lost indexing, a sitemap that never listed a new page. Use when the user asks how the site or an app page is doing in search, why there is no traffic, whether pages are indexed, what to improve, to check the analytics or funnel, after shipping a new page, or for a periodic read ("איך האתר בגוגל", "למה אין תנועה"). It runs a full on-page technical audit and reads the live search and analytics data. It does NOT set up the measurement stack, and does NOT write the product's words — `copy-edit` and the profile own those.
yybd/app-factory · ★ 5 · AI & Automation · score 77
Install: claude install-skill yybd/app-factory
# web-seo — what a site is doing in search, and what to do about it **Where the site is, and what it is made of**, comes from `.claude/seo-site.json` in the site's own repo — hostnames, which directory serves which property, the publish command, and the credentials for the live half. It is optional: without one, the repo is the site. The shape is in [references/site-descriptor.md](references/site-descriptor.md). **A baseline is worth more than any single reading.** Search numbers only mean something against an earlier measurement of the same thing, so the first run of this skill on a site is a baseline, and every later one is compared to it. Keep it where the site keeps its own notes. Work in two passes and in this order: **audit the pages** (`audit.py` — offline, seconds, everything actionable now), then **read Google's response** — a snapshot of Search Console and analytics. **No client for those APIs ships here**; GA4 is read through Google's `analytics-mcp` server when the machine has it (§3), and the rest of the snapshot is whatever export or dashboard the site already has. It can only report what the pages already earned. Diagnosing from the response alone is how a markup problem gets mistaken for a demand problem. It does **not** set up measurement — that exists — and it does not write the product's words: those come from `<slug>/profile.md` and are edited by `copy-edit`. It decides *what* needs saying and where; that skill decides how it reads. This file is the m