// the find
AgriciDaniel/claude-seo
Universal SEO skill for Claude Code. 25 sub-skills + 18 sub-agents covering technical SEO, E-E-A-T, schema, GEO/AEO, backlinks, local SEO, maps intelligence, semantic clustering, e-commerce SEO, international SEO, Google APIs, and PDF/Excel reporting. Optional DataForSEO, Firecrawl, and Banana extensions.
A Claude Code plugin that runs 25 SEO sub-skills and 18 specialist agents in parallel — technical audits, schema validation, E-E-A-T scoring, AI-search optimization, local SEO, and more. It's aimed at SEO consultants and agencies who want LLM-driven audit synthesis inside their Claude Code workflow rather than switching to a separate tool. Not a crawler replacement for Screaming Frog; more of an analysis and recommendation layer on top of what you can fetch.
The falsifiability requirement on every recommendation is genuinely unusual — each finding carries an explicit 'how would we know this failed?' check and a leading indicator, which forces the output to be actionable rather than just observational. The 4-tier Google API credential system is well designed: you get real value at Tier 0 (PageSpeed, CrUX 25-week history) without being forced to set up OAuth before you can try anything. The schema deprecation tracking is current and specific — FAQPage removed May 2026, ClaimReview and CourseInfo retired June 2025, with replacement guidance per type. The parallel agent dispatch (up to 15 simultaneously) means a full site audit finishes in minutes rather than sequential tool runs.
The whole thing is markdown files and Python scripts wired together through Claude's agentic loop — there's no deterministic execution guarantee, so audit consistency across runs depends entirely on how Claude interprets the SKILL.md orchestration instructions that day. Star count (8793) versus fork count (1276) is suspicious for a niche Claude Code plugin that's under a year old; growth charts in the README are from the author's own site, which is a self-referential proof point. The SPA/hydration limitation is real and documented honestly, but 'run seo-visual and compare manually' is not a solution, it's a workaround that requires you already know which pages have the problem. Single maintainer with no disclosed testing infrastructure beyond 326 passing tests on skill manifests — there's no indication of what happens when Google's APIs change their response shapes or when Claude Code updates break the agent dispatch model.