Most Claude Code stacks have a content-strategy skill (keyword research, search intent, meta description writing) and a launch-readiness skill (OG metadata, sitemap, robots). Between those two layers there is a gap nobody owns: technical SEO. Schema markup that actually matches your content type. The GEO and AEO layer that decides whether ChatGPT and Perplexity will ever cite your URL. Backlink hygiene, local SEO if you have a physical surface, OG image generation that uses your real brand tokens. None of that lives in the strategy skill. None of it lives in launch-readiness. It lives in nothing.
AgriciDaniel released claude-seo v1.9.0 on April 15, 2026. Nineteen sub-skills, twelve subagents, three extensions (DataForSEO, Firecrawl, Banana). It sits precisely in the gap. The brief this week tagged it [INSTALL NOW] because it is the rare case where a public skill closes a real hole without overlapping anything we already own.
The install
Two lines of bash. The first adds the marketplace from the GitHub repo. The second installs the plugin from that marketplace.
claude plugin marketplace add AgriciDaniel/claude-seo
claude plugin install claude-seo@agricidaniel-seo
The marketplace add takes about ten seconds (it clones the repo, validates the manifest, caches it locally). The install is instant. After this, every Claude Code session in any project on your machine has the SEO sub-skills available. They are invoked by description matching, the same way the rest of the skill ecosystem works.
What you get
The most immediately useful sub-skill is the technical SEO audit. Run it against any live URL and it returns a structured report: meta tags coverage, schema markup completeness, page-level performance flags, internal linking density, robots / sitemap coverage, and a score across six dimensions. The output is opinionated and does not hedge, which is why we picked it over the Next.js-specific alternatives that produce nicer scoring tables but vaguer recommendations.
The Banana extension adds command-driven OG image generation. Point it at a page and a brand prompt and it produces a 1200x630 PNG you can drop into your app/opengraph-image.tsx. We were already generating OG images by hand for every Defrag surface; the Banana flow makes it a one-line slash command per page going forward. This alone earns the install.
The GEO and AEO layers (Generative Engine Optimization, Answer Engine Optimization) are newer additions and worth knowing about even if you do not use them today. As more traffic shifts from Google search results to ChatGPT and Perplexity citations, the criteria for ranking in those answer engines diverges from classical SEO. The sub-skills here cover the structured-data patterns that make a URL more likely to be quoted by an LLM with attribution. Test them on one piece of content and watch what changes.
What to test first
Run the technical audit against your most-visited page. The first audit is uncomfortable because it surfaces every gap you knew was there and several you did not. The second audit, after you have fixed the obvious flags, is when the skill earns its place. From there it becomes part of the launch-readiness pre-flight: every new page gets the audit before it ships, and the audit's blocker-tier flags gate the merge.
The Banana extension is the second thing to try. Generate one OG image for a page that currently has none. The output quality is a function of how good your brand prompt is; invest fifteen minutes writing the prompt for your project's locked palette and type and the rest of the surface generates from that single template forever.
Skip the local SEO sub-skill unless you actually have a physical surface to rank for. Skip the backlink hygiene flow unless you are at the scale where backlinks compound. Both are real capabilities and both are wasted attention for most readers of this publication.
Why this earned the [INSTALL NOW] tag
Three filters: it is allowlist-clear (over 1k stars, MIT licensed, recent commit), it does not overlap anything we already own (different layer than seo-strategy, different scope than launch-readiness), and the install cost is two minutes for a capability we would otherwise build from scratch over the next month. Most weeks the brief surfaces nothing that passes all three filters at once. This one did.