Prototype · 2026 · Claude agents
Next.js SEO skill for Claude Code
A custom Claude skill that turns page-specific Next.js SEO requirements into reusable guidance, then measures the result against a five-case eval set.
Problem
Generic SEO advice often misses route-specific schema, canonical, pagination, data-source, and Cloudflare caching requirements.
My role
Skill design, page-type reference material, evaluation cases, grading criteria, and benchmark review
Stack
How the work moves
- Step 01Page requestRoute, content type, and stack constraints
- Step 02Skill routingRelevant page-type guidance only
- Step 03ImplementationMetadata, schema, content, and cache
- Step 04Eval setExpectation-based comparison
Why a skill instead of a long prompt
SEO implementation changes with the page type. A homepage, review, news post, pricing page, and paginated product listing need different metadata, structured data, content semantics, and caching decisions. The custom skill packages those differences as focused references for a Next.js stack using Supabase and Cloudflare Pages.
The routing instructions load only the relevant page-type guidance. That keeps the working context smaller while still covering concrete implementation details such as canonical URLs, JSON-LD types, indexation rules, semantic markup, and Cloudflare cache paths.
Evaluation method
The skill was refined against five cases: a review article, pricing page, news post, homepage, and product listing. Each case has explicit expectations, and the benchmark stores the generated output, grading evidence, timing, and token use for runs with and without the skill.
In the recorded single-run benchmark, the skill-assisted outputs passed 34 of 34 expectations across the five cases. The comparison outputs without the skill passed 20 of 34. The largest gap appeared on the product listing case, where the skill run handled canonicalization, pagination indexation, nested structured data, Supabase pagination, and Cloudflare caching as specified by the eval.
What the result does and does not show
The artifact supports a narrow conclusion: for this five-case set and recorded run, the skill improved adherence to the stack-specific requirements encoded by the graders. It is not a claim about search rankings, traffic growth, or performance across every model and website.
The useful engineering pattern is the combination of reusable instructions and executable evaluation. The skill captures the intended approach; the eval set makes regressions and missing details visible.