aleclindz/seo-skill-bench — explained in plain English
Analysis updated 2026-05-18
Compare real world performance of different Claude Code SEO skills before choosing one.
Test your own SEO skill against fixture sites with a hidden answer key.
See which SEO skills avoid hallucinating unnecessary fixes on already correct pages.
Understand how a deterministic, answer key based benchmark can be designed and scored.
| aleclindz/seo-skill-bench | peng-zhihui/cheese-uestc | karpathy/notpygamejs | |
|---|---|---|---|
| Stars | 110 | 110 | 109 |
| Language | JavaScript | JavaScript | JavaScript |
| Last pushed | — | 2017-01-13 | 2023-10-17 |
| Maintenance | — | Dormant | Dormant |
| Setup difficulty | moderate | moderate | easy |
| Complexity | 3/5 | 2/5 | 2/5 |
| Audience | developer | general | developer |
Figures from each repo's GitHub metadata at analysis time.
Running the benchmark requires Claude Code access and headless execution against fixture sites, reading the leaderboard needs no setup at all.
seo-skill-bench is a benchmark that compares different SEO focused skills built for Claude Code and similar coding agents. Instead of trusting a skill's description or its star count on GitHub, this project actually installs each skill in a real agent session and points it at test websites that have known problems planted inside them on purpose. Each test site is a small but realistic setup: a live static website, a slightly out of date source code repository, and a sample Google Search Console data export, along with a hidden answer key listing exactly what is wrong with the site and which recommendations would be traps, meaning things that sound like fixes but are not actually needed. Because the correct answers are known ahead of time, most scoring is done automatically by comparing what the skill found and recommended against that answer key, rather than relying on a judge's opinion. The project follows a few strict rules to keep the comparison fair: every skill gets the exact same short prompt with no hints about what to look for, each skill runs in its own isolated workspace so results cannot leak between entrants, and the scoring rubric is written down and frozen before any runs happen. A small part of the score, judging the overall quality of strategy suggested, still needs a human style judgment call, so that part is scored by an anonymized panel that does not know which skill produced which answer. The README includes a public leaderboard showing how several real SEO skills scored across categories like defect detection, avoiding traps, and overall judgment, along with notes that the maintainer's own skill is included as an entrant and has previously placed last in one of these runs. Full scoring weights and methodology are documented separately in the repository.
An open benchmark that runs Claude Code SEO skills for real against test websites with planted, known defects and scores them against a hidden answer key.
Mainly JavaScript. The stack also includes JavaScript, Node.js, Claude Code.
No license information is stated in the visible README, so reuse and redistribution terms are unclear.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.