Best Claude Skills by Topic, Tested

Skills grouped by the work you want done. Each was tested twice on the same task, once with the skill loaded and once without it, and each line below says how many helped.

A skill tested in more than one run has a result from each, and one kind of work is scored two ways, so a topic can hold more results than skills. They are never added together.

  • Writing Skills

    6 of 6 results helped clearly on at least one model.

    5 results are level at the top, and nothing on this site separates them.

  • Accessibility

    1 of 6 results helped clearly on at least one model.

    Best of them: skills/accessibility-audit

    Helped on Claude Haiku 4.5; no reliable difference on Gemini 3.1 Flash Lite and GPT-5 mini. Measured against no skill loaded.

  • Product Management

    2 of 10 results helped clearly on at least one model.

    Best of them: skills/pm-spec-writing

    Helped clearly on Claude Haiku 4.5, Gemini 3.1 Flash Lite and GPT-5 mini. Measured against no skill loaded.

  • SEO

    1 of 5 results helped clearly on at least one model.

    Best of them: skills/seo-onpage

    Helped on Gemini 3.1 Flash Lite; no reliable difference on Claude Haiku 4.5 and GPT-5 mini. Measured against no skill loaded.

  • Brand Voice

    2 of 2 results helped clearly on at least one model.

    Best of them: skills/brand-voice

    Helped clearly on Claude Haiku 4.5, Gemini 3.1 Flash Lite and GPT-5 mini. Measured against no skill loaded.

  • Code Review

    0 of 8 results helped clearly on at least one model.

    Nothing tested helped on any model.

Set up, not yet run

  • Test writing: set up for testing, not yet run.

Within a topic, the order is how many models the skill helped on, which is a count and never the size of the change. Nothing here ranks one topic against another: two topics are two answer keys, and there is no shared scale between them. Every skill, by kind of work.