Skip to content
alyn

Methodology

The Alyn Score: how we rank AI skills.

One score from 1 to 99 for every skill. Popular is not enough: the score also checks how well the skill is made, whether it is rising and whether anyone still maintains it.

45%

Popularity

GitHub stars and installs, on a log scale so giants do not flatten everything else.

25%

Quality

Six checks on the SKILL.md itself: does it say when to use it, is it deep enough, does it ship resources?

15%

Momentum

New stars in the last 7 days. Rising skills get noticed before they are famous.

15%

Freshness

Last push to the repository. Abandoned skills slowly sink.

Quality, 25%

Six checks on every SKILL.md

These are the things that decide whether an agent actually uses a skill and follows it well. Each skill page shows which checks it passes.

  1. 01

    Says when to use it

    The description tells the agent when to load the skill (“Use when…”). Without it, agents rarely trigger the skill.

  2. 02

    Clear description

    80 to 1,024 characters: long enough to match tasks, short enough for the agent's skill list.

  3. 03

    Real instructions

    1,500 to 40,000 characters of instructions. Not a stub, not a book.

  4. 04

    Ships resources

    Scripts, references or templates next to the SKILL.md.

  5. 05

    Open license

    A license on the skill or the repository, so you may use it.

  6. 06

    Documented

    A README with a cover image or a project homepage.

Weekly top 10

What is hot this week

The weekly list uses a different mix: 50% new stars this week, 35% real usage on Alyn (CLI installs, MCP loads, copies, votes) and 15% the Alyn Score. One skill per pack.

See this week's top 10
Is the Alyn Score just popularity?

No. Popularity is 45%. A quarter of the score comes from quality checks on the SKILL.md itself, and 30% from momentum and freshness, so a well-made new skill can outrank an old famous one.

Why log scale?

Stars range from a few hundred to hundreds of thousands. On a log scale, going from 1K to 10K stars counts as much as 10K to 100K, so the top repositories do not flatten everyone else.

Can someone game it?

Stars and installs come from GitHub and the skills CLI. Alyn's own usage signals (CLI installs, MCP loads, votes) count one install per person per hour, votes need an account, and bots are filtered out.

Do you test whether skills really work?

Independent studies show skills raise pass rates a lot on average (SkillsBench: 24.3% to 40.6%). Alyn checks the structure that makes a skill work in practice and ranks by what people actually keep using. Per-skill benchmark runs are planned for a small set of popular skills.

How often does it update?

Every night. Stars, installs, new skills and momentum refresh with the nightly sync; live numbers on skill pages update on every visit.