A small, curated collection of high-quality skills — and an experiment in measuring what makes a skill good.
Most "skills" in the wild are a generic paragraph with a nice filename; you could regenerate them from the title alone. Loooom is the opposite bet: a tight set of skills held to a deliberately high bar, each scored against a rubric so quality is visible instead of vibes. The web app lives at loooom.xyz; this is the main repo.
More on the why: Loooom: Curated Skills for People Who Don't Code.
npx loooom add mager/beginner-japaneseCreates .claude/skills/beginner-japanese/SKILL.md in your project. Claude Code picks it up automatically — and because the skills are agent-agnostic, you can copy a SKILL.md into any AI just as well.
Every skill is scored before it ships — that's the whole point of the project. Scoring runs on a free model (Groq), so grading the catalog costs nothing.
npm install -g promptfoo
# Free key at console.groq.com
export GROQ_API_KEY=your-key-here./bin/eval-all.sh # all skills
npx promptfoo eval --config plugins/beginner-japanese/promptfooconfig.yaml # one skill
npx promptfoo view # results in the browserScores write to eval-scores.json and show on loooom.xyz/browse.
Evals run on every push that touches plugins/*/promptfooconfig.yaml or plugins/*/skills/**, and nightly at 02:00 UTC. Results are committed back to eval-scores.json by the eval bot. Requires GROQ_API_KEY set as a GitHub Actions secret.
plugins/
└── my-plugin/
├── promptfooconfig.yaml # quality tests
└── skills/
└── my-plugin/
└── SKILL.md # the skill itself
| Command | Description |
|---|---|
npx loooom add <author>/<plugin> |
Install a skill into .claude/skills/ |
npx loooom list |
List installed skills |
npx loooom help |
Show help |
- Node.js 18+
MIT