📊 Catalog size
Before creating a skill, it helps to look at the size of the territory. In the skills.sh harvest, the catalog has 39.366 skills, from 5.075 source repos, totaling 53.9 million installs, organized into 16 topic groups. It’s large enough to already have almost everything — and specific enough to still have gaps.
💡 Averages are misleading
53.9M ÷ 39.366 would be ~1.370 installs per skill “on average.” But average is the worst measure here—the distribution is so skewed that almost no skill is close to it. That’s exactly what the next topic shows.
📉 The power law
The distribution of installs follows a power law: many at the bottom, very few at the top. The vast majority of skills live in the long tail, with almost no installs, while a tiny handful attract a huge audience.
✗ The long tail (most)
- ✗60,2% have fewer than 100 installs
- ✗30,8% range from 100 to 999
- ✗91% of the catalog has never exceeded 1,000
- ✗Publishing isn’t the same as being used
✓ The top (the exception)
- ✓7,7% between 1k and 9.9k installs
- ✓1,1% between 10k and 99k
- ✓Only 0.3% (131 skills) exceed 100k
- ✓Getting there requires solving a real pain point
💡 Honest reading
This isn't discouraging—it provides direction. Being in the long tail is the default state. The right question isn't "how do I reach the top 0.3%" but "what specific pain point does my skill solve better than any existing alternative?"
🎯 Focus
The power law has a dramatic consequence at the edge: the The 100 most installed skills account for 43,7% of all installs. Almost half of 53.9 million are in 100 names — out of nearly 40 thousand. It’s a “winner takes almost all” effect.
The absolute top of the catalog
the focus in one line:
100 skills (0,25% do catálogo) = 43,7% dos installs
~39.266 skills (99,75%) = 56,3% restantes
💡 What to take away
Competing head-on with find-skills or frontend-design is nearly impossible—they have brand recognition, distribution, and momentum. The smart move is not to duel at the top, but to find where the top hasn’t reached.
🗂️ The 16 Topic Groups
The catalog is divided into 16 groups. Comparing skills count with installs reveals what matters: some groups have many offerings; others have few skills capturing a lot of demand. See the main ones.
| Group | Skills | Installs |
|---|---|---|
| 🤖 AI/LLM/Agents | 12.693 | 12,9M |
| 🎨 Web & Frontend | 4.694 | 4,5M |
| ☁️ DevOps & Cloud | 907 | 8,3M |
| 🛠️ Dev Tooling & Code | 796 | 4,4M |
| 📈 Marketing & SEO | 582 | 2,3M |
| 🎬 Media & Creative | 550 | 1,85M |
| 🔌 Backend & APIs | 2.685 | 1,38M |
| 🗄️ Data & Databases | 1.385 | 0,79M |
| 🧪 Testing & QA | 1.339 | 0,60M |
📈 High demand for a skill
DevOps & Cloud: only 907 skills, but 8.3M installs — ~9.150 per skill. A few names (the microsoft azure family) capture almost everything.
Marketing & SEO e Dev Tooling also have a good install-to-skill ratio.
📉 High supply, fierce competition
AI/LLM/Agents: 12.693 skills competing for 12.9M — ~1.016 per skill. It’s the most crowded group in the catalog.
Backend & APIs has 2.685 skills for just 1,38M installs — lots of people, not much pie.
💡 Don’t look only at size
The volume of skills in a group measures its competition; installs per skill measure need met. The less obvious groups (Office, Finance, Productivity) have few offerings — an opportunity for anyone who does it well.
🏢 Source concentration
Skills aren’t the only thing that concentrates— source too. Despite the 5.075 repos, a few account for a disproportionate share of installations. The same names keep appearing at the top of almost every group.
vercel-labs — dominates Web & Agents
find-skills (1.8M), vercel-react-best-practices (443k), web-design-guidelines (358k), agent-browser (332k), vercel-composition-patterns (195k). It leads two of the largest groups on its own.
microsoft — owner of DevOps/Cloud
The azure-skills family (microsoft-foundry, azure-ai, azure-deploy, azure-diagnostics, azure-prepare) gets ~358–360k each. That's why DevOps has such a high install/skill ratio.
anthropics & larksuite — reference and strong niches
anthropics: frontend-design (488k), skill-creator (246k). larksuite/cli: lark-slides (139k) dominates Office. Each rules its own territory.
💡 Brand + distribution matter
A great skill in an unfamiliar repo starts at a disadvantage compared with a familiar name. That reinforces the lesson: for the community, the path is niche, not direct competition with vercel-labs or microsoft.
🧭 What the map teaches
Bringing it all together, the skills.sh landscape offers three practical lessons that guide the rest of the course. They’re not opinions—they come straight from the numbers.
Discover what already exists before creating
With 39,366 skills, there’s a good chance your idea already exists. Search skills.sh first (npx skills find ...). If there’s already a good one, install it. Reinventing the wheel only fattens the long tail.
Target empty niches
Don’t compete in AI/Agents (12.693 crowded skills) or with find-skills. Look for something specific: your domain, your tool, your team convention—where the big names haven’t reached.
Play on the right side of the curve
The right goal isn’t “having lots of installs” — it’s solving a real pain point better than the alternatives. Do that and the install count will follow. The quality that sets the top apart from the 60.2% is the subject of Track 2.
mental checklist before creating:
[ ] busquei no skills.sh? já existe algo bom? [ ] meu grupo é lotado (muita oferta) ou vazio? [ ] que dor específica eu resolvo melhor? [ ] minha description deixa claro o QUE e o QUANDO?
💡 The bridge to the rest of the course
Trail 1 gave you the lay of the land: what a skill is and how the market works. From here, the course gets hands-on — quality (T2), anatomy (T3), the creation loop (T4), and what to consider when publishing (T5).
✅ Module Summary
Next:
1.3 — ⭐ The Best Ones to Start With (Starter Pack). The canonical skills everyone should install first — and the recommended minimum kit.