Verdict: If you are installing one Claude skill today, install find-skills — it sits at #1 on the skills.sh leaderboard with 3.4M installs (extracted 2026-09-13), roughly three times the runner-up, because it is the discovery layer everything else routes through. If your agent writes code, grill-me is the highest-value single skill at 1.1M installs; it interrogates you about requirements before a line of code is generated. If you build interfaces, frontend-design is the best official Anthropic option, sitting #6 with 881.3K installs.
TL;DR
- find-skills (vercel-labs/skills) is #1 at 3.4M installs — install it first, it finds the rest for you.
- grill-me (mattpocock/skills) is #2 at 1.1M installs and the best organic pick for coding agents.
- frontend-design (anthropics/skills) is #6 at 881.3K installs and Anthropic's only top-10 slot.
- The top ten come from just three authors: on the live skills.sh board Matt Pocock holds seven slots, Vercel Labs two, Anthropic one.
- Install counts partly measure onboarding funnels rather than merit — bundlers inflate their own numbers.
- Last verified: 2026-09-13.
The top of the leaderboard at a glance
Every install figure below is from the skills.sh leaderboard, extracted 2026-09-13.
| Skill | Author | Installs | Job | Install it if |
|---|---|---|---|---|
| find-skills | Vercel Labs | 3.4M (skills.sh) | finds and installs other skills mid-task | always — first |
| grill-me | Matt Pocock | 1.1M (skills.sh) | interrogates requirements before building | your agent writes code |
| frontend-design | Anthropic | 881.3K (skills.sh) | official UI design-quality skill | you ship interfaces |
| agent-browser | Vercel Labs | 840.6K (skills.sh) | browser automation for the agent | your agent drives the web |
| vercel-react-best-practices | Vercel Labs | 708.5K (skills.sh) | React coding standards | you ship React |
What are the best Claude skills right now, and which should you install first?
Order matters more than choice. The best Claude skills to install, in sequence:
-
find-skills — the package-manager-shaped entry point. It teaches Claude Code to search the open skills registry mid-task and recommend packages by install count and reputation (Skillselion). Install with
npx skills add https://github.com/vercel-labs/skills --skill find-skills, which drops it into.claude/skills. - grill-me — a requirements interrogation skill. Before your agent starts building, it asks the questions you skipped. The value is negative work avoided: builds that head in the wrong direction get caught in the first minute rather than the fortieth.
- frontend-design — Anthropic's own UI skill, useful when you want output that looks deliberate rather than default. If that is your main use case, our walkthrough of Claude's design skills for building beautiful websites goes deeper on what it actually produces.
Everything after those three is situational: vercel-react-best-practices if you ship React, agent-browser if your agent drives a browser (both in the table below).
Why does find-skills win when it does not do the interesting work?
Because the leaderboard rewards whoever owns onboarding, not whoever writes the best skill. find-skills is infrastructure: install it and it installs others, so its count aggregates the ecosystem's growth rather than its own usefulness.
The same effect shows up elsewhere. setup-matt-pocock-skills sits at #8 with 823.5K installs and is a per-repo configuration skill that scaffolds the assumptions his other skills depend on, as Skillselion describes it. The tracker o-mega.ai makes the point plainly: find-skills and setup-matt-pocock-skills are bundlers, so their counts measure onboarding funnels rather than adoption on merit, which is why o-mega.ai calls grill-me the top organic skill after it overtook frontend-design (756.3K against 742.3K on their snapshot, about four weeks of data).
Read the leaderboard as two separate charts, then: one for distribution mechanics, one for skills people chose deliberately.
Why is grill-me the best pick if your agent writes code?
Because the expensive failure in agentic coding is not bad syntax, it is confident work on the wrong specification. grill-me inverts the default order: it questions you first, then lets the agent build.
The growth data supports it being more than a curiosity. Skillselion records find-skills adding +224 installs a day and grill-me +188 a day, with tdd at +148 and grill-with-docs at +143 — the interrogation-and-testing cluster is where daily momentum sits. A sibling skill, grilling, is separately at 687.0K installs.
The tradeoff is real. grill-me adds friction to every task, including the trivial ones. For a one-line fix the questioning is overhead you will resent. Keep it scoped to features, not chores.
Why does Anthropic hold only one of the top ten slots?
Anthropic defined the skill format and publishes the official repo at github.com/anthropics/skills, where each skill is a folder containing a SKILL.md with name and description frontmatter. Its document skills (docx, pdf, pptx, xlsx) power Claude's own document handling and are source-available rather than open source; most others carry Apache 2.0.
Yet in the top ten it holds a single position, frontend-design (skills.sh). Skillselion's snapshot of the same board counted the same three-author concentration as 5-4-1 before Matt Pocock's newer skills climbed. Its other entries sit further down: skill-creator at 379.5K, pdf at 195.0K, docx at 186.9K, xlsx at 167.4K, webapp-testing at 155.5K.
Aggregate publishing tells a different story from individual ranks. ByteDance's Lark/Feishu bundle is the largest aggregate publisher at 16.4M total installs across its skills, with lark-doc alone at 686.5K; microsoft/azure-skills aggregates 8.7M. Anthropic set the standard and platform vendors took the volume. Anthropic's own engineering explainer on agent skills is still the clearest description of what the format is for.
Why does the skills layer matter more than which model you pick?
Because frontier models are converging on the constraint-following tasks that used to separate them, which pushes differentiation up into workflow.
Our own bench: across three trials each on an identical seven-constraint article-planning task, Gemini 3.8 Flash (High) and Claude Opus 4.6 (Thinking) both scored 17 of 17 on machine-checked constraint adherence. Median wall time was 23 seconds for Gemini against 67 seconds for Opus (first-party measurement, n=6, measured 2026-09-13).
If both models follow instructions equally well, the leverage is in what instructions they get — which is exactly what a skill is. That reframes tooling choices too: see our Antigravity vs Cursor vs Claude Code comparison, Hermes bot mode vs standalone Claude, the Claude Code pricing breakdown and a less conventional task imagination skill.
How big is the skills catalogue, and can you trust the count?
No single authoritative number exists — any figure is one tracker's snapshot with its own inclusion rules. Skillselion indexes 58,979 Claude Code skills; claudeskills.info counts 42,238 across 61 task-specific leaderboards. Larger figures circulating informally do not survive contact with primary sources, so treat catalogue size as marketing and install trend as the signal. The MIT-licensed vercel-labs/skills CLI supports Claude Code, OpenCode, Codex, Cursor and many more clients, so skills are becoming portable rather than Claude-only.
The verdict
Install find-skills first because discovery compounds. Add grill-me if your agent writes code, and accept the friction on small tasks as the price of avoiding wrong-direction builds. Add frontend-design if you ship interfaces. Ignore install counts as a quality signal beyond the top few, since bundlers distort them, and prefer daily growth plus the number of independent trackers listing a skill.
FAQ
Q: Which Claude skill has the most installs?
A: find-skills from vercel-labs/skills, at 3.4M installs on the skills.sh leaderboard as extracted on 2026-09-13, ahead of grill-me at 1.1M.
Q: How do I install find-skills?
A: Run npx skills add https://github.com/vercel-labs/skills --skill find-skills, which installs it into your project's .claude/skills directory, per Skillselion's install guide.
Q: Is grill-me better than frontend-design?
A: For coding work, yes. o-mega.ai recorded grill-me overtaking frontend-design as the most-installed non-bundled skill, 756.3K against 742.3K on their snapshot. For interface work, frontend-design is still the better fit.
Q: Are Anthropic's official skills open source?
A: Partly. The anthropics/skills repo publishes most skills under Apache 2.0, but the document skills covering docx, pdf, pptx and xlsx are source-available rather than open source.
Q: Do Claude skills work in other tools?
A: Yes. The vercel-labs/skills CLI states support for Claude Code, OpenCode, Codex, Cursor and further clients, and is MIT licensed.
Last verified: 2026-09-13. Corrections: none to date.
Top comments (0)