The find-skills Skill for Claude Code: What It Does and How to Install It

Claude Code can only use the skills you have installed - unless you give it a skill whose whole job is finding more. That is exactly what find-skills does, and adoption shows it: no other skill comes close.

Key takeaways

• find-skills by vercel-labs is the #1 agent skill by adoption: 2,357,672 installs and 21,669 GitHub stars (skills.sh registry, GitHub), ahead of every other entry in the AI & Agents category.
• It triggers on questions like "how do I do X" or "is there a skill for X", searches the open registry, and recommends installable SKILL.md packages instead of letting the agent improvise.
• It ranks recommendations by install counts and source reputation, preferring battle-tested owners like vercel-labs and anthropics over unknown packages.
• Momentum is still strong: about 224 new installs in the last day alone (skills.sh hot view, 2026-07-06).
• The catalog tracks 12 known copies of find-skills totaling about 1.4k installs (Skillselion catalog, 2026-07-06). Install the vercel-labs original; the copies freeze older revisions.

What is the find-skills skill?

find-skills is a discovery skill: it turns "can Claude do X?" into a registry search that ends with an installed, working capability. Under the hood it checks the skills.sh leaderboard first, then runs npx skills find <query> for keyword search, and verifies quality signals like install counts and source reputation before recommending anything. Think of it as a package manager interface for agent capabilities - the same reflex as reaching for npm instead of writing a library from scratch.

How do you install find-skills?

One command, from your project root:

npx skills add https://github.com/vercel-labs/skills --skill find-skills

It installs into .claude/skills of the current project. Claude Code picks it up automatically; the official Claude Code skills docs cover global installs and how skills load. After installing, ask something like "find a skill for Playwright testing" and the agent searches the ecosystem instead of guessing.

How does find-skills decide what to recommend?

It checks the install-ranked leaderboard before running any search, so well-proven skills surface first: vercel-labs/agent-skills and anthropics/skills entries carry six-figure install counts each (skills.sh). When the leaderboard has no match it falls back to keyword search and weighs source reputation. That bias toward proven sources is the point - it keeps your agent from installing an abandoned one-off. For a human-curated equivalent, start from the ranked shortlist of agent-building skills.

Which copy should you install?

find-skills has been copied widely: 12 known copies with about 1.4k combined installs sit in the catalog next to the original's 2,357,672 (Skillselion catalog, 2026-07-06). The copies freeze whatever revision existed when they were made, while the vercel-labs original was last updated on 2026-06-05 (GitHub). Always install from vercel-labs/skills - the canonical listing links the right repo.

When should you skip find-skills?

Two cases. First, if your agent already has the Skillselion MCP connected, load_skill covers discovery mid-task with no install step at all - the skill loads into context and unloads when you are done. Second, locked-down CI: the skills.sh security audit trail rates find-skills MEDIUM risk because it shells out to npx to install packages. Review that trade-off before enabling it in pipelines that run unattended.

A complete starter stack

• find-skills (2,357,672 installs) - discovery, so the rest of this list stays current
• brainstorming by obra (261,134 installs) - shape the design before any code
• grill-me by mattpocock (460,658 installs) - pressure-test the plan you just shaped
• frontend-design by anthropics (628,875 installs) - distinctive UI output instead of generic defaults

Common pitfalls

• Installing a fork instead of the vercel-labs original - 12 copies exist and none of them get updates.
• Accepting every recommendation: each installed skill adds context weight, so keep the set lean and run npx skills check for updates instead of re-adding.
• Reading install counts as a quality guarantee in niche domains - a low-install skill may just be new; check the repo before you rely on it.

If you install one skill this week, make it the one that installs the others.

Common questions