ai.withmb · free guide · keyword SKILLS

The skill that finds skills, with one safety rule

find-skills is a free skill from Vercel that searches a directory of more than a million Claude skills and installs the right one for your task. Here is the setup, the rule that keeps it safe, and how to read a skill page in ten seconds.

#1 on skills.sh 4.4M installs 39.9K GitHub stars MIT license

Setup, one line

Run this in your terminal. It adds the find-skills skill to Claude Code from the official Vercel repo.

npx skills add vercel-labs/skills --skill find-skills -a claude-code

Prefer the browser? Open skills.sh/vercel-labs/skills/find-skills and copy the install command from the page. The repo is github.com/vercel-labs/skills.

Then just tell Claude what you are trying to do, in plain words: "I need to write weekly content for my business", "I want to build a small app", "help me research suppliers". find-skills searches the directory, ranks what fits, and can add it to your setup.

The one rule: check before you install

A skill is a file that tells your AI what to do on your computer. The find-skills file itself offers a fast install flag (-y) that skips confirmation. That is fine for the official skill, dangerous for everything it finds. So paste this into your CLAUDE.md once, and it applies to every skill from then on.

Before installing ANY skill, stop and show me:
1. the author and the original source repo ("Originally from")
2. the GitHub star count (under 100 stars = treat with skepticism)
3. the security audit results (Gen Agent Trust Hub, Socket, Snyk)
4. the full SKILL.md text from the original source
Then wait for my explicit yes. Never install on your own.
Never use -y or any flag that skips my confirmation.

Why each line matters: installs come from CLI telemetry and can be gamed, stars are harder to fake, audits catch known bad behavior, and the SKILL.md is the actual instruction your AI will follow. Reading it takes one minute and it is the whole game.

Read a skills.sh page in 10 seconds

  1. Installs (big number, e.g. 4.4M): popularity only. It is telemetry, so treat it as a hint, not proof.
  2. GitHub stars (e.g. 39.9K): real people vouching for the source repo. Under 100, slow down.
  3. Originally from: the true source repo. Copies of popular skills exist under other names with few stars. Always follow this link home.
  4. The three audits: Gen Agent Trust Hub, Socket, Snyk. PASS is good. WARN means read the SKILL.md twice before saying yes.

Real example from the board: a reddit-automation skill shows 1.4M installs but only 41 stars and a Snyk WARN. Big number, weak trust signals. That is exactly the page where the rule above saves you. Meanwhile find-skills itself: 4.4M installs, 39.9K stars, Trust Hub PASS, Socket PASS, Snyk WARN (read its SKILL.md, it is short and public).

Three prompts to use this week

Find a skill for a job

the everyday use
start here
I need to [write my weekly newsletter / build a simple booking page / research competitors].
Use find-skills to search the directory.
Show me the top 3 with author, stars, audits and SKILL.md.
Do not install anything until I pick one and say yes.

The weekly repeat audit

find work worth handing over
weekly
Here is what I repeated by hand this week: [list 3 to 5 tasks].
For each one, use find-skills to check if a well trusted skill exists.
Rank by trust signals, not install count. Show your evidence.

Clean up what you never use

keep the setup small
monthly
List every skill installed in my setup with its source and last use.
Flag ones I have not used in 30 days or that come from unknown authors.
Propose removals. Wait for my yes before removing anything.

Known sources worth trying first

vercel-labs/skills

home of find-skills, MIT, 39.9K stars
official

anthropics/skills

Anthropic's own official skill repo
official

github.com/anthropics/skills · document and workflow skills from the people who make Claude

Anything else

the directory is open, anyone can publish
check first

That is the point of the rule. Author, stars, audits, SKILL.md, then your yes. Every time.

PART OF WICK ACADEMY - FROM THE @AI.WITHMB SERIES
© 2026 THE WICK FIRM - PRIVACY · TERMS @AI.WITHMB · GUIDES · ACADEMY