Agent Skills With skills.sh: Give Your Coding Agent Real Procedures

If you use AI coding agents daily, you have probably felt this gap.

Your agent can write code, but it does not always follow the same reliable procedure you would. It might skip a check, miss a step, or take a path that looks reasonable and still wastes time.

Skills are a practical way to close that gap.

What Are Skills?

Skills are reusable capabilities for AI agents. They package procedural knowledge that helps an agent complete a task with a consistent playbook.

You can think of a skill as a small, focused set of instructions your agent can pull in when it needs to do a specific job.

A good skill can include:

1. code generation patterns
2. domain or framework expertise
3. tool integrations
4. checklists and workflows

Getting Started: Install the Skills CLI

You can install skills with the skills CLI.

Run this:

pnpm dlx skills add vercel-labs/agent-skills

That command downloads the skill and configures it for your agent.

Skills work with popular AI coding agents including Claude Code, Cursor, Windsurf, and others. Compatibility depends on the specific skill, so it is worth checking the skill README.

How Skills Are Ranked

The skills leaderboard on skills.sh is ranked using anonymous telemetry from the CLI.

When people install skills, aggregated installation counts help surface the most useful ones.

What is collected:

• the skill name
• the skill files
• a timestamp

What is not collected:

• personal information
• device identifiers
• usage patterns inside your editor

If you want to opt out, set this environment variable:

DISABLE_TELEMETRY=1

Browse Skills

If you are not sure what to install, start on the skills homepage and browse the leaderboard. It is the fastest way to discover what other teams are using.

Security and Trust

Skills are hosted in GitHub repositories, so the same rules apply as any dependency.

Vercel runs routine audits to assess skills for malicious content, but they do not guarantee the quality or security of every skill listed on the site.

If you find an issue, you can report it at security.vercel.com.

For day to day use, the best practice is simple:

• read the repository
• scan the skill files
• install only what you understand

The Reality

Skills do not make an agent smarter by magic. They make it more consistent.

If you have a workflow you repeat every week, it is a perfect candidate for a skill. The goal is to remove variance in execution so you spend time on decisions instead of process.

Try It Yourself

If you want to add skills today:

1. Visit skills.sh and find a skill you would actually use
2. Install it with npx skills add <owner>/<skill-name>
3. Ask your agent to use it on a real task and compare the result to your normal flow

When the procedure is explicit, the output gets better.