Documentation menu

Skills

What are skills?

Skills are reusable instruction sets that extend what your agent can do. They’re markdown files with structured instructions, invoked via slash commands — but they’re more powerful than that sounds.

A skill can define an entire multi-step workflow. For example, you could build a /ship skill that spawns a subagent to review your code changes, blocks the push if it finds issues, asks you how to proceed, then commits and pushes when the review passes. Or a /seo-audit skill that crawls your site, checks meta tags, analyzes page speed, and produces a structured report. These are real skills people have built.

Skills aren’t prompt templates. They’re closer to programs — they can use your connected apps, spawn subagents, create files, install packages, and run shell commands. They have the full power of the agent’s computer behind them.

Using skills

There are two ways skills get used:

Manually. Type / in the chat composer to see your installed skills. Select one and the agent follows its instructions. Some skills ask for input, others just run.

Proactively. When your request matches an installed skill, the agent picks it up automatically. If you’ve installed a /frontend-design skill and ask “build me a landing page,” the agent reaches for it without you invoking it explicitly. You don’t have to memorize what’s installed — the agent knows what it has available.

Your skills library

The skills panel in the sidebar shows everything you have installed. You can:

  • Browse your installed skills and see what each one does
  • View the full skill instructions — the exact markdown the agent follows
  • Edit any skill in a Monaco editor. Skills are just text, so tweaking them is instant
  • Search for new skills from the marketplace without leaving the app

Skills live in your workspace’s skills directory as regular files — you can read, edit, or delete them from the file explorer or terminal too.

The marketplace

The skills marketplace is powered by skills.sh, a community skills directory maintained by Vercel with tens of thousands of installs across multiple AI agent platforms.

You can discover and install skills two ways:

Browse yourself. Open the skills panel, search by keyword, preview what a skill does, and install it with one click. The full skill instructions are visible before you install — no blind trust required.

Ask your agent. Tell it “find me a skill for writing marketing copy” or “is there a skill for database migrations?” The agent searches the marketplace, suggests relevant matches, and installs them for you — but only after you confirm. It won’t install anything without your approval.

Installed skills appear in your slash command menu immediately.

Security. skills.sh performs security scanning on listed skills and accepts security reports through Vercel’s security team. That said, skills are just instruction files — they tell the agent what to do, and the agent still operates within your workspace’s security model (credential proxy, network access controls, container isolation). You can always read a skill’s full instructions before installing it.

Creating your own skills

A skill is just a directory with a SKILL.md file. The platform creates a skills directory in your workspace — you’ll see it after installing your first skill or creating one manually.

my-skill/
├── SKILL.md
└── (optional helper files)

The SKILL.md file has YAML frontmatter and markdown instructions:

---
name: my-skill
description: What this skill does (shown in the slash command menu)
---

Instructions for the agent to follow when this skill is invoked.

## Steps
1. First, do this
2. Then do that
3. Finally, check this

The description matters — it’s what appears in the slash command autocomplete and what the agent uses to decide when to proactively invoke the skill. Be specific about when to use it.

Tips for good skills

  • Be specific. Vague instructions produce vague results. Tell the agent exactly what to do, in what order, and what to check.
  • Include decision logic. “If the review passes, proceed to commit. If it fails, present the issues and ask the user.”
  • Reference files. Skills can include helper files alongside SKILL.md — reference docs, templates, checklists. The agent can read them.
  • Leverage the full computer. Skills can install packages, hit APIs, create databases, spawn subagents — anything the agent can do normally, a skill can orchestrate.
  • Test iteratively. Run the skill, see what happens, adjust the instructions. Skills are just text — editing them is instant.