Vercel Skills 101

The Package Manager Your AI Agents Were Missing

Agent Skills: The New Standard

If you’ve been following the AI space lately, you’ve likely heard about Agent Skills. Pioneered by Anthropic, this open standard allows us to package specialized instructions, tools, and scripts into a modular format.

It’s a brilliant architectural shift: instead of bloating an AI’s context with a 10,000-line system prompt, you provide “dormant” manuals. The agent only reads and “activates” them when it actually needs to perform a specific task.

In my previous post, Demystifying Coding Agents , I took a deep dive into the problems Skills solve and why they are the natural evolution of context management for coding agents.

The Manual Installation Problem

But here’s the catch: A standard is not a manager.

While tools like Claude Code have built-in ways to fetch skills, the rest of the ecosystem is a fragmented mess. If you’re a developer jumping between Cursor , Windsurf , Claude Code , and OpenClaw ( or just using one that doesn’t have built in way to install Skills), you’re currently living in the “Manual Installation Era.”

To give your agent a new capability today, you usually have to:

1. Hunt down a GitHub repo or a SKILL.md from anywhere.

2. Download the directory or file manually.

3. Manually paste it into a specific hidden folder, like .cursor/skills or .windsurf/skills

We’ve essentially regressed to the days before npm or pip, where “installing” a library meant dragging a ZIP file into your project and praying your environment variables were correct. It’s manual, it doesn’t scale, and it’s a massive barrier to building truly portable AI agents.

A Package Manager for Skills

Vercel Skills solves exactly this. It isn’t trying to create a new standard, it is the Package Manager for the existing one.

Think of it as the “NPM for Agents.” It provides a CLI and a central registry (skills.sh) to find, install, and manage these open-standard skills across every AI agent in your workflow.

In the first installment of My Digital Arsenal , we discussed how tools like uv and pip revolutionized Python development, Vercel’s Skills CLI brings that same level of sanity to the AI ecosystem.

How to Use Vercel Skills

The beauty of Vercel Skills is that there is nothing to install. It lives in the cloud and runs on your machine via npx, mirroring the “zero-config” philosophy of modern dev tools.

1. Discovery: Finding Your Edge

Instead of scouring GitHub for the right instructions, you can search the registry directly from your terminal — npx skills find

This triggers an interactive, searchable list. If you’re looking for something specific, you can pass a keyword (npx skills find react for example)

2. Installation: Adding Powers to Your Agents

Once you’ve found a skill (e.g., vercel-labs/agent-skills), you can download it with the command npx skills add vercel-labs/agent-skills

This starts an interactive process where you choose the exact skills you want, select which AI agents you are using, and decide if you want the skill at the global or repo level.

How it works under the hood: Vercel Skills clones the repo and then download the skills into a central .agents/skills directory. For every agent you use, it copies or symlinks the relevant files into the appropriate folders (like .cursor/skills or .windsurf/skills).

It supports multiple source formats:

# GitHub shorthand

npx skills add vercel-labs/agent-skills

  
  
  Direct path to a specific skill within a repo


npx skills add https://github.com/vercel-labs/agent-skills/tree/main/skills/web-design-guidelines

  
  
  Local development paths


npx skills add ./my-local-skills

1. Managing Your Arsenal

Vercel Skills doesn’t just “drop and forget” files. It manages the lifecycle of your skills:

• npx skills list: List installed skills

• npx skills check: Check for available skill updates

• npx skills update: Update all installed skills to latest versions

• npx skills remove [skills]: Remove installed skills from agents

4. Verification: “What Can You Do?”

After installing a skill, the best way to test it is to simply ask your agent. Try asking your AI agent:

“What skills do you have access to right now?”

If everything is wired up correctly, the agent will have acsses to the new skill and be able to “learn” it easily, ready to be “woken up” when the task demands it.

You can see the full list of commands at vercel-labs/skills.

Sharing your own Skill

publishing your own skill is just as easy. To demonstrate, I built a simple GitHub repository: anuk909/Skills.

It currently contains:

• gh-pr-review: Improves agent capabilities for GitHub PRs (based on agynio/gh-pr-review).

• git-worktree-workflow: A skill I built (based on some cursor rule from my college) tha finally made the worktree workflow usable for me.

To try them out, just run: npx skills add anuk909/skills

Conclusion

Vercel Skills is the missing link for anyone serious about using AI agents. It solves the “manual installation” headache and makes specialized knowledge portable across our entire stack.

However, we are still in the early days. Unlike npm packages, Skills don’t yet have a robust way to manage versions or dependencies. While this simplicity makes them easy to deploy today, it will be interesting to see if the ecosystem evolves toward the complexity of traditional package managers.

What about you? Have you started using Skills in your workflow yet? What’s your preferred method of installing them today?

---

Originally published on AI Superhero