Raju H
BootcampsSign in
Blog

Claude Skills Explained: How to Install Them and Which Ones Are Worth It

What Claude Skills (Agent Skills) actually are, the correct way to install them (the viral posts get the command wrong), and a practical filter for which skills are worth adding.

Raju H

AI Builder · Dhaka

· 4 min read

TL;DR

A "Claude Skill" (more precisely, an Agent Skill) is a packaged specification that tells an AI agent how to do one job well — an SEO audit, a code review, a PRD draft. You install them with npx skillkit install <owner>/<repo> (the viral posts saying npx skills add are wrong). They work with Claude Code, Cursor, Codex, and any agent supporting the Agent Skills spec. This is the correct, no-hype guide to what they are, how to add them, and how to decide which are worth keeping.

What is a Claude Skill?

A skill is a folder of instructions — usually a SKILL.md file plus supporting references — that defines a repeatable task: its inputs, the steps to perform, and the output format. When you install a skill, your AI agent gains the ability to invoke it by name.

The key mental model: a skill is a saved spec. Instead of re-typing a detailed prompt every time you want, say, an SEO audit, someone has encoded that prompt-plus-process into a skill once, and now any agent can run it consistently. That's why spec-driven AI and skills are two sides of the same coin — a skill is just a spec good enough to package and share.

Skills follow the open Agent Skills spec, so they're not Claude-locked. The same skill runs in Claude Code, Cursor, Codex, Windsurf — anything that implements the spec.

How to install a skill (the correct way)

The viral marketing posts tell you to run npx skills add .... That command doesn't exist and will error. The real command uses skillkit:

npx skillkit install <github-owner>/<repo>

For example, to install Corey Haines's popular marketing skills:

npx skillkit install coreyhaines31/marketingskills

That fetches the skills from the repo and installs them where your agent can find them. After installing, your agent can invoke any skill in the pack.

Two-step setup from scratch

If you're starting with nothing:

  1. Install the agent. For Claude Code: install VS Code, then Claude Code (follow Anthropic's official setup). For Cursor or another agent, install that instead.
  2. Add the skills. In your terminal: npx skillkit install <owner>/<repo>.

Then invoke a skill from your agent. That's the whole loop.

Note: skill ecosystems and tooling move fast. Always check the specific repo's README for its exact install instructions — some publish pre-built .skill files (e.g. the mysticaltech fork) that install differently. When a blog post and a repo's README disagree, follow the README.

How to tell if a skill is worth installing

Not every skill pack deserves a slot in your setup. Here's the filter I use:

1. Who wrote it? The best skills encode real domain expertise. The marketing-skills repo is good because Corey Haines is an actual conversion specialist who runs an agency — the skills are his real process, packaged. A skill written by someone with no domain track record is just a fancy prompt. Check the author's credibility in the skill's domain.

2. Is the output a draft or a decision? Good skills produce drafts you refine — an audit, a first-pass copy, a structured analysis. Be skeptical of any skill that claims to make business decisions for you. (See why "AI replaced my agency" is overstated.)

3. Can you read the SKILL.md? Open the skill's source on GitHub before installing. If the SKILL.md is a clear, well-structured spec, it'll behave predictably. If it's vague hand-waving, the output will be too. Read the spec; it tells you exactly what you'll get.

4. Does it fit your actual workflow? A skill you'll run weekly earns its slot. A skill you install because a viral post hyped it, then never touch, is clutter. Install for a real recurring task, not FOMO.

What good skill packs exist right now

A few worth knowing about (verify each repo's current state before installing):

  • coreyhaines31/marketingskills — CRO, SEO, copywriting, analytics, ad creative, A/B testing. The one in the viral posts; genuinely good.
  • Built-in / first-party skills — your agent likely ships with some (code review, doc generation). Check your agent's docs.
  • Domain-specific community packs — search GitHub for agent skills <your-domain>. Apply the four-point filter above before installing any.

The principle underneath

Skills are popular for a reason that has nothing to do with hype: they make good prompting reusable. The hard part of getting useful work from an AI agent was never the agent — it was knowing precisely what to ask for and how to constrain it. A skill captures that knowledge once so you don't re-derive it every time.

If you internalize one thing: a skill is only as good as the spec inside it. The best way to get value from skills — and to write your own — is to understand how to write a good spec in the first place. Once you can write the spec, you can write the skill.

What to do next

  1. Install one skill pack for a task you do weekly: npx skillkit install <owner>/<repo>.
  2. Read its SKILL.md first so you know what to expect.
  3. Run it on a real task, review the output, keep the skill only if it earns its slot.
  4. When you find yourself re-typing the same detailed prompt repeatedly, that's your signal to write it as a spec — and then as your own skill.

To set up a project for agents from scratch, two free tools help: the AGENTS.md generator for the rules file, and the AI agent project starter kit, which generates the rules file plus a subagent and a slash command together.

Sources

This site uses analytics to improve future bootcamps. Privacy & cookies