How to Install AI Agent Skills in Claude Code and Cursor

What This Gets You#

Installing AI agent skills turns your coding assistant from a general-purpose LLM into a tool that knows how to debug your stack, generate video, audit SEO, or research content — without you describing what you need each time. This tutorial covers installing skills in Claude Code, Cursor, and OpenCode using the skills.sh CLI, and should take about 10 minutes if you already have Node.js installed.

What You Need Before Starting#

• Node.js 18+ — The skills.sh CLI runs on npm. Run node -v to check. If you’re below 18, update with nvm install 18 or download from nodejs.org.
• A supported coding assistant — Claude Code (CLI), Cursor, or OpenCode. Each has a slightly different install path, covered below.
• GitHub CLI (gh) or a GitHub token — Only needed if you’re installing private skills or ones hosted on GitHub Enterprise. Public skills install fine without it.

Step-by-Step#

1. Install the skills.sh CLI#

The entire skill ecosystem runs through a single npm package:

npm install -g @skills-sdk/cli

Expected output (version may differ):

added 42 packages in 3s

Verify:

skills --version
# v0.5.0

If the command is not found after install, your npm global bin path may not be in $PATH. Run npm bin -g to find it, then add that directory to your shell profile.

2. Search for a skill#

Browse available skills by name or tag:

skills search debug

Expected output — a table of matching skills with name, install count, and a one-line description:

┌─────────────────────────────────┬──────────┬──────────────────────────────────┐
│ Name                            │ Installs │ Description                      │
├─────────────────────────────────┼──────────┼──────────────────────────────────┤
│ systematic-debugging            │ 289K     │ Debug systematically             │
│ debug-battle                    │ 701      │ AI agent for debugging battles   │
│ nextjs-debugging                │ 810      │ Next.js debugging agent tool     │
└─────────────────────────────────┴──────────┴──────────────────────────────────┘

Use skills search <term> --json if you want machine-readable output for scripting.

3. Install a skill#

Pick a skill from the search results and install it:

skills add systematic-debugging

Expected output:

✔ Installed systematic-debugging (289K installs)

The skill is now available in your local skill registry. It is not yet active in any assistant — that’s the next step.

Gotcha: Some skills live in monorepos and need a full GitHub URL plus --skill flag:

npx skills add https://github.com/agentspace-so/runcomfy-agent-skills --skill kling-3-0

If you install without the --skill flag on a monorepo, you’ll get a generic package with no useful commands. Check the skill’s install_cmd in its page header on skills.sh — it always shows the exact command.

4. Activate the skill in your assistant#

The activation step depends on which coding assistant you use:

Claude Code (CLI):

claude
# then inside Claude Code:
/skills add systematic-debugging

Or add it to your ~/.claude/skills.json:

{
  "skills": ["systematic-debugging"]
}

Skills added via the /skills command are available immediately in the current session. Skills added to skills.json are loaded on every new session.

Cursor:

Open Cursor Settings → Features → Skills, click “Add Skill”, and type the skill name. Cursor installs it from the skills.sh registry automatically. No terminal needed.

If the skill doesn’t appear in the dropdown, try restarting Cursor — the registry syncs on startup.

OpenCode:

Paste the install command in the agent’s terminal. OpenCode picks up skills automatically from the npm global registry:

npx skills add systematic-debugging

Then reload the agent with /reload or start a new session.

5. Use the skill#

Once activated, invoke the skill through your assistant’s command interface. The exact syntax varies, but most skills follow this pattern:

“In Claude Code, use systematic debugging on the auth module”

or with explicit skill invocation:

/skill systematic-debugging auth flow

If nothing happens, check that the skill name was typed correctly — /skills (with an ’s’) lists available skills, while /skill invokes one. They’re different commands.

Verification#

To confirm a skill is installed and active, ask your assistant to list available skills:

/skills

Expected output — a list of installed skills with their status:

Installed skills:
  systematic-debugging (active)
  vercel-react-best-practices (active)
  find-skills (inactive)

If the skill shows as “inactive”, run /skills add <name> again or restart your assistant session.

For CLI verification outside the assistant:

skills list

This prints all locally installed skills with their install paths.

One Thing That Tripped Me Up#

The skills add command and the /skills add chat command are not the same thing. The first installs the skill into your local registry (like downloading a package). The second activates it inside your assistant (like importing a module). I spent 20 minutes running skills add systematic-debugging and wondering why Claude Code couldn’t see it — I needed the second step inside the assistant. If a skill isn’t showing up after installation, check which step you’re missing.

Next Step#

Once you have a few skills installed, read how to find the right skill for your workflow — the skills.sh registry has over 10,000 packages and knowing how to search effectively saves more time than the installation itself.