Guides · Updated
How to Use Claude Skills
How to install and run Claude skills in Claude Code, on claude.ai and in other agents, including the npx skills add command and what to do when a skill does not trigger.
To use a Claude skill, put its folder where Claude can find it, then either ask for something that matches the skill's description or call it by name. In Claude Code that means a folder under ~/.claude/skills/ or .claude/skills/, and a command such as /skill-name. On claude.ai it means uploading the skill as a zip file in settings. The rest of this guide covers each route, starting with the quickest.
If you are not sure what a skill is yet, read what are Claude skills first.
Install with npx skills add
Most published skills live in GitHub repositories. The open-source skills CLI from Vercel Labs installs them with one command:
npx skills add vercel-labs/agent-skills
The argument is a GitHub owner/repo. A full GitHub or GitLab URL, any git URL, or a local path also works. The CLI detects which coding agents you have installed and asks where to put the skill.
The options you will use most:
| Option | What it does |
|---|---|
-l, --list | List the skills in the repository without installing |
-s, --skill <name> | Install only the named skills |
-a, --agent <agent> | Install to specific agents, for example claude-code or codex |
-g, --global | Install to your user directory instead of the current project |
-y, --yes | Skip the confirmation prompts |
So this installs one skill, for Claude Code only, for every project on your machine:
npx skills add vercel-labs/agent-skills --skill frontend-design -g -a claude-code
Without -g, the skill goes into the current project, which for Claude Code is .claude/skills/. That is the right choice when the skill should be committed and shared with your team. When you install interactively, the CLI offers a symlink, which keeps one copy that every agent points to, or a plain copy.
The same tool manages what you have installed: npx skills list shows installed skills, npx skills update updates them, npx skills remove removes them, and npx skills find searches for new ones.
The CLI collects anonymous usage telemetry by default. Setting DISABLE_TELEMETRY=1 turns it off.
Install by hand in Claude Code
A skill is only a folder, so you can also copy it into place yourself. Claude Code looks in these locations:
| Location | Path | Applies to |
|---|---|---|
| Personal | ~/.claude/skills/<skill-name>/SKILL.md | All your projects on this machine |
| Project | .claude/skills/<skill-name>/SKILL.md | This repository, and anyone who clones it |
| Plugin | Inside an installed plugin | Wherever the plugin is enabled |
Download or clone the skill, then move its folder so that SKILL.md sits directly inside a folder named after the skill.
Run a skill
There are two ways to trigger a skill in Claude Code.
Let Claude pick it. Ask for something that matches the skill's description. A skill described as "Summarizes uncommitted changes and flags anything risky" will load when you ask "What did I change?"
Call it yourself. Type / and the skill name:
/summarize-changes
Skills that come from a plugin carry the plugin's name as a prefix, such as /commit-commands:commit.
Some skills only work one way. A skill author can set disable-model-invocation: true so that only you can run it, which is common for anything with side effects such as deploying or committing. The opposite setting, user-invocable: false, makes a skill background knowledge that only Claude loads.
To see what is installed, type / to open the command menu, or ask Claude "What skills are available?"
Use skills on claude.ai
On claude.ai, the pre-built skills for PowerPoint, Excel, Word and PDF files work without setup: Claude uses them when you ask for a document.
Custom skills are uploaded. Anthropic's documentation gives the route as Settings > Features, where you upload the skill folder as a zip file. This is available on Pro, Max, Team and Enterprise plans with code execution enabled. An uploaded skill belongs to the user who uploaded it.
A skill written for Claude Code may need trimming before it uploads. claude.ai accepts only the frontmatter fields in the open standard (name, description, license, compatibility, metadata, allowed-tools) and rejects a file that contains others.
Use skills in other agents
The format is an open standard, so other agents read the same folders. With the skills CLI, add --agent to target one:
npx skills add vercel-labs/agent-skills -a codex
Each agent has its own folder and its own way of calling a skill. Codex reads skills from .agents/skills in a repository, and you mention a skill by typing $ or running /skills. OpenClaw loads skills from a skills folder in its workspace and has its own openclaw skills install command. Check your agent's documentation for the rest; the agent pages list skills by agent.
When a skill does not trigger
Anthropic's troubleshooting advice for a skill that is not used when you expect it:
- Check that the description includes words you would naturally say.
- Check that the skill appears when you ask "What skills are available?"
- Rephrase your request so it is closer to the description.
- Run it directly with
/skill-name.
If you have many skills installed, Claude Code may shorten some descriptions to fit its budget for the skill listing, which removes the keywords Claude matches on. Running /doctor shows what the listing costs.
If a skill triggers too often, the fix is the reverse: make its description more specific.
Before you install anything
A skill can instruct the agent to run commands, and it can ship scripts. Read SKILL.md and any scripts before you install a skill from someone you do not know, and prefer repositories that are maintained. In Claude Code, also look at the skill's allowed-tools line, which lists tools the skill may use without asking you.
When you are ready, browse the directory. Each skill page has the install command for your agent.
Questions people ask
What does npx skills add do?
It runs the open-source skills CLI, which fetches a repository, finds the SKILL.md folders in it, and installs the ones you choose into the skills folder of each agent you select. By default it installs into the current project; the -g flag installs for your user account instead.
Where does Claude Code look for skills?
In ~/.claude/skills/ for personal skills that apply to all your projects, in .claude/skills/ inside a project, and inside any enabled plugin.
How do I run a skill manually?
In Claude Code, type a slash followed by the skill name, for example /summarize-changes. Skills from a plugin are prefixed with the plugin name, as in /my-plugin:review.
How do I remove a skill?
Delete its folder from ~/.claude/skills/ or .claude/skills/, or run npx skills remove with the skill name if you installed it with the skills CLI. A skill that came with a plugin is removed by disabling or uninstalling the plugin.
Can I use the same skill in Codex or Cursor?
Usually. The skills CLI installs to many agents with the --agent flag, and skills that follow the open Agent Skills format are read by any agent that supports it. Features specific to Claude Code may not carry over.