How to Install Claude Skills from GitHub

Skill directories and install commands verified 31 July 2026 against Anthropic's Claude Code skills documentation.

A skill is just a folder with a SKILL.md file inside, plus any optional scripts or reference files. That simplicity is exactly why so many useful ones live on GitHub — Anthropic's own example repo, community collections, and single-purpose skills you stumble across. This guide walks through every safe way to install Claude Skills from GitHub, how to confirm it worked, and the security checks you should never skip. The steps work the same on the models currently available, including Claude Opus 5, Claude Sonnet 5, and Claude Haiku 4.5, and we review this guide regularly so the paths and commands stay accurate; a copy-and-verify install takes under 30 seconds once you know where the folder goes. We have walked this exact path in our own terminal, so the guidance here is grounded in doing rather than describing.

If you want to write your own instead of downloading someone else's, that is a different job — follow our guide on how to build a Claude skill. Here we are focused on adding skills that already exist. And if you are new to the feature entirely, the primer on Claude Code skills explains what they are and why they matter.

What you are actually installing

A directory whose entrypoint is SKILL.md — YAML frontmatter naming the skill and describing when to use it, then a Markdown body of instructions. Bundled scripts and reference files load only when the skill actually runs.

Before you install Claude Skills from GitHub, it helps to know what a skill is under the hood. According to the official Claude Code skills documentation, each skill is a directory with SKILL.md as its entrypoint. That file has two parts: YAML frontmatter (a name and a description that tells Claude when to use the skill) and a Markdown body containing the actual instructions Claude follows when the skill runs.

The important detail is the folder. When you "install" a skill, you are placing that folder — the one that contains SKILL.md at its root, not nested a level deeper — somewhere Claude Code looks. The SKILL.md must sit directly inside the skill folder, like my-skill/SKILL.md, or it will not be recognized.

Skills can also bundle extra files: helper scripts, reference documents, or example collections. Those load only when the skill actually runs, so a skill with a big reference file costs you almost nothing until it is needed. It also means a skill can carry code that executes on your machine — hold that thought for the security section.

Method 1: copy the skill folder into your skills directory

Clone the repo, find the directory holding SKILL.md, and copy it into ~/.claude/skills/ for every project or .claude/skills/ for one repo. The file must sit exactly one folder deep or Claude never sees it.

This is the most direct way to install Claude Skills from GitHub, and it works for any repo no matter how it is structured. You get the folder onto your disk, then drop it where Claude Code watches.

Step one — get the files. Clone the whole repository or download it as a ZIP:

git clone https://github.com/owner/repo.git

Step two — find the skill folder. Inside the repo, locate the directory that contains a SKILL.md. Some repos hold one skill at the root; larger collections hold many, each in its own subfolder.

Step three — place it in your skills directory. Claude Code loads skills from two locations you control:

Copy the skill folder into ~/.claude/skills/ if you want it everywhere, or into your project's .claude/skills/ if you want to share it with a repo and its collaborators. For example:

mkdir -p ~/.claude/skills
cp -r repo/skill-name ~/.claude/skills/skill-name

After copying, confirm the result is ~/.claude/skills/skill-name/SKILL.md — one folder deep, with SKILL.md at its root.

Step four — pick it up. Claude Code watches these directories, so adding a skill to an existing skills folder takes effect within your current session. But if ~/.claude/skills/ did not exist when your session started and you just created it, restart Claude Code so the new directory can be watched. When in doubt, start a fresh session — it is the reliable move.

Copying skills into this project's .claude/skills/ on our own Windows setup, the one mistake we made more than once was the nesting. A lot of repos wrap the actual skill a folder deeper than you expect, so a careless cp -r lands you with .claude/skills/repo-name/skill-name/SKILL.md and Claude never sees it. Now, every time we install Claude Skills from GitHub, we run a quick check that SKILL.md sits exactly one level down before we even try to invoke the skill — it has saved us far more time than the copy itself takes.

That is the whole flow. Clone, find the folder, copy it in, restart if needed. Once you have done it once, adding new skills becomes muscle memory, and it pairs nicely with the workflow habits in our Claude Code tips and tricks roundup.

Method 2: install via the plugin marketplace

When a repo ships as a plugin marketplace, register it with /plugin marketplace add owner/repo and install with /plugin install name@marketplace. Claude Code manages the files, and namespaced plugin skills cannot collide with your own.

Many GitHub repos are not just loose skill folders — they are packaged as Claude Code plugin marketplaces that bundle skills (and sometimes commands, hooks, and MCP servers) together. For those, you do not copy anything by hand. You add the marketplace and install the plugin, and Claude Code manages the files for you (see the official Claude Code plugins documentation).

The commands follow this shape inside a Claude Code session:

/plugin marketplace add owner/repo
/plugin install <name>@<marketplace>

The first command points Claude Code at the GitHub repository acting as a marketplace. The second installs a specific plugin from it. The exact plugin name and marketplace identifier come from the repo's own README, so always defer to the instructions the repo publishes rather than guessing. Plugin skills live under a plugin-name:skill-name namespace, which means they can never collide with your personal or project skills.

This method is cleaner for skills you expect to update, because reinstalling or updating the plugin is a single command rather than a manual re-copy. For the full walkthrough of adding marketplaces and managing plugins, see our guide to the Claude Code plugin marketplace. If a repo offers both a manual copy and a marketplace install, the marketplace route is usually the lower-maintenance choice.

Method 3: the Claude apps and API

Outside Claude Code, skills arrive through each surface's own mechanism — enabled or uploaded in the app's settings, or supplied through the API — rather than a local folder. Same SKILL.md concept, different delivery.

If you are not in Claude Code but in the Claude apps (desktop or web) or building against the Claude API, skills are provided through that surface's own mechanism rather than a local ~/.claude/skills/ folder. In those contexts you generally enable or upload the skill through the product's settings or the API's skill support, rather than dropping a folder onto your filesystem.

The concept is the same — a SKILL.md folder with instructions — but the delivery differs by surface. Check the documentation for the specific product you are using, since the enable-or-upload flow evolves and differs from the Claude Code file-based approach described above.

Where to find skills worth installing

Start with Anthropic's official example repository, then widen to well-known community collections and curated lists. Official repos are the safest starting point precisely because someone is accountable for what is in them.

Anthropic maintains a public example skills repository on GitHub with reference skills you can study or adapt. It is a good, trustworthy starting point precisely because it is official. Beyond that, community collections and curated lists gather skills for everything from document handling to note-taking. If you want ideas, our roundup of the best Claude skills highlights ones worth your time, and there are niche gems too — like a dedicated Claude skill for Obsidian if you live in a notes vault. The most popular community bundle is the claude superpowers skills collection, which ships dozens of planning, testing, and debugging workflows in one install.

Wherever the skill comes from, the installation mechanics are identical: it is still a folder with a SKILL.md, and you still use Method 1 or Method 2 to install Claude Skills from GitHub into your setup.

Choosing how to install Claude Skills from GitHub

Copy the folder for a single skill or a quick test, use the marketplace for bundles and anything you expect to update, and use the app or API mechanism when you are working outside Claude Code.

Here is how the three approaches line up so you can match one to your situation.

Most people who install Claude Skills from GitHub use Method 1 for a one-off skill and Method 2 when a repo ships a whole bundle. There is no wrong answer — they land the skill in a place Claude can reach.

How do you verify the skill actually loaded?

Start a fresh session and invoke it directly by typing a slash plus the directory name, or trigger it by describing a task its frontmatter matches. Nothing happening usually means a mis-nested SKILL.md.

Copying files is not the same as a working skill, so confirm it before you rely on it. Start a fresh Claude Code session in a project where the skill should be available. You can invoke a skill directly by typing / followed by its name — the directory name becomes the command, so ~/.claude/skills/summarize-changes/ gives you /summarize-changes.

The other confirmation is automatic activation: because the skill's description frontmatter tells Claude when to use it, ask Claude to do something that matches that description and watch whether the skill kicks in. If it does not appear, the usual culprits are a mis-nested SKILL.md (buried one folder too deep), a directory created after the session started (restart to fix), or a typo in the folder path. Fix the folder, restart, and check again.

Security: only install skills you trust

A skill can carry scripts that execute on your machine, and GitHub skills are not security-scanned. Read the SKILL.md, open every bundled script, favor official repos, and review project-level skills before trusting a workspace.

This is the part you cannot skip. A skill can include instructions and scripts that run on your machine. Installing one is closer to running someone else's code than to reading a blog post — treat it that way. GitHub skills are not security-scanned, so the burden of vetting is on you.

Before you install Claude Skills from GitHub from any source you do not already trust:

• Read the SKILL.md. It is plain text. Read the instructions Claude will follow and make sure nothing looks manipulative or destructive.
• Inspect every bundled script. Open any .sh, .py, or other executable files in the skill folder and understand what they do before the skill can run them.
• Prefer official and well-known repos. Anthropic's own example repo and widely-used community collections are far safer bets than an unknown one-off.
• Be extra careful with project-level skills. A skill checked into a repo's .claude/skills/ can request broad tool access once you trust the workspace, so review project skills before accepting a repository's trust dialog.

None of this is meant to scare you off — skills are genuinely useful and most are harmless. But "it was on GitHub" is not a substitute for reading the code. Vet first, install second. That single habit is what separates a productive setup from a compromised one, and it applies to everything you add to Claude Code, not just skills.