Guides · 2026-10-16

Claude Code Skills Examples You Can Install and Run

The short answer

A Claude Code skill is a folder with a SKILL.md file: a name, a short description, and the steps Claude should follow. Claude loads it when your request matches the description, or you run it with /skill-name. The useful examples are the jobs you already repeat: review a diff, add tests, debug a failure, write a commit message.

What goes in the file

Keep the frontmatter to name and description. Put the job in the description first, because that is what Claude matches on. The body is the procedure: what to read, what to check, and when to stop and ask. A skill that tries to do five jobs gets used for none of them.

Example: review the current diff

Ask for bugs, security issues, data loss, breaking changes, and missing tests. Limit the review to the diff plus nearby context, and have it name the file and the line. This is the skill you run before you commit, while the change is still local.

  • Read the diff, not the whole repository
  • Flag risk before style nits
  • Name the test that should exist and does not

Example: write tests in the project's framework

Point the skill at the code you just changed and tell it to use the test runner and style already in the repo. One focused test that fails for the right reason is worth more than a file of boilerplate. Re-run the suite before it claims the work is done.

Example: debug a failing test

A debug skill should reproduce the error, state one hypothesis, check it, and only then change code. Skipping the reproduce step is how you get a pile of guesses. When the hypothesis is wrong, it writes that down and tries the next one.

Example: commit message and PR description

Have the commit skill read the staged diff and draft a Conventional Commit. It should ask before it commits. A PR skill writes the title, what changed, how you tested, and the risk. Neither one pushes or merges.

Install a set instead of pasting prompts

You can copy a single SKILL.md into .claude/skills/ for the team, or into ~/.claude/skills/ for yourself. Claude Code watches those folders, so a new skill shows up in the same session. Type / to see it. The same folders upload as skills on claude.ai, because the frontmatter stays on the portable name and description fields.

Skip writing them from scratch

The Claude Code Skills Pack has eight of these ready to install: review changes, write tests, debug an error, refactor safely, security audit, commit message, PR description, and docs. The installer backs up anything with the same name. If you live in Cursor instead, the Cursor Rules Pack is the file-scoped version of the same idea. For a review that posts on the pull request itself, the PR Code Review Agent is a GitHub Action, separate from the local review skill.

Related

FAQ

Where do Claude Code skills live?

In the repo at .claude/skills/ so the team gets them through git, or in ~/.claude/skills/ for every project on your machine. Type / or run /skills to list them.

How is a skill different from a long prompt?

A prompt disappears when the chat does. A skill stays on disk, loads only when the description matches, and runs the same steps every time. You can still edit the file.

Will these replace Claude Code's built-in commands?

Not if the names avoid the bundled ones, such as /code-review and /debug. Pick names like review-changes and debug-error.

Agent Drop: the weekly newsletter

One new agent template, one tool worth using, three links. Free, weekly.