Claude Code Skills: A Practical Guide

Claude Code skills are reusable instructions, packaged as a SKILL.md file, that teach the agent how to do a specific job well. When a request matches a skill's trigger, Claude loads its guidance and runs the task the way the author intended, instead of improvising from scratch. Think of a skill as a senior colleague's playbook the agent can pull off the shelf on demand.

This is the practitioner's companion to Anthropic's official docs. The docs define skills; this guide is about which ones are worth installing and how they behave once you move off Claude Code.

How Claude Code skills work

A skill is a folder with a SKILL.md at its root. The frontmatter carries a name and a description, and the body holds the instructions. The description is what the agent matches against, so a sharp description is what makes a skill trigger reliably. Skills can bundle extra files: scripts the agent runs, reference docs it reads on demand, and on Claude Code they can also use allowed-tools, hooks, and MCP servers.

Claude Code reads skills from ~/.claude/skills (personal) and from a project's .claude/skills (shared with the repo). That second location is why a skill can travel with a codebase and work for the whole team.

The SKILL.md format

The format is deliberately small. A minimal skill is a name, a description, and a body of plain Markdown instructions. The power comes from progressive disclosure: the agent reads the short description first, and only loads the full body and bundled files when the skill actually fires. That keeps context lean until the skill is needed.

How to choose a good skill

Most skill lists rank by stars. We rank by whether the skill earns its place on a real team:

1. Does the description map to a clear, recurring job, not a vague theme.
2. Does it back its work with something concrete, scripts, checks, or a defined process, rather than just lecturing the model.
3. Is the source maintained, and is the install honest about what it needs.
4. How does it behave on other agents. A skill that hard-depends on a Claude-only feature is fine, as long as you know that before you standardize on it.

Cross-agent reality

SKILL.md is an open format. Codex reads .agents/skills and needs the name and description frontmatter. OpenCode has native skill support and also reads .claude/skills and .agents/skills. So most skills port without edits. What does not port cleanly is the Claude-specific plumbing: hooks, some MCP setups, and bundled behavior can degrade or no-op elsewhere. Every skill in our catalog carries a compatibility note for exactly this reason.

Where to go next

• See the best Claude Code skills for our tested shortlist.
• Browse the full curated skills marketplace by role.
• New to the concept? Start with what are Claude skills.
• Ready to install? Follow how to use Claude skills.