Skills
A skill is a set of instructions packaged so you can invoke it with one command.
Instead of typing the same prompt every time you want Claude to review a pull request, write /review-pr. The skill contains the instructions, the quality gates, and the expected output format. You invoke it. Claude follows it.
Skills are how repeatable work becomes one-line commands.
How It Works
Skills are markdown files. They live in .agents/skills/ at your project root — one folder per skill, one SKILL.md file inside. The file contains whatever Claude needs to execute the workflow: steps, rules, examples, output formats.
Invoke a skill in Claude Code:
/skill-name
Claude loads the skill file and follows it. You can pass arguments:
/commit -m "fix login flow"
Skills can also load automatically when Claude recognises a relevant situation — or you can wire them to run as part of a hook (a script that fires on every file save, for example).
Built-In Skills
Claude Code ships with a library of common skills. A few worth knowing:
| Skill | What it does |
|---|---|
/commit | Writes a commit message from your staged changes |
/fix-links | Finds and repairs broken internal links |
/review-pr | Reviews a pull request and flags issues |
/perfect-day | Morning and evening session structure |
Run / in the Claude Code prompt to see what's available in your project.
Building Your Own
Any instruction you type more than three times is a candidate for a skill.
The structure:
.agents/skills/
└── my-skill/
└── SKILL.md
SKILL.md contains:
- Frontmatter — name, description, argument hint, trigger phrases
- Instructions — what Claude should do, step by step
- Quality gates — what done looks like
- Questions — what to ask if inputs are unclear
The AAIF standard governs the frontmatter format so skills work across different AI tools, not just Claude. One skill definition, any agent.
Composing Skills
Skills can call other skills. A publishing workflow might invoke /fix-links, then /review-pr, then /commit — all from a single /publish command.
This is how the system scales. Small, correct skills compose into larger workflows without becoming unmaintainable.
Context
- Config Architecture — where skills fit in the broader agent-agnostic setup
- Claude Code — how skills, hooks, and rules work together in the Claude Code loop
- Agents — when to use subagents instead of skills for parallel work
- Standards — the AAIF skill specification
- Onboarding — practical walkthrough of invoking your first skill
Questions
What's the highest-value repeatable workflow in your current work that you haven't turned into a skill yet?
- How do you decide when a workflow is stable enough to package as a skill versus keeping it as an improvised prompt?
- What's the difference between a skill that saves time and one that just moves the work somewhere else?
- If skills are the interface between your judgment and the machine, what does a library of 20 skills say about how you think?