Skills
Reusable prompt modules that encode your expertise into callable workflows. Still just text.
What Are Skills?
Skills are markdown files that define reusable workflows Claude can invoke. They're like recipes or checklists that encode your expertise into a format Claude can follow.
Instead of explaining your code review process every time, you create acode-review skill. Claude invokes it with /code-reviewand follows your exact process.
The key insight: Skills let you encode "how I do things" once, then apply that expertise consistently across all future work.
Anatomy of a Skill File
Skills live in ~/.claude/skills/[skill-name]/SKILL.md. Here's the structure:
--- name: quality-check description: Automated quality audit for projects. Checks CLAUDE.md compliance, folder structure, git safety, and dev environment. allowed-tools: Read, Grep, Glob, Bash --- # Quality Check Skill Run automated quality checks following my protocols. ## When to Activate - Opening a project with CLAUDE.md - User asks to "check quality" or "audit" - Before major commits or deployments ## Checks Performed ### 1. CLAUDE.md Compliance - **Line count**: Target <200 lines - **Memory triggers**: Brief 1-liners, not paragraphs - **@imports**: External references for detail ### 2. Folder Structure - **3-5 files per folder**: Flag folders with >5 files - **Archive READMEs**: Every archive folder needs context ## Output Format [Define the expected output structure] ## Quick Commands ```bash # Count CLAUDE.md lines wc -l CLAUDE.md ```
Frontmatter
Name, description, and which tools the skill can use.
When to Activate
Trigger phrases or situations that invoke this skill.
Process/Checks
The actual steps Claude should follow.
Output Format
How results should be structured and presented.
Example Skills
🔍 code-review
Systematic code review for quality, correctness, and maintainability. Covers logic, error handling, type safety, patterns, and performance.
Trigger: /code-review or "review this code"
🚀 deployment-checklist
Pre-deployment verification for production releases. Build verification, environment parity, performance baselines, rollback planning.
Trigger: /deployment-checklist or before deploying
📊 estimating-time
Estimate time and cost for development tasks. Uses historical data and client-type pricing from your portfolio.
Trigger: /estimating-time or "how long will this take"
📋 tracking-scope
Detect and flag scope creep in client projects. Compares requests against agreed baselines, suggests handling approaches.
Trigger: Phrases like "can we also", "quick change", "while you're at it"
Creating Your Own Skill
- 1.Identify a repeating workflow
What do you explain to Claude over and over? That's a skill candidate.
- 2.Create the directorymkdir -p ~/.claude/skills/my-skill
- 3.Write SKILL.md
Add frontmatter (name, description, tools), activation triggers, steps to follow, and output format.
- 4.Test it
Invoke with
/my-skilland iterate on the process.
Claude Code 2.1 Tip: Skills now support hot-reload - edit your SKILL.md file and changes apply immediately without restarting your session.
Skills vs CLAUDE.md
| Aspect | CLAUDE.md | Skills |
|---|---|---|
| Purpose | Passive context | Active workflows |
| When loaded | Always, at session start | On-demand, when invoked |
| Scope | Project-specific | Cross-project reusable |
| Content | Context, conventions, memory | Processes, checklists, workflows |
Think of it this way: CLAUDE.md is "what Claude should know". Skills are "what Claude should do".