Claude Skills: Create, Install, and Use Skills Across Claude.ai, Claude Code, and the API

Claude Skills turn your expertise, procedures, and best practices into reusable capabilities that Claude applies automatically — across Claude.ai, Claude Code, and the API, without modification. If prompts are instructions you repeat, Skills are instructions you package.

A Skill is a folder with a SKILL.md file containing Markdown instructions, plus any reference files, executable scripts, or code Claude needs to do the job.

Why Skills Matter Now

Skills went from a niche feature to a cross-industry standard in 2026. Anthropic positioned Skills as the primary way to package agent knowledge — and the pattern is spreading. Google announced that Gemini's Gems are being retired in favor of Skills, and open-source harnesses from OpenCode to Jcode and Pi all consume the same SKILL.md-style format.

The key properties:

• Consistent output — Claude applies your documented procedure every time.
• Organizational knowledge — package company procedures so new team members get expert-level results.
• Build once, run everywhere — the same Skill works in Claude.ai, Claude Code, and the API.
• Stackable — combine Skills for multi-step workflows; Claude uses what's needed when it's needed.

SKILL.md Anatomy

A Skill is a folder with a SKILL.md file containing instructions in Markdown. Add reference files, executable scripts, or code — whatever Claude needs to do the job.

my-skill/
├── SKILL.md          # Required — the instructions
├── LICENSE.txt       # Optional — licensing terms
├── examples/         # Optional — reference material
│   └── 3p-updates.md
└── scripts/          # Optional — executable helpers
    └── format.py

The SKILL.md frontmatter:

---
name: internal-comms
description: A set of resources to help me write all kinds of internal communications, using the formats that my company likes to use. Claude should use this skill whenever asked to write some sort of internal communications.
license: Complete terms in LICENSE.txt
---

The Body

The body tells Claude how to use the Skill — when to trigger it, and what procedures to follow:

## When to use this skill

Use this skill for:
- 3P updates (Progress, Plans, Problems)
- Company newsletters
- FAQ responses
- Status reports

## How to use this skill

1. **Identify the communication type** from the request.
2. **Load the appropriate guideline file** from the `examples/` directory.
3. **Follow the instructions** in that file for formatting and tone.

If the communication type doesn't match any existing guideline, ask for clarification.

## Keywords

3P updates, company newsletter, company comms, status reports, internal comms

The description and Keywords sections matter for automatic loading — Claude matches the current task against them to decide whether to inject the Skill.

Creating a Skill

Option 1: Use the Skill Creator (recommended for beginners)

Describe what you want, and Claude generates the folder structure, formats the SKILL.md file, and bundles your resources. Start in Claude.ai → Settings → Capabilities, or just ask: "Create a Sales Call Prep skill based on this conversation."

Option 2: Build it by hand

mkdir -p my-skill
cd my-skill

---
name: code-review
description: Guides Claude through a structured code review following the team's standards. Use whenever reviewing a pull request or changeset.
---

mkdir examples scripts

Installing Skills

Skills live in the .claude/skills/ directory for Claude Code, and in your Claude.ai capabilities settings for the web/app. For Claude Code:

mkdir -p .claude/skills
# Copy your skill folder in
cp -r my-skill .claude/skills/

Skills also work across the ecosystem: OpenCode loads skills from .opencode/skills/, Jcode from ~/.jcode/skills/, and Pi from .pi/ project resources — most using the same SKILL.md structure. See the Agent Skills guide for the cross-tool standard.

Skills vs Prompts, Projects, MCP, and Subagents

The distinction: Skills are instructions you package once and reuse across many contexts; MCP gives the agent capabilities (tools), while Skills give it method (how to do things well). Subagents are workers; Skills are manuals those workers can read.

Best Practices

• Write the description for loading, not for humans — include trigger phrases and keywords so Claude knows when to apply it.
• Keep instructions procedural but not rigid — state the outcome and the checklist, not a step-by-step script (see why simplified prompts work better).
• Bundle examples — reference implementations in examples/ beat inline prose.
• Stack, don't bloat — separate concerns into multiple small Skills instead of one huge manual; Claude composes them as needed.
• Version with the repo — commit Skills alongside the code they govern.

Related Pages

• Agent Skills (Agents Section) — cross-tool SKILL.md standard
• Gems Are Retiring: Move to Skills (Blog) — Google's Gems→Skills migration
• Prompting Claude Code — CLAUDE.md patterns
• Agent Memory Architectures — how agents persist context