Claude SkillsLesson 1 of 15

Introduction to Claude Skills

Skills package instructions, checklists, and multi-step procedures into reusable commands. Write once, invoke with /skill-name or let Claude load them when your request matches.

What are Claude Skills?

A skill is a SKILL.md file that gives Claude a procedure, checklist, or set of instructions. Claude loads it when relevant, or you invoke it directly with /skill-name.

Skills work across Claude.ai, Claude Code, and the API. They follow the open Agent Skills standard, so the same format is portable across tools.

Unlike CLAUDE.md, a skill's full content loads only when the skill runs. Large reference docs or step-by-step procedures cost almost nothing in context until you need them.

When should you create a skill?

Create a skill when you notice any of these patterns:

You paste the same checklist or instructions into every new session.
A section of CLAUDE.md has grown into a full step-by-step workflow rather than a simple fact.
A workflow always runs the same sequence — review, commit, deploy — and you want that bundled.
You want your whole team to get the same output from recurring tasks like PR reviews or standups.

Skills vs CLAUDE.md

Topic	CLAUDE.md	Skill (SKILL.md)
Content type	Facts, conventions, project rules	Procedures, checklists, workflows
When it loads	Every session from the start	Only when the skill is invoked
Context cost	Always uses tokens	Zero until invoked
Invocation	Automatic — always active	Auto (by description) or manual (/name)
Best for	Code style, branch naming, test runner	Deployments, PR reviews, commit workflows
Scope	Project or personal	Enterprise, personal, project, or plugin

Anatomy of a SKILL.md file

Every skill is a directory named after the command you type. Inside, SKILL.md is the only required file. It has two parts:

YAML frontmatter — between --- markers. The description field is the most important; Claude uses it to decide when to auto-invoke the skill.
Markdown content — instructions, checklists, templates, and shell commands that inject live data.

~/.claude/skills/summarize-changes/SKILL.md

---
name: summarize-changes
description: Summarises uncommitted git changes and flags anything risky.
---

## Current changes

!`git diff HEAD`

## Instructions

Summarise the changes above in 2–3 bullet points, then list any risks such as missing
tests or hardcoded values.

The !`git diff HEAD` line is dynamic context injection: Claude Code runs the command before sending the skill to Claude, so it always sees live data — not a static placeholder.

How to invoke a skill

Mode	How it works	Example
Automatic	Claude reads each skill's description and loads it when your request matches.	What did I change?
Direct	Type `/skill-name` to invoke it yourself.	/summarize-changes

Terminal

# Method 1 — let Claude decide

$ claude

> What did I change?

→ Loading skill: summarize-changes

Changes: Added rate limiter to /api/users

Risk: no tests added for the limiter

# Method 2 — invoke directly

$ claude

> /summarize-changes

→ Running skill: summarize-changes

Changes: Added rate limiter to /api/users

Where skills live

Where you store a skill determines who can use it. When skills share the same name, enterprise overrides personal, and personal overrides project. Plugin skills use a plugin-name:skill-name namespace.

Scope	Path	Available to
Enterprise	Managed settings	All users across your organisation
Personal	~/.claude/skills/<name>/SKILL.md	All of your projects on this machine
Project	.claude/skills/<name>/SKILL.md	This project only — commit to Git for the team
Plugin	<plugin>/skills/<name>/SKILL.md	Wherever the plugin is enabled

Bundled skills in Claude Code

Claude Code ships with ready-to-use bundled skills. Unlike fixed commands, these are prompt-based — they give Claude detailed instructions and let it orchestrate the work.

Command	What it does
/simplify	Simplifies complex code or explanations to be more readable
/debug	Systematically debugs an issue using logs, errors, and code
/batch	Applies the same transformation across multiple files
/loop	Repeats a task until a condition is met or you stop
/claude-api	Helps you call the Claude API directly from code

Create your first skill in 3 steps

A personal commit helper that stages all changes and writes a commit message.

Create the skill directory

mkdir -p ~/.claude/skills/commit

Write SKILL.md

~/.claude/skills/commit/SKILL.md

---
name: commit
description: Stage all changes and write a descriptive commit message.
disable-model-invocation: true
allowed-tools: Bash(git add *) Bash(git commit *)
---

Stage all changes and write a commit message that follows this format:

<type>(<scope>): <summary>

Types: feat, fix, chore, docs, refactor, test
Keep the summary under 72 characters.
Write a short body if the change needs context.

disable-model-invocation: true prevents Claude from triggering /commit on its own — this is a workflow you always run yourself.

Test the skill

Terminal

$ claude

> /commit

→ Running skill: commit

[git add .]

feat(auth): add token refresh with exponential backoff

1 file changed, 34 insertions(+)

Ways to build a skill

Approach	When to use it
Write SKILL.md by hand	Full control over every line. Best for simple, focused skills.
Ask Claude in chat	Describe what you want; paste the generated SKILL.md into your skills folder.
/skill-creator plugin	Guided Q&A writes and tests the full skill — includes Eval, Improve, and Benchmark modes.

Before you continue

Personal path: ~/.claude/skills/<name>/SKILL.md; project path: .claude/skills/<name>/SKILL.md.
Invoke manually with /skill-name; auto-trigger uses the description field.
Use disable-model-invocation: true for skills you want to trigger yourself.
Dynamic data injection: !`shell-command` in SKILL.md runs before the skill loads.
Next: activating skills in your Claude.ai account.

What's Next

Next: activating skills in your Claude account and seeing how they appear in real conversations.

← PREVIOUS

Choosing the Right Claude Tool

Use Skills in Claude Account