[Claude Code] How to Create and Use Skills - Your Own Slash Commands

If you're rewriting the same prompt every time

When you use Claude Code, this kind of situation comes up quickly:

"Write a blog post. Oh, but always use this style... do I have to explain it again?"

Typing a long context every time, like "write in Korean, use a developer tone, and follow this structure," is tedious. If you make one mistake, the output can come back in a completely different style.

This is exactly where Skills help.

Skills are Claude Code's custom slash commands. If you save a prompt pattern in a file, you can bring it back with a short command like /blog-write. After using it myself, I kept thinking, "I wish I had known this sooner."

Skill in one line

Skill = a Markdown file that contains a prompt template + slash command registration

No complicated setup. Create one file and it becomes a slash command.

Preparation

First, let's look at where skill files are stored.

~/.claude/skills/
└── skillname/
    └── skill.md ← This one file is the entire skill.

~/.claude/skills/ is the global skills directory. It can be used in any project.
The skill folder name becomes the slash command name.
- ~/.claude/skills/blog-write/ → /blog-write
- ~/.claude/skills/gitpush/ → /gitpush

Step 1. Create `skill.md`

Let's create a skill folder and file. As an example, we'll make a simple commit message generator.

mkdir -p ~/.claude/skills/commit-msg

Then create ~/.claude/skills/commit-msg/skill.md.

---
name: commit-msg
description: Create a Korean conventional commit message based on git diff.
user-invocable: true
---

Analyze the current changes and write a conventional commit message in Korean.

## Rules
- Start with one of the following: `feat:`, `fix:`, `refactor:`, `docs:`, `style:`, or `chore:`
- Keep the title within 50 characters
- If multiple files changed, summarize the key changes
- Keep English technical terms, but explain in Korean

## Output format
Output only the commit message. No extra explanation.

The file structure has two main parts.

Section	Content
Frontmatter (between `---`)	Skill metadata: `name`, `description`, `user-invocable`
Body	The actual prompt that gets sent to Claude

Frontmatter fields

---
name: commit-msg # Slash command name (/commit-msg)
description: one-line description # Description shown in the `/` list
user-invocable: true # Register it so it can be used as a slash command
---

If user-invocable: true is missing, it won't appear in the slash command list. You must include it.

Step 2. Use the skill

At the Claude Code prompt, type / to see the list of registered skills.

/commit-msg

After you press Enter, the body prompt in skill.md is automatically passed to Claude. It will analyze the current git diff and generate a commit message right away.

Step 3. Pass arguments with `$ARGUMENTS`

If you want to pass extra input into a skill, use the $ARGUMENTS variable.

For example, if your blog-writing skill takes a topic as an argument:

---
name: blog-write
description: Write a developer blog post in Markdown.
user-invocable: true
---

Write a blog post on the following topic.

## Style
- Korean, friendly, and practical tone
- Include code examples
- Audience: developers seeing this for the first time

## Topic
$ARGUMENTS

When using it, type:

/blog-write How to create and use skills in Claude Code

The text "How to create and use skills in Claude Code" is substituted into $ARGUMENTS.

Step 4. Use `skill-creator` to make skills

Claude Code has a built-in skill called skill-creator. It creates skills for you.

/skill-creator blog-write: Create a skill for writing developer blog posts

Claude will read your skill description and automatically create the skill.md file. This is faster when you're not sure how to structure the prompt.

Common patterns

1. Basic skill structure

---
name: skill-name
description: one-line description
user-invocable: true
---

Write the prompt here.

2. Skill with arguments

---
name: skill-name
description: description ($ARGUMENTS supported)
user-invocable: true
---

topic: $ARGUMENTS

Please write ... based on the topic above.

3. File path structure

~/.claude/skills/
├── blog-write/
│   └── skill.md
├── commit-msg/
│   └── skill.md
└── gitpush/
    └── skill.md

Skill vs. repeated prompt

	Type it manually every time	Skill
Input size	Full long prompt	Single `/skillname` line
Consistency	Can vary each time	Always the same prompt
Arguments	Not available	Available via `$ARGUMENTS`
Sharing	Difficult	Easy to share as a file

Troubleshooting

`/skillname` doesn't show up in the list

Make sure user-invocable: true is in the frontmatter
Make sure the file path is ~/.claude/skills/skillname/skill.md (case-sensitive)
Try restarting Claude Code

The prompt doesn't work as expected

Extra text around $ARGUMENTS can confuse Claude
The more clearly you specify the output format, the more stable the result will be
"Output ~" is more consistent than "Do ~ for me"

Changes to `skill.md` aren't reflected

After saving the file, start a new conversation in Claude Code and it will be reflected
It may not be reflected immediately in an existing conversation

Summary - core flow at a glance

1. Create a folder
   ~/.claude/skills/skill-name/

2. Write skill.md
   ---
   name: skill-name
   description: description
   user-invocable: true
   ---

3. Use it
   /skill-name

If you use Skills well, repetitive prompt work drops dramatically and your prompt quality becomes much more consistent. For anything you repeat often, Skills are worth setting up once.