create-skill

A Claude Code skill that helps you create new Claude Code skills. Invoke it with /create-skill [name] [description] and it will scaffold a well-structured SKILL.md with the correct frontmatter, file layout, and patterns.

Purpose

Creating Claude Code skills involves getting several things right: YAML frontmatter fields, variable substitution syntax, tool permissions, invocation control, and directory placement. This skill encodes all of those conventions so you can describe what you want and get a working skill created for you.

Prerequisites

• Claude Code installed and working
• No additional dependencies — this skill is pure instructions, no scripts

Usage

/create-skill my-skill-name Does something useful

Arguments:

• First argument: skill name (lowercase, hyphens, max 64 chars)
• Remaining arguments: description of what the skill should do

Claude will create the skill directory and SKILL.md at the appropriate location.

What it covers

Skill locations

Location	Path	When to use
Personal (default)	~/.claude/skills/<name>/SKILL.md	Available across all projects
Project	.claude/skills/<name>/SKILL.md	Project-specific skills

Frontmatter fields

Field	Purpose
name	Slash command name
description	What the skill does (used for auto-invocation matching)
allowed-tools	Tools allowed without permission prompts
model	Model override when skill is active
argument-hint	Autocomplete hint for arguments
disable-model-invocation	Prevent Claude from auto-invoking (for destructive actions)
user-invocable	Set false for background knowledge only
context	Set to fork for isolated subagent execution
agent	Subagent type (Explore, Plan, general-purpose)

Variable substitutions

Variable	Expands to
$ARGUMENTS	All arguments passed when invoking
$0, $1, $2, ...	Specific arguments by position
${CLAUDE_SESSION_ID}	Current session ID

Dynamic context injection

Shell commands can be executed at skill load time to inject dynamic context (e.g., current git branch, environment info).

Patterns included

1. Simple slash command — basic instruction skill
2. Command with arguments — positional argument handling
3. Background knowledge — non-invocable context (architecture docs, conventions)
4. Restricted tool access — read-only exploration skills
5. Isolated subagent task — forked context with dynamic command output
6. Multi-argument command — file patterns, migrations

Limitations

• Skills are a Claude Code feature — they don't work outside of Claude Code
• The SKILL.md file must be under 500 lines; use supporting files for longer content
• Dynamic context injection runs at skill load time, not at invocation time
• Variable substitution syntax uses $ARGUMENTS / $0 etc. — these are Claude Code-specific, not shell variables

License

MIT