Guides the creation of new Agent Skills, including generating SKILL.md structure, descriptions, instructions, and optional supporting files/scripts...
This skill helps create high-quality, reusable Agent Skills for Claude Code.
Understand the Goal
Clarify what the new skill should do, key triggers (e.g., keywords like "PDF", "commit message"), and scope (keep it focused!).
Choose Skill Type and Location
~/.claude/skills/new-skill-name/ .claude/skills/new-skill-name/ (recommended for team sharing)Generate SKILL.md Structure
Use this template:
---
name: your-skill-name (lowercase, hyphens only)
description: What this skill does + when to use it (include triggers!). Keep under 1024 chars.
# Optional: allowed-tools: Read, Grep, Bash (etc., for restricted access)
---
# Skill Title
## Instructions
Detailed step-by-step guidance for Claude.
## Examples
Show 2-3 concrete before/after examples.
## Best Practices / Checklist
- Bullet points for quality standards
Add Supporting Files (Optional but Recommended)
reference.md: Detailed docs, schemas, APIs examples/ folder: Sample inputs/outputs scripts/ folder: Helper Python/bash scripts (e.g., data processors) templates/ folder: File templatesReference them like: See reference.md for details.
Make Description Trigger-Happy
Good example: "Analyze React components for performance and accessibility. Use when reviewing UI code, optimizing frontend, or mentioning React/JSX."
Bad: "Helps with code."
Test the Skill
Optional: Generate Files Automatically
Use the helper script below to scaffold a new skill directory.
Usage: python scripts/scaffold_skill.py 'Skill Name' 'Brief description with triggers'
Run it via Bash tool when needed: python scripts/scaffold_skill.py "Commit Message Generator" "Generate conventional commit messages from git diffs. Use when creating commits or reviewing changes."
allowed-tools for safe/read-only skills