Skill Creator

Create high-quality custom Skills for Claude Code.

Core Principles

Concise is Key — Only add information Claude doesn't already know; every piece must justify its token cost
Appropriate Freedom — High freedom uses text guidance, low freedom uses specific scripts
Progressive Loading — Metadata always loaded, body loaded on trigger, resources loaded on demand

Creation Workflow

1. Define Use Cases → Gather concrete scenarios and trigger conditions
2. Plan Contents    → Identify needed scripts, references, and assets
3. Initialize       → Run scripts/init_skill.py
4. Write Content    → Complete SKILL.md and bundled resources
5. Validate & Pack  → Run scripts/package_skill.py

Skill Structure

my-skill/
├── SKILL.md           # Required: Main file
├── scripts/           # Optional: Executable scripts
├── references/        # Optional: Reference docs (loaded into context on demand)
└── assets/            # Optional: Templates, images, etc. (not loaded into context)

SKILL.md Specification

Frontmatter (Required)

---
name: my-skill              # kebab-case, max 64 characters
description: Concise description of functionality and trigger scenarios. Must include WHEN to use.
---

Body Structure Patterns

| Pattern | Best For | Example | |---------|----------|---------| | Workflow-based | Sequential processes | PDF form filling, document approval | | Task-based | Multiple independent operations | Image processing (crop/rotate/compress) | | Reference-based | Standards or specifications | Brand guidelines, coding standards | | Capability-based | Integrated systems | Product management, data analysis platform |

Scripts

Initialize New Skill

python scripts/init_skill.py <skill-name> --path <output-dir>

# Example
python scripts/init_skill.py my-analyzer --path ~/skills

Validate and Package

python scripts/package_skill.py <skill-folder> [output-dir]

# Example
python scripts/package_skill.py ~/skills/my-analyzer ./dist

Quality Checklist

[ ] Description includes trigger conditions (WHEN to use)
[ ] SKILL.md body < 500 lines
[ ] No redundant docs (README, CHANGELOG, etc.)
[ ] Scripts tested and working
[ ] Large reference docs split out with guidance on when to read them

Common Patterns Reference

See references/patterns.md for:

Sequential workflow pattern
Conditional branching pattern
Output template pattern
Example-driven pattern