name: skill-creator description: Create Claude skills following best practices. Use when building new skills, packaging skill folders, or improving existing skills. Triggers on requests to create skills, make skills, build capabilities, or package skill directories. allowed-tools: Bash, Read, Write, Edit, Task

Skill Creator

Create effective, token-efficient Claude skills.

Core Principles

1. Concise - Challenge every token. Only add what Claude doesn't already know.
2. Appropriate freedom - Match specificity to task fragility (see references/workflows.md)
3. Progressive disclosure - Metadata → SKILL.md → references (loaded as needed)
4. Favor subagents - Design skills to use Task tool for parallel work where beneficial

Skill Structure

skill-name/
├── SKILL.md              # Required: frontmatter + instructions
├── scripts/              # Optional: executable code
├── references/           # Optional: documentation loaded as needed
└── assets/               # Optional: files used in output

Creation Workflow

1. Understand - Gather concrete usage examples (use subagents for parallel research)
2. Architect - Use skill-architect to design subagents for the skill
3. Plan - Identify reusable resources (scripts/references/assets)
4. Initialize - Run scripts/init_skill.py <name> --path <dir>
5. Implement - Create resources, write SKILL.md with agent definitions
6. Validate - Run scripts/validate_skill.py <skill-dir>
7. Register - Add skill to .claude/claude.md under "Auto-Triggered Skills"
8. Package - Run scripts/package_skill.py <skill-dir> (creates .skill file)
9. Iterate - Improve based on real usage

Writing SKILL.md

Frontmatter (Critical)

---
name: lowercase-hyphenated
description: What it does AND when to trigger. Include specific triggers, keywords, use cases. This is the PRIMARY triggering mechanism.
allowed-tools: Read, Edit, Write, Task  # Include Task for subagent support
---

The description must contain ALL "when to use" info - body is only loaded AFTER triggering. Include Task in allowed-tools to enable subagent parallelization.

Body Guidelines

• Keep under 500 lines; split to references if longer
• Use imperative form ("Run the script" not "You should run")
• For multi-step processes, give overview early
• For output formats, use templates or examples (see references/output-patterns.md)

Resource Types

Do NOT Include

• README.md, CHANGELOG.md, INSTALLATION_GUIDE.md
• Auxiliary documentation not needed by Claude
• Duplicate info (lives in SKILL.md OR references, not both)

Validation Checklist

• name and description in frontmatter
• Description includes what + when to trigger
• SKILL.md under 500 lines
• Scripts tested and working
• No extraneous files
• References cited in SKILL.md

See references/workflows.md for workflow patterns and references/output-patterns.md for output formatting.