skill-creator

---

name: skill-creator description: Guide for creating effective skills. This skill should be used when users want to create a new skill or update an existing skill that extends Claude with specialized knowledge, workflows, or tool integrations. license: See LICENSE.txt

Purpose

Provide clear, imperative guidance to design, initialize, validate, and package modular skills that extend Claude. Emphasize progressive disclosure, reusable resources, and high-quality metadata.

When to Use

Use when a user requests to create or improve a skill that bundles specialized workflows, tools, domain knowledge, or assets. Trigger when queries mention new skills, packaging skills, initialization, validation, or distribution.

How to Use

1. Understand concrete usage examples and confirm scope.
2. Plan reusable resources: scripts, references, assets.
3. Initialize a new skill directory using the initializer.
4. Edit SKILL.md in imperative style; keep references in separate files.
5. Validate and package into a distributable zip.

Initialize a New Skill

Run the initializer to scaffold a skill directory with required structure and a SKILL.md template.

Copyscripts/init_skill.py <skill-name> --path <output-directory>

Outputs:

• //
  • SKILL.md (with YAML frontmatter placeholders)
  • scripts/
  • references/
  • assets/

Edit the Skill

• Write SKILL.md in imperative/infinitive form; third-person phrasing.
• Keep detailed schemas, policies, and API docs in references/.
• Store templates, fonts, icons, boilerplate code in assets/.
• Add deterministic or frequently reused code to scripts/.

Validate and Package

Run the packager to validate structure and metadata, then create a distributable zip.

Copyscripts/package_skill.py <path/to/skill-folder> [./dist]

Validation checks (summary):

• SKILL.md exists with frontmatter containing name and description.
• Description is concise (≤120 words) and uses third-person (“This skill should be used when …”).
• Optional folders (scripts/, references/, assets/) are organized; references/ excludes binaries.

See references/validator_criteria.md for full criteria.

Progressive Disclosure

1. Metadata (name + description) — always loaded.
2. SKILL.md body — load when the skill triggers (<5k words).
3. Bundled resources — load or execute on demand.

Notes

• Keep SKILL.md lean; move details to references/.
• Include grep patterns in SKILL.md only when references are large.
• Maintain non-interactive defaults in scripts; provide --help.