Skill Creator

Skills are modular packages that extend Claude's capabilities with specialized knowledge, workflows, and tools. They transform Claude from a general-purpose agent into a specialized one equipped with procedural knowledge.

Core Principle: Concise is Key

The context window is a public good. Only add context Claude doesn't already have. Challenge each piece: "Does this justify its token cost?"

Prefer concise examples over verbose explanations.

Installation

OpenClaw / Moltbot / Clawbot

bash

npx clawhub@latest install skill-creator

Skill Anatomy

code

skill-name/
├── SKILL.md (required)
│   ├── YAML frontmatter (name, description)
│   └── Markdown instructions
└── Bundled Resources (optional)
    ├── scripts/      - Executable code
    ├── references/   - Documentation loaded as needed
    └── assets/       - Files used in output

SKILL.md Frontmatter (required)

yaml

---
name: skill-name
description: |
  WHAT: One sentence describing what this skill does.
  WHEN: When to use it (contexts, user phrases).
  KEYWORDS: trigger phrases in quotes
---

The description is the only thing Claude reads to decide if the skill triggers. Be comprehensive.

Body (Markdown)

Instructions and guidance. Only loaded AFTER skill triggers. Keep under 500 lines.

Bundled Resources

scripts/

Executable code for deterministic or repeated tasks.

•Use when: Same code is rewritten repeatedly, or reliability is critical
•Example: scripts/rotate_pdf.py

references/

Documentation loaded into context as needed.

•Use when: Claude should reference while working
•Examples: schemas, API docs, domain knowledge, detailed guides
•If >10k words, include grep patterns in SKILL.md
•Keep information in EITHER SKILL.md OR references, not both

assets/

Files used in output, not loaded into context.

•Use when: Output needs these files
•Examples: templates, images, icons, boilerplate code

Creation Workflow

Step 1: Understand with Examples

Skip only if usage patterns are clearly understood.

Ask:

•"What functionality should this skill support?"
•"Give examples of how this skill would be used?"
•"What would a user say to trigger this skill?"

Step 2: Plan Reusable Contents

For each example, identify:

•What scripts would help? (repeated code)
•What references would help? (domain knowledge)
•What assets would help? (templates, files)

Step 3: Initialize the Skill

bash

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

Creates:

•SKILL.md template with proper frontmatter
•Example scripts/, references/, assets/ directories

Step 4: Edit the Skill

Implement resources first (scripts, references, assets). May require user input for domain-specific content.

Test scripts by running them to ensure no bugs.

Write SKILL.md with:

•Clear frontmatter description
•Workflow steps in imperative form ("Run X", not "You should run X")
•Links to references with when-to-read guidance

Step 5: Package the Skill

bash

scripts/package_skill.py <path/to/skill-folder>

Validates structure and creates .skill file for distribution.

Step 6: Iterate

Use the skill on real tasks. Notice struggles. Update and test again.

Progressive Disclosure Patterns

Keep SKILL.md lean. Split content when approaching 500 lines.

Pattern 1: High-level guide with references

markdown

## Advanced features
- **Form filling**: See [FORMS.md](FORMS.md)
- **API reference**: See [REFERENCE.md](REFERENCE.md)

Pattern 2: Domain-specific organization

code

bigquery-skill/
├── SKILL.md (overview and navigation)
└── references/
    ├── finance.md
    ├── sales.md
    └── product.md

Pattern 3: Framework/variant organization

code

cloud-deploy/
├── SKILL.md (workflow + selection)
└── references/
    ├── aws.md
    ├── gcp.md
    └── azure.md

Degrees of Freedom

Match specificity to task fragility:

Freedom Level	When to Use	Example
High (text)	Multiple valid approaches	General guidelines
Medium (pseudocode)	Preferred pattern exists	Scripts with parameters
Low (specific scripts)	Fragile, consistency critical	Exact command sequences

Quality Checklist

• Description includes WHAT, WHEN, KEYWORDS
• SKILL.md under 500 lines
• References are one level deep (not nested)
• Long references have table of contents
• Scripts are tested and work
• No README.md, CHANGELOG.md, or extra documentation
• Information lives in ONE place (SKILL.md or references, not both)

NEVER

•Create README.md, INSTALLATION_GUIDE.md, CHANGELOG.md, or auxiliary docs
•Duplicate information between SKILL.md and references
•Include "When to Use" sections in the body (put in description)
•Create deeply nested reference hierarchies
•Add scripts without testing them
•Exceed 500 lines in SKILL.md