skill-creator
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's capabilities with specialized knowledge, workflows, or tool integrations.
Author
Category
Development ToolsInstall
Download and extract to your skills directory
Copy command and send to AI Agent for auto-install:
Skill Creator — Complete Guide to Creating Claude Skills
Skill Overview
Skill Creator is a complete guide to skill development that helps you create efficient, reusable skill packages for Claude, transforming a general-purpose AI into a specialized agent with domain expertise, customized workflows, and tool integrations.
Applicable Scenarios
When you want Claude to master specialized knowledge in a particular field or execute a customized workflow, use this guide to design and develop a skill package from scratch.
When an existing skill needs architectural optimization, new functionality, or bug fixes, refer to the iterative process in this guide for improvements.
When an organization needs to package internal knowledge, business processes, and API integrations into reusable skill assets for distribution and management across teams, use this guide.
Core Features
Provides a six-step development process covering requirements analysis, resource planning, template initialization, content creation, packaging, and distribution. Each step includes clear tools and validation methods to ensure that skills follow best practices.
Guides you in designing a three-layer loading system (metadata → SKILL.md → resource files), allowing Claude to load detailed content only when needed. This ensures complete functionality while optimizing context management.
Clearly defines the use cases and best practices for
scripts/ (executable code), references/ (documentation references), and assets/ (output resources), helping you properly organize reusable skill components and improve development efficiency and reusability.Frequently Asked Questions
How do I create a Claude skill from scratch?
First, clarify the skill’s specific use cases and functional requirements. Then use scripts/init_skill.py <skill-name> to generate the standard template. Next, populate the scripts/, references/, and assets/ directories according to your plan. Finally, write SKILL.md and use scripts/package_skill.py to validate and package the skill. The complete six-step process is detailed in the skill documentation.
What must the SKILL.md file contain?
SKILL.md must contain YAML frontmatter metadata (the name and description fields) as well as a Markdown body. The metadata determines when the skill is triggered. The body should include the skill’s purpose, applicable scenarios, and instructions for using its resources. The description should use an objective, third-person description and clearly explain “under what circumstances this skill should be used.”
What is the difference between the scripts/, references/, and assets/ directories?
These three directories store different types of reusable resources: scripts/ contains executable code (such as Python scripts) for tasks requiring deterministic reliability; references/ contains documentation and reference materials (such as API documentation and data structures) for Claude to load into context as needed; assets/ contains files used in outputs (such as templates, icons, and fonts), which are not loaded into context but are copied directly for use.
What is the progressive disclosure design principle?
Progressive disclosure is a context-management strategy for skills that divides information into three levels: metadata (name + description), which is always loaded and contains approximately 100 words; the SKILL.md body, which is loaded when the skill is triggered and contains approximately 5,000 words; and packaged resources, which are loaded as needed with no fixed limit. This ensures that Claude can use the skill correctly while avoiding unnecessary context consumption.
How do I validate and package a skill for distribution?
Use the command scripts/package_skill.py <skill-folder>. The script automatically validates the skill’s format, structure, and description quality. After validation succeeds, it generates a ZIP distribution package with the same name. If validation fails, it reports the specific errors; fix them and package the skill again. You can specify an output directory with scripts/package_skill.py <skill-folder> ./dist.
What reusable resources should a skill include?
Choose resources according to the specific needs of the skill: if the same code is repeatedly rewritten, place it in scripts/; if documentation or a knowledge base needs to be consulted, place it in references/; if the output requires templates or other materials, place them in assets/. The principle is to avoid duplicated work and context bloat by modularizing deterministic, reusable content.
How should I design a skill architecture suited to my business?
Analyze which knowledge, processes, and tools in your core business scenarios are used repeatedly, and abstract them into skill components. Give priority to storing business logic scripts in scripts/, domain knowledge documents in references/, and business templates in assets/. Use the progressive disclosure principle to design the information hierarchy, ensuring that Claude can find the right guidance when needed.
How should I iterate on and improve a skill after it has been created?
Observe the skill’s performance in actual use and record any problems or efficiency bottlenecks. Based on the feedback, update the guidance in SKILL.md and adjust the organization of the content in scripts/, references/, and assets/. Then validate and package the skill again. During iteration, keep the skill’s core purpose unchanged and focus on improving usability and efficiency.