Create New Skills

This skill acts as a comprehensive guide and generator for developing custom Agent Skills within the Claude Code ecosystem. It is designed for engineers, developers, and power users who want to extend Claude's capabilities by packaging domain-specific expertise, workflows, and tools into modular, reusable components. By following this skill's structured creation workflow, you ensure that your agents remain context-efficient through progressive disclosure, only loading relevant metadata, instructions, and resources when specifically triggered.

• Standardized structure: Enforces a consistent directory layout using .claude/skills/ for project-level or local sharing via git.

• Multi-file support: Guides the integration of diverse file types, including .md for documentation, .sh/.py for executable logic, .yaml/.json for configuration, and SQL for database operations.

• Progressive disclosure architecture: Implements a three-level loading strategy (metadata, instructions, resources) to optimize the LLM context window.

• Best practice enforcement: Provides step-by-step guidance on crafting SKILL.md files with proper YAML frontmatter, including name, description, and tool restriction definitions.

• Pattern library: Integrates with reference documentation like multi-file patterns, agent skill architecture, and design principles to ensure high-quality, maintainable code.

• Use this skill whenever you need to automate repetitive tasks or embed specific expert workflows directly into your agent's environment.

• Always define clear triggers in your frontmatter so Claude knows exactly when to activate your custom skill.

• Keep instructions actionable and numbered; prioritize atomic, modular designs over monolithic scripts.

• Utilize the resource-loading feature for complex logic, keeping your main instruction file clean and focused on guidance rather than execution code.

• Ensure all scripts are set to executable (chmod +x) and follow the recommended project structure for seamless integration with team-shared repositories.