Agent Skill Manager & Creator

Welcome to the Agent Skill Manager. This skill outlines the core philosophy and comprehensive guiding principles for building, editing, and maintaining high-quality, efficient, and modular Agent Skills.

🧠 Core Philosophy: Simplicity & Progressive Disclosure

When creating or refactoring a skill, you must first assess its complexity.

• For simple skills: A single SKILL.md file is entirely sufficient. Do not create unnecessary folders or files if the core logic and instructions are brief, straightforward, and easily fit within the agent's context window.
• For complex skills: Your critical responsibility is designing for Progressive Disclosure. You must proactively split complex content into multiple files to manage the target agent's context window efficiently. Do not overload the main skill file.

Progressive Disclosure Workflow (For Complex Skills)

1. Discovery: Only the name and description (YAML frontmatter) are loaded at the target agent's startup. Keep them keyword-rich and highly descriptive so the agent knows exactly when to use the skill.
2. Activation: The newly created SKILL.md file is read in its entirety upon task activation. It must remain strictly under 500 lines. Focus this file only on what the skill does, high-level rules, and providing a map to auxiliary files.
3. Execution (Optional File Splitting): When a skill requires complex logic, long text, or templates, structurally split them into optional subdirectories so the target agent loads them only on demand:
   • references/: Extract detailed technical docs, step-by-step workflows, API references, or long-form domain instructions into highly focused, small markdown files here.
   • assets/: Isolate static resources like document templates, config templates, or data schemas here.
   • scripts/: Move executable code (e.g., Python or Bash) here. Scripts must be self-contained, provide clear error messages, and handle edge cases gracefully.

🏗️ Structural Principles