skill-creator

Effective skill descriptions balance clarity with token efficiency by front-loading the core task, including "Use when" scenarios, and embedding error keywords that trigger automatic suggestions. Progressive disclosure—layering detail from essential to optional—ensures Claude recognizes the skill's purpose without bloating metadata.

Token optimization for skill metadata eliminates redundancy by separating tagline (quick hook), core task (precise function), and full description (layered scenarios). Concise YAML frontmatter and strategic keyword placement maximize context efficiency while maintaining discoverability signals.

Design skills using progressive disclosure: craft a tagline that states the benefit, write a core task description that includes specific use cases and trigger keywords, and structure the full description with problem statement, features, and quick-start examples. Claude matches these signals to user queries automatically.

Check that your skill's keywords, core task description, and tagline directly address the problem users face and include error scenarios that trigger suggestions. Audit for token waste—redundant explanations, component names, or structural vocabulary—and refactor using progressive disclosure to sharpen discovery signals.

Skills fail discovery when descriptions bury the core task, use vague language instead of task verbs, omit error keywords and "Use when" scenarios, or lack progressive disclosure. Overly long metadata and self-promotional framing also waste tokens and obscure the functional purpose Claude uses to match suggestions.

Structure skill content in layers: tagline states the outcome, keywords target discovery, core task describes exact function and use cases, and full description adds context only when needed. This constraint-aware design keeps metadata lean while ensuring Claude has enough signal to recognize when the skill applies.