What problem does it solve? Writing a Claude Code Skill that actually triggers reliably and follows Anthropic's progressive-disclosure guidelines is error-prone: descriptions under-trigger, SKILL.md files bloat past 500 lines, and nobody validates or tests the result. This Skill turns an idea into a validated, versioned skill package through a structured five-mode workflow. ## Core Features & Use Cases - Five operating modes: create (7-step gated pipeline with interview, eval design, scaffolding, validation, testing, versioning), refactor (analyze plus safe auto-fixes), validate, test, and version bumping. - Best-practices enforcement: bundled validator checks frontmatter constraints, third-person descriptions, kebab-case naming, the 500-line rule, nested references, Windows paths, and placeholder markers. - Evaluation-driven development: designs eval cases before writing SKILL.md, then runs them via a subagent harness and grades outcomes. - Bilingual workflow: interacts in Korean or English while always producing English skill artifacts with verbatim trigger phrases in both languages. - Use Case: A user says "make a skill that converts CSV to Markdown" — the agent interviews them, drafts eval cases, scaffolds the package from templates, validates it, tests it, and logs version 0.1.0. ## Quick Start Ask the agent to create a new skill for your task, for example by saying "create a skill that converts CSV files to Markdown tables", and it will guide you through the full pipeline.
