skill-creator_skill

Overview

This skill designs effective Claude Code skills with concise descriptions, progressive disclosure, and error-prevention patterns. It guides freedom-level choices, token-efficiency techniques, and quality standards so skills are discoverable and reliable. Use it to scaffold, audit, or optimize skills for production use.

How this skill works

It inspects skill metadata, description length and discovery keywords, file layout, and resource layering to recommend changes. It enforces progressive disclosure: minimal metadata, on-demand full docs, and optional deep references. It also analyzes token usage and suggests pruning, relocation to references, or template tightening to reduce context footprint.

When to use it

• Creating a new Claude Code skill from scratch
• Improving an existing skill's description to increase discovery
• Optimizing token usage for the conversation context window
• Structuring long documentation into layers (metadata, main docs, references)
• Debugging why a skill isn't auto-triggering or being discovered

Best practices

• Write a 250–350 character description with two short paragraphs: what it builds and when to use it
• Be ruthlessly concise—move examples and long guides to references or assets
• Choose freedom level per task: high (advice), medium (patterns), low (exact scripts)
• Include 3–5 "Use when" scenarios and 2–3 distinctive error keywords for discoverability
• Test skills in a real project and validate code examples and package versions before marking production

Example use cases

• Scaffold a new skill via the create-skill command and generate required files and frontmatter
• Refactor an overlong skill doc: move heavy examples to references/ and shorten metadata
• Audit descriptions to add specific technologies and error keywords so Claude triggers the skill
• Compress token usage by replacing multiple examples with one canonical example and links to references
• Add scripts/ for deterministic tasks (migrations, scaffolding) and references/ for deep API docs