skill-creator

Create a Claude Skill by building a .claude/skills/my-skill/ directory with a SKILL.md file containing YAML frontmatter (name, description) and core instructions. Use progressive disclosure to structure information across three levels: Level 1 metadata, Level 2 instructions, and Level 3 resources, keeping the core under ~5k tokens. Test in a conversation to validate it triggers correctly.

Progressive disclosure loads information on-demand rather than all at once, keeping core instructions lean while detailed content stays in optional references, scripts, or assets. This approach improves efficiency by reducing context overhead and letting Claude access domain expertise only when needed during conversations.

Codify project patterns, API workflows, database schemas, or framework conventions directly into your Skill's SKILL.md instructions and optional bundled content. Structure repeating context—like frequent explanations or configuration rules—as self-contained skills that Claude can reference automatically without re-explaining each time.

Yes. Beyond the required SKILL.md, add optional references/, scripts/, or assets/ directories for detailed content. Scripts must behave deterministically and never load code into context; references store supplementary material that progressive disclosure can retrieve when needed.

YAML frontmatter fields (name and description) have defined character limits. The SKILL.md body is constrained to approximately 5,000 tokens to maintain efficiency while capturing essential domain logic and instructions.

Test your Skill by using it in real Claude conversations to confirm it triggers correctly and responds as intended. Iterative refinement through actual usage ensures the workflow is testable, effective, and properly structured across its three-level architecture.