skill-development

To create a Claude Code plugin skill, structure your project with a `SKILL.md` file, `scripts/`, `references/`, and `assets/` directories. This ensures modular, self-contained extensions for AI agents are discoverable and efficient.

Best practices for Claude plugin metadata involve using progressive disclosure, imperative third-person writing style, and organizing content into modular, self-contained references. High-quality metadata ensures skills are discoverable and effective.

Progressive disclosure in AI agent skill development is a best practice that structures documentation to reveal information sequentially. It prevents overwhelming the agent by organizing detailed content into `references/` rather than the main `SKILL.md`.

Organize Claude Code plugin documentation by placing core instructions in `SKILL.md` and detailed background content in `references/`. This structure facilitates progressive disclosure and keeps extensions self-contained.

Yes, you can use `scripts/` and `references/` components to build modular AI agent extensions. These directories keep your skill self-contained by separating executable logic and detailed documentation from the main metadata.

You should use an imperative, third-person writing style for Claude skill descriptions. Adhering to this style and structuring trigger phrases correctly ensures your AI agent processes requests efficiently.