skill-creator

To scaffold a new skill, run the initializer script to generate a SKILL.md file with required YAML frontmatter, create deterministic helper scripts for execution, and establish reference directories for deep details.

The best way to package skills for distribution is to enforce root containment and reject symlinks and traversal paths during the packaging workflow, ensuring safe and reliable delivery without directory escape vulnerabilities.

YAML frontmatter provides the trigger metadata, specifically the name and description fields, which enables reliable skill discovery and ensures proper runtime loading by keeping execution contracts concise.

Yes, you can generate optional UI metadata by creating an agents/openai.yaml file, which supports user-facing discovery and runtime loading while keeping the main execution contract lean in SKILL.md.

Moving deep details into references reduces runtime context costs and prevents prompt stuffing, keeping the execution contract in SKILL.md lean while maintaining reliable, discoverable, and dependable skill performance.

Yes, deterministic helper scripts are required for unstable workflow steps to guarantee reliable execution, standardize repeatable scaffolding, and reduce distribution mistakes across skill updates.