Introduction

Summary

A skill is a reusable instruction package for an AI agent. The same idea can be shared across models, but each platform may require a different wrapper format.

Core purpose: encode repeatable workflow, domain rules, and reusable resources.
In Codex, a skill is a folder with required SKILL.md (YAML frontmatter with name and description) plus optional scripts/, references/, and assets/.
Skills are partially portable across models: reusable content can be shared, but trigger metadata and loading behavior are usually platform-specific.
Best portability pattern:
1. Keep one model-agnostic core doc in references/.
2. Maintain small per-platform adapter files (SKILL.md, Claude/Gemini instruction wrappers).
Read a skill in this order: description trigger intent → workflow steps → referenced files → scripts.
Edit safely: change small sections, test one real task, then iterate.

Typical Codex skill structure: my-skill/ ├── SKILL.md ├── scripts/ ├── references/ └── assets/
Minimal SKILL.md frontmatter: --- name: my-skill description: Use when ... ---
Cross-model sharing approach:
- Shared: references/workflow.md, scripts/*.py
- Codex wrapper: SKILL.md
- Other model wrapper: tool-specific instruction file pointing to same references/scripts