Agent Skills
This is an OpenAI-specific overview. For the complete deep-dive — including the SKILL.md format, skills vs tools vs MCP, sub-agent orchestration, and the full ecosystem — see Agent Skills — Full Deep Dive.
OpenAI’s SKILL.md Support
OpenAI adopted the SKILL.md open standard within 48 hours of Anthropic’s December 2025 release. Codex, the Agents SDK, and ChatGPT all read the same SKILL.md files from the same directory layout.
# Same format works across Codex, Claude Code, and Cursor.codex/skills/ # Codex.claude/skills/ # Claude Code.cursor/skills/ # CursorCodex Skills
---name: review-securitydescription: Audit code for OWASP Top 10 vulnerabilities---
## Process1. Scan changed files for common vulnerability patterns2. For each finding, assess severity (Critical/High/Medium/Low)3. Provide CWE reference and remediation code4. Output as a GitHub issue with labelsAgents SDK Skills
from agents import Agent, Skill
agent = Agent( name="Security Auditor", model="gpt-5.4-mini", skills=[Skill.from_file(".codex/skills/review-security/SKILL.md")])Portability
A skill written for Claude Code works in Codex and vice versa. The 32+ tools supporting SKILL.md include:
- Codex, Claude Code, Cursor, GitHub Copilot, VS Code
- Windsurf, Aider, JetBrains Junie, AWS Kiro, Amp
- Google Gemini CLI, ChatGPT
Going Deeper
The full Agent Skills deep dive covers:
- The complete SKILL.md format with all sections
- Advanced examples with scripts and assets
- Sub-agents and orchestration patterns
- When to use skills vs MCP vs function calling