What Is the Agent Skills Open Standard?

SKILL.md started as Anthropic's format for Claude Code. It's now an open standard called Agent Skills, adopted by multiple AI coding tools including OpenClaw (247K+ GitHub stars), Codex CLI, Cursor, and Gemini CLI. This matters because it means skills are portable: build once, use across agents.

What the standard defines

The Agent Skills specification covers the file structure (a directory with a SKILL.md file), frontmatter format (YAML with required name and description fields, plus optional fields), instruction format (markdown below the frontmatter), supporting files (scripts, references, assets directories), and progressive disclosure (metadata loaded first, full instructions loaded on demand).

The standard is intentionally minimal. It defines enough structure for skills to be discoverable and portable, without constraining what agents can do with them.

Which tools support it

As of 2026, the following tools support the SKILL.md format: Claude Code (Anthropic), Codex CLI (OpenAI), Gemini CLI (Google), GitHub Copilot (via VS Code agent skills), Cursor (with manual placement), and several community tools like Cline, Windsurf, and OpenCode.

Each tool may extend the standard with additional features (Claude Code adds context forking, Codex adds openai.yaml metadata), but the core SKILL.md format works across all of them.

Why an open standard matters

Before the Agent Skills standard, every tool had its own customization approach. Cursor used .cursorrules, Claude Code used /commands, Copilot used instructions files. If you switched tools or used multiple tools, your customizations didn't transfer.

The open standard changes this. A skill written for Claude Code can be copied into Codex's skills directory and it works. A team can standardize on SKILL.md for their conventions and every developer uses them regardless of which agent they prefer.

For marketplace creators, this is even more significant. A skill listed on Agensi is compatible with every major AI coding tool. That's a much larger market than building for one tool alone.

What's not standardized

Some features remain tool-specific. Claude Code's context: fork for subagent execution doesn't work in Codex. Codex's openai.yaml metadata isn't recognized by Claude Code. Cursor doesn't have native skill discovery (you invoke them manually).

In practice, most skills that stick to the core format (SKILL.md with standard frontmatter and markdown instructions) work everywhere without modification. Skills that use advanced features may need minor adjustments for cross-tool compatibility.

Where to learn more

The specification is published at agentskills.io. Anthropic's reference implementation and example skills are on GitHub.

For a practical introduction, read our complete guide to SKILL.md. For the technical reference, see the SKILL.md Format Reference. Browse portable skills on the Agensi marketplace.