The Agent Skills Open Standard: SKILL.md Works Across Claude, Codex & Cursor

For a long time, if you wanted the same workflow instructions available to every AI coding agent, you maintained three separate files: CLAUDE.md for Claude Code, AGENTS.md for Codex, and .cursorrules for Cursor. Three copies, three formats, zero shared tooling. Every change meant editing the same guidance three times.

That fragmentation is now optional. The Agent Skills open standard — built around a single SKILL.md file — lets one skill work across roughly forty agents, including Claude Code, Codex, Gemini CLI, GitHub Copilot, VS Code, Goose, and OpenCode. Here's what the standard actually requires, where each tool looks, and how to keep one portable library instead of three.

Where the Standard Came From

Anthropic shipped Agent Skills across Claude apps, Claude Code, and its Developer Platform API on October 16, 2025. Nine weeks later, on December 18, 2025, it published the format as an independent open standard at agentskills.io — following the same playbook it used for MCP, which was donated to the Linux Foundation on December 9, 2025. Today governance sits under the Linux Foundation's Agentic AI Foundation.

Adoption moved fast. OpenAI and Microsoft added support within roughly 48 hours, Codex gained official SKILL.md support in December 2025, and Cursor followed with version 2.4 on January 22, 2026. By mid-2026 the agentskills.io showcase listed around forty products supporting the standard.

What the Format Requires

At its core, a skill is a folder containing a SKILL.md file: YAML frontmatter at the top plus a markdown body with the instructions. The spec requires only two frontmatter fields:

• name — 1–64 characters, lowercase, numbers and hyphens only, and it must match the parent directory name.
• description — 1–1024 characters that say both what the skill does and when to use it. This is the trigger.

Everything else is optional: license, compatibility, metadata, an allowed-tools list, and supporting scripts/, references/, and assets/ subdirectories. Keep the body under roughly 500 lines and ~5,000 tokens, pushing long material into referenced files.

Progressive Disclosure Keeps It Cheap

The standard is built around progressive disclosure: when an agent starts a session, it loads only each skill's name and description — roughly 50–100 tokens per skill. It reads the full SKILL.md body only when it decides the skill is relevant to the current task. That's why a skill library costs almost nothing to carry until you actually need a given skill. For more on how that differs from a big CLAUDE.md, see our guide to why AI agents are better with your SOPs.

Fair Warning: Discovery Paths Still Differ

The file format is shared, but where each agent looks is not. Don't assume zero-config portability.

• Claude Code reads .claude/skills/<name>/SKILL.md in the repo, and ~/.claude/skills/ for personal skills.
• Codex resolves .agents/skills/, searching from the current directory up to the repo root, then user and system paths.
• Cursor reads .cursor/skills/<name>/SKILL.md, added in version 2.4.
• Gemini CLI and others use their own expected locations.

The practical pattern is to keep one authoritative copy — Codex's .agents/skills/<name> is the common project-native path — and symlink or point each agent at it. Tools like the Vercel skills CLI install one source into any agent at project or global scope, with copy or symlink modes. Validate frontmatter before linking anywhere so a malformed name or description never drifts across mirrored directories.

The Portability Bet Is Winning

The direction of the market is clear: portability is winning. Anthropic's open SKILL.md format has become the de facto standard, and both OpenAI and Google support compatible formats. Cursor's marketplace plugins remain a proprietary holdout — a Cursor plugin only works inside Cursor, while a pure SKILL.md skill transfers. For teams not locked into one tool, that portability gap is the single biggest decision factor.

Write your skills in the open format and they work in Claude Code, Codex, and Gemini CLI today — and in whatever comes next. Lock-in is the enemy of good tooling, and the skill format is one place you can avoid it entirely. That's also why Claudia exports plain, portable SKILL.md files: a recorded workflow becomes a skill you can hand to whatever agent your team adopts, not a file hostage to one vendor.

Want to see the format up close? Read how to create a Claude Code skill step by step, then record once and automate forever across every agent that reads the standard.