Agent Skills as an Open Standard: What It Means if Your Team Uses ChatGPT or Gemini

Agent Skills began as a Claude feature and became an open standard in December 2025, published at agentskills.io. A skill is a folder with a SKILL.md file of instructions that any compatible AI agent can load. For teams using several assistants, it means a method written once can travel between tools, though support still varies by product.

What is the Agent Skills open standard?

It is a published, vendor-neutral format for packaging instructions an AI agent can load and follow. Anthropic introduced skills for Claude in October 2025, then released the format as an open standard in December 2025. The standard's site describes skills as "a lightweight, open format for extending AI agent capabilities".

Openness matters because a skill encodes how your business does something. If that method only works in one vendor's product, switching tools means rewriting it. With an open format, a skill is closer to a document your business owns that several tools can read.

The move follows a familiar pattern. Anthropic did the same with the Model Context Protocol, introducing it in 2024 and donating it to the Linux Foundation's Agentic AI Foundation in December 2025.

What is in the Agent Skills specification?

The specification is short. A skill is a folder containing a required SKILL.md file with metadata, at minimum a name and a description, followed by instructions. The folder can also hold scripts, reference documents and templates. Agents read only the name and description at first, then load the rest when a task matches.

According to agentskills.io, loading happens in three stages:

1. Discovery: the agent reads each skill's name and description.
2. Activation: when a task matches, it reads the full SKILL.md.
3. Execution: it follows the instructions, using bundled files or scripts as needed.

That design lets an agent keep many skills available without filling its working memory. Simon Willison, a widely read developer, called the specification "deliciously tiny", and also noted it is "quite heavily under-specified" in places, such as how extra metadata should be named.

Which tools support Agent Skills?

Support has grown quickly but varies by product. According to Simon Willison's December 2025 write-up, the standard's site listed adoption by OpenCode, Cursor, Amp, Letta, goose, GitHub and VS Code, and OpenAI added skills to its Codex documentation shortly after. Anthropic supports skills across Claude's apps, Claude Code and its API.

Two cautions apply. First, support in a developer tool does not always mean support in the same vendor's chat product for everyday staff. Second, features differ: some tools support scripts inside skills, others only instructions. Check the current documentation for each assistant your team uses, and test a real skill before relying on it.

What does it mean if your team uses ChatGPT or Gemini?

It means your methods can be written in a format that is not tied to one assistant. Where your tools support the standard, the same skill folder can work across them. Where they do not yet, the SKILL.md file is still clear, readable instructions that can be adapted as custom instructions or project guidance.

For mixed teams, the practical benefit is a single source for each method. Instead of maintaining a proposal method in a custom GPT, another in a Claude project and a third in someone's notes, the business keeps one skill folder and points each tool at it as support allows.

The deeper benefit is ownership. According to agentskills.io, skills package "procedural knowledge and company-, team-, and user-specific context into portable, version-controlled folders". Kept in files the business controls, they outlast any one vendor relationship.

How do you write skills that travel between tools?

Keep each skill focused on one job, put the essential method in plain Markdown in SKILL.md, keep the description specific about when to use it, include an example of good output, and avoid depending on features only one product offers. Store skills in a shared, versioned folder the business owns.

A short checklist:

• One job per skill. Narrow skills are easier to test and move.
• Plain instructions first. Scripts are powerful but less portable.
• A precise description. It decides when an agent uses the skill.
• An example. One good output teaches more than rules.
• Owned storage. A shared, version-controlled folder, not one person's account.

The guide to writing a Claude skill covers each step.

Skills also work best on top of shared facts, so the method has accurate information to apply, which is what a company brain holds.

Skein writes client skills in the open format and keeps them in files the client owns; the skill library lists hundreds of open skills by function. That suits a team using more than one AI assistant, or one that wants its methods to outlast its tools. A team committed to a single assistant still benefits, mainly from the ownership.

Frequently asked questions

What is the Agent Skills standard?

It is an open, vendor-neutral format for packaging instructions an AI agent can load and follow. A skill is a folder with a SKILL.md file containing a name, description and instructions, plus optional scripts and resources. Anthropic published it as an open standard in December 2025 at agentskills.io.

Do Agent Skills work with ChatGPT?

Support is growing but varies by product. According to Simon Willison's December 2025 write-up, OpenAI added skills to its Codex documentation. Check current documentation for the specific ChatGPT product your team uses, and test a real skill before relying on it.

Who created Agent Skills?

Anthropic developed Agent Skills for Claude and introduced them in October 2025, then released the format as an open standard in December 2025. The standard's site invites contributions from the wider ecosystem, and a range of agent tools have adopted it.

Why does an open standard for skills matter?

Skills encode how a business does its work. An open format means those methods are not locked into one vendor's product, so they can travel between tools and outlast any single supplier. Kept in files the business owns, they become a lasting asset.

How is a skill different from a custom GPT?

A custom GPT bundles instructions and knowledge into one assistant inside ChatGPT. A skill is a portable folder of instructions that compatible agents load when a task matches, and it follows an open standard. Skills are designed to be reused across tools; a custom GPT stays in one product.

Related reading