Use Claude-style skills (SKILL.md) with any AI model

Skills are one of the most useful ideas in AI agents: instead of pasting the same long instructions into every chat, you write them once as a file, and the agent loads them when a task calls for them. Claude popularised the format - a folder with a SKILL.md file in it - and plenty of teams now keep their skills in GitHub repos.

The good news: nothing in that format is tied to one model. Here's how skills work, how to write good ones, and how to use them with GPT, Gemini or open models.

What a skill is

A skill is a folder with a SKILL.md file. The file starts with a short header (YAML front matter) and then the instructions:

• name - a short id, like "landing-page".
• description - one or two sentences on when to use it. This is the most important line in the file.
• the body - the actual instructions: steps, rules, examples, a checklist.

The folder can also hold reference files the instructions point to - a style guide, a template, example output.

Why skills work so well

The agent doesn't read every skill all the time. It sees only each skill's name and description - a few words each - and opens the full instructions when a task matches. That keeps the prompt small, so you can have dozens of skills without slowing every answer down, and the agent still follows the detailed playbook when it matters.

Why any model can use them

A skill is plain text. Any model that can follow instructions can follow a skill, as long as the tool around it does two things: tells the model which skills exist, and lets it load one when needed. Open models like GPT-OSS and Qwen, Gemini, Mistral and GPT models can all do that through ordinary tool calls.

Two honest caveats:

• Smaller models follow long skills less reliably. Keep skills short and concrete, and put the most important rules first.
• Some skills include scripts. Running them needs a sandbox; a tool that only reads skills will use the instructions and reference files, not run the code.

Skills worth writing for a team

• Brand voice: how you write, words you never use, three short examples.
• Landing page or UI style: your colours, type scale, spacing and components, so every page the agent drafts looks like yours.
• Code review checklist: what your team always checks before merging.
• Client report: the structure, tone and sections you always send.
• Hackathon pitch: a two-minute script structure and a slide outline.

How to write a good one

• Make the description about when, not what. "Use when drafting or editing a landing page or marketing page for our product" beats "Landing page guidelines".
• Lead with rules, then examples. Models copy examples closely, so make them good.
• One job per skill. Two short skills beat one that tries to do everything.
• Keep it in GitHub. The team can review changes like code, and everyone's agent gets the update.

A word on safety

A skill is instructions your agent will follow. Only add skills from repositories you trust, and read a skill before adding it - the same way you'd read a script before running it.

How Teamski runs skills on any model

In Teamski, an owner or admin pastes a GitHub repo link in Settings → Skills, and every SKILL.md in it is added to the project. Each agent sees the skills' names and descriptions and reads a skill - and any file it mentions - when a task matches, whichever model is answering: the free built-in model, your own keys, or key rotation. Skills update from GitHub about once a day, and private repos work through your GitHub connection. See also AI agents for teams.