Notion Rolls Out Team-Wide Skills for AI Agents
Notion has officially launched Skills for teams: your best work turns into a skill that an agent loads automatically. And thanks to MCP, that same skill can be used by Codex and Claude Code too.
The idea is simple. Feeding context to an agent is nothing new — system prompts, AGENTS.md files, rule files tucked into a project. The real problem is different: every tool and every person ends up with their own copy of these rules, and over time they drift apart. A skill packages that knowledge once. After that, the agent loads it automatically whenever it's relevant, and works the way your team already does things.
Here's the mechanic, straight from the announcement: ask Notion Agent to turn your team's best example of a piece of work into a skill, then share it with colleagues. The skill isn't locked inside Notion, either — any external agent can connect to Notion via MCP (Model Context Protocol, an open standard for exchanging context) and pull the instructions it needs. One set of skills lives in Notion, but it runs in Codex, Claude Code, or any other MCP-compatible agent.
This is actually how I've set up the writing instructions for this blog. They live in Notion, and Codex pulls them in for every new task. I edit them like regular pages, and the agent always works from the current version.
The practical value here is that team context stops living in people's heads and pinned messages. It becomes an artifact you can discuss and refine. It also lowers the barrier for new hires: an agent equipped with a team skill already knows how things are done around here.
Where to start today: pick one process your team does regularly and does well. Drafting posts, triaging inbound requests, writing up meeting notes, drafting newsletters — anything like that. Ask an agent to turn your best example into a skill, then test it on the next real task. If quality holds up, start packaging the rest of your processes the same way.
If your team is already working with AI agents but the instructions are scattered across people's heads and scratch notes, team skills are a solid excuse to finally turn that knowledge into a system. It's especially useful for anyone who owns a process and needs consistent, predictable output from an agent.
As more teams package their know-how into shareable, portable skills, the gap between "an agent that follows generic instructions" and "an agent that works like your best person" is going to shrink fast — and it's worth getting ahead of that shift now rather than later.