Notion is turning one of the most useful parts of working with AI—figuring out the right instructions—into something you can actually package and reuse.

With Notion 3.7, the company has introduced AI Skills, reusable instruction sets that teach Notion Agent how to perform a particular type of work. The bigger idea is that a good prompt no longer has to remain buried in an old chat or trapped in one employee’s head. It can become a shared, maintainable workflow that an entire team can use, and even take with them to other AI agents.

What are Notion AI Skills?

A Skill is essentially a Notion page containing instructions for how an AI agent should handle a specific kind of task.

That could be something simple, such as proofreading or reformatting text, or something much more specific to a company’s workflow. Notion gives examples including reviewing feature proposals against product, design and engineering standards, creating new-hire checklists according to company policies, and drafting job descriptions using an organization’s approved language and leveling framework.

The important distinction is that a Skill is designed around a repeatable process rather than being just another one-off prompt. You can define the goal, the steps, standards, examples and desired output, giving the agent a consistent set of instructions to follow.

And because the Skill is a Notion page, it remains part of the workspace rather than becoming another isolated AI configuration.

How to build a Skill in Notion

Notion says you don’t have to start with a blank page and manually construct everything.

Open a Notion Agent chat, type / and select the + option to create a new Skill. You can also simply tell Notion Agent that you want to create a Skill, after which the agent will guide you through the process.

The easiest way to build a useful Skill is to describe how you already perform the task. If you normally give an AI a particular piece of context, follow a certain sequence of steps, apply specific standards or expect a particular format, include those details when creating the Skill.

Notion can also help identify potential Skills from work you’re already doing. That makes the feature less about becoming a prompt engineer and more about turning existing workflows into reusable instructions.

You can also turn an existing Notion page into a Skill, which is particularly useful for teams that already have documentation describing how certain work should be done.

The best Skills tend to focus on one type of work. Instead of creating one enormous instruction set that tries to handle everything, a team could have separate Skills for turning customer interviews into research summaries, preparing executive briefs, reviewing product proposals or formatting campaign copy.

How to use a Skill

Once you’ve created or enabled a Skill, using it is pretty straightforward.

In a Notion Agent chat, type / and select the Skill you want. You can then provide additional context before running it. Notion also allows Skills to be used automatically when the agent determines that one is relevant to the request.

Skills can also be used directly while working on a Notion page. Highlight the content you want to change or review, open the editor menu and select the appropriate Skill.

For teams using Custom Agents, a Skill can become one component of a larger automated workflow. Notion says a Custom Agent can be given access to a Skill page through its tools and access settings, with instructions telling the agent when that Skill should be used.

That means a Skill doesn’t necessarily have to be the entire job. It can be one standardized step inside a much larger agent workflow.

How to share Skills with your team

This is where the feature starts becoming more interesting for organizations.

Notion automatically places newly created Skills into a private Skills database. As a team builds more of them, Skills can be organized into databases based on the team’s needs. A marketing group, for example, could maintain a collection of writing, campaign and launch Skills in one shared location.

Because Skills are essentially Notion pages, they use Notion’s existing sharing and permission system.

You can share an individual Skill when only certain people need a particular workflow, or share an entire Skills database when you want a team to have access to the full collection. People with access can run the Skill, while users with editing permissions can modify the underlying instructions.

That also gives teams a single place to maintain their AI workflows. Instead of five employees using five slightly different versions of the same prompt, the team can maintain one Skill and update it as its process changes.

How to export a Notion Skill

Perhaps the most interesting part of the update is that Notion isn’t keeping these Skills locked inside Notion.

A Skill can be downloaded and used with supported local AI agents, including Claude Code, Codex, Cursor, Gemini and Grok. Notion exports the instructions as a SKILL.md file along with approved supporting files.

That effectively turns Notion into a place where a team can maintain its AI operating instructions while still using different AI tools elsewhere.

For example, a company could create a code-review Skill in Notion, maintain the official version there, and then export it for use with its preferred coding agents. A content team could maintain its editorial guidelines as a Skill and take that workflow into a supported AI environment outside Notion.

There is one important catch: downloaded Skills don’t automatically stay synchronized with the Notion version. If the Skill is changed in Notion, the downloaded copy needs to be downloaded again. Notion adds a badge to indicate that a newer version is available.

So Notion is acting more like the source of truth than a live synchronization layer for exported Skills.

Why this matters

The interesting thing about Notion’s approach isn’t simply that it has added another AI feature. It’s trying to solve a problem that has become increasingly obvious as companies adopt AI agents: the valuable part isn’t always the model itself. It’s knowing exactly how you want that model to perform a recurring task.

A great workflow often consists of dozens of little decisions—what information to look at, what standards to apply, what to ignore, how to structure the output and what should happen next. Traditionally, that knowledge gets passed around through documentation, Slack messages, training sessions and increasingly long prompts.

Skills give that knowledge a more structured home.

And exporting those instructions as SKILL.md files makes the idea considerably more portable. A team doesn’t necessarily have to choose between keeping its documented workflow in Notion and using another AI agent. It can maintain the instructions in one place and bring copies of those instructions into other supported tools.

That could make Skills particularly useful as AI agents become less like chatbots and more like coworkers handling repeatable processes.

Notion 3.7 also adds several other agent-focused features, including Custom Agent sub-agents, automated post-meeting workflows, Custom MCP connections for tools such as GitHub and Amplitude, and a public beta of its Agent SDK. But Skills may be the piece that quietly becomes useful every day: write down how the work should be done once, then let different agents follow the same playbook.