Notion + Claude Code: Build AI Skills Businesses Pay For

Author: Arseny Shatokhin · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Notion + Claude Code: Build AI Skills Businesses Pay For · Channel: Arseny Shatokhin · 12:28 · Watch on YouTube

Topics: AI agents, Enterprise AI applications, Task and workflow automation, Claude Code, AI developer tools

Summary

The video demonstrates how Notion’s shared AI skills library can help teams capture process knowledge and make updated skills available across AI agents. The creator builds a script-writing skill from interviews, writing documents, and real examples from a video production agency, then connects the Notion library to GitHub and Claude for distribution and syncing. Feedback from a later draft is added to the shared skill so future users can benefit from it. The video also presents shared skills as a small, practical service businesses can buy before considering a larger AI operating system installation.

Key points

Individual AI skills can become outdated when teams keep separate versions and leave feedback trapped in personal chats.
Notion pages let technical and non-technical team members review and edit shared skills.
A Notion skill can be installed into an agent with a command, but a direct install is only a snapshot.
The creator builds a script-writing skill using producer interviews, writing documents, frameworks, and published scripts.
Testing the skill on another project reveals feedback that can be incorporated into the shared process.
A scheduled GitHub sync distributes Notion skills to agents such as Claude; the demonstrated sync runs hourly.
Lessons from one draft can update a Notion skill and become available to the next person using it.
The creator suggests offering a small engagement covering two real workflows as an entry point to larger AI implementation work.

Chapters

Start	Chapter	Summary
00:00	Why teams need shared AI skills	The video introduces the problem of disconnected skill versions and shows how Notion pages can serve as a shared workspace for humans and agents.
02:05	From personal skills to team knowledge	The creator explains the limits of snapshot installs and GitHub-only workflows, and positions Notion as the team-facing source of truth.
04:10	Building a script-writing skill	Using interviews and real agency materials, the creator drafts a skill, tests it on another project, and identifies feedback to incorporate.
06:16	Syncing the skill through GitHub	The walkthrough connects Notion to a private GitHub skills repository and configures the skills for use in Claude.
08:22	Sharing lessons across projects and agents	A reviewed script informs an update to the shared skill, and the video describes another setup that syncs agent instructions from Notion into a deployed platform.
09:54	Offering shared skills as a business service	The creator recommends starting with a small engagement for real workflows, then using demonstrated value to propose a larger AI installation.

Quotes

AI skills are supposed to turn the way your people do their work into a process that any AI can follow. — 00:00

GitHub in this case just becomes the distribution layer on top of Notion. — 03:39

The lesson is now in the original Notion skill where it added this new step under the workflow section. — 08:53

A shared skills library is how you make the feedback and the experience from every project stay inside the business — 10:25

Next time you pitch a business, don’t start with a generic audit. — 11:57

Concepts and entities

AI skills, Notion, Notion AI, Claude Code, Codex, Notion MCP, GitHub, Cursor, Represent, biblioteca compartilhada de skills

Full transcript (original language)

00:00 AI skills are supposed to turn the way your people do their work into a process that any AI can follow. But as soon as your team has more than one person, this is where they usually break. One person improves the skill in their own chat, but someone else keeps using the older version. And all the feedback stays trapped inside separate chats across all the agents and all the team members that you have. And as soon as someone leaves, they take all that feedback and just leave with it, so your business doesn’t even improve. Notion just released a shared AI skills library that solves

00:31 this exact issue. Humans and agents can finally edit the same skills as any normal Notion pages, and those changes sync automatically into Cloud Code, Codex, and all the other agents that your team might use. I just sold this exact setup to a video production agency for $1,000. In this video, I’m going to show you how it works, then build their script writing skill from the real workspace and sync it across their whole team and agents that they use. But to show you first how easy it is to get started, let’s first install one of my own Notion skills into Cloud Code with a

01:03 single command. Okay, so here’s one of my skills that essentially writes copy for cold email campaigns. And if I want to use this page as a skill in Notion, I simply hit on three dots right here, and then under use with AI, I can hit use as AI skill. And then, I need to move the skill to a special skills database. To install it, I can just run the NPX skills latest add notion {dash} {dash} global. So, I’m installing it globally, so it’s available across all my projects. And

01:34 then I can hit enter, and I will see all my Notion skills from that database. So, I can just select email copywriter skill like this, and this will install it across all my projects for all my agents. So, then if I run Cloud Code, you can see that I have the skill now right here. So, then I can simply select the skill and just say, “Hey, create copy for this campaign. 9 minutes later, Claude says it’s finished and if I open the same campaign in Notion, the complete copy is already

02:05 there. So, it didn’t just leave me with a draft inside the terminal. It actually followed my skill, found the correct project, and then saved the result where my team actually works. But, this direct install is only a snapshot. So, if I change the skill in Notion, I still have to rerun the same command. Later, I’ll show you how to sync the whole library to GitHub so your team can use the same updated skills across Claude, ChatGPT, and Codex without installing every page separately. Okay, so installing a skill for yourself is easy. The problem starts

02:37 when a whole team has to keep improving it, which is what every business should do. And every time someone gets a better result, corrects the agent, or figures out a better way to do the work, that learning is really important and it stays inside their own chat. And usually, even if they update that skill, it only updates their own version. And so, the next person still starts from an older version and repeats the same mistakes. And a GitHub repo technically gives you a place to store skills, but most writers, producers, sales people, or operators don’t work in GitHub.

03:09 Markdown files are just difficult to review and even when an agent wants to update the skill, it has to mess with commits, pull requests, and the repository. And so, Notion dramatically changes this picture because the skill becomes a normal Notion page that the whole team can actually read and edit. And even if they’re not technical, they can still review, comment, write, and edit in the same convenient Notion interface. And then, agents can also read and update the same pages through Notion MCP as well. So, GitHub in this

03:39 case just becomes the distribution layer on top of Notion. And Notion stays as the source of truth because Notion is a lot more convenient to use than GitHub. And even non-technical people can collaborate with each other and with their agents in Notion. And this way every approved campaign example and decision can stay with the business. So the next project doesn’t start from scratch, it starts from what the previous person or the agent already learned. And if someone leaves, that knowledge still stays with the business. Okay, so now I’m going to show you how

04:10 this actually works for a real client. So I recently sold this exact setup as a pilot to represent a video production agency covering just two workflows. Script writing and producer client communication. So I’ve now had two interviews with their producers. Each one took approximately 45 minutes. And instead of giving you some hypothetical template, I’m going to go into their actual workspace, find the scripts, revisions, and the client feedback and build the script writing skill from the way their team actually works.

04:40 Okay, now let me build this for a real client. So I interviewed John who writes the launch video scripts at Represent. So I asked him questions like what information must be available before he can write a good draft, where do I find the client strategy, audience, offer, or the key messages, how does he structure his writing, if he uses any frameworks or techniques, where did he learn them from, and where should the agent actually store the outputs. And this is a script that John wrote that Represent actually published. And I’m giving this

05:11 agent the interview, the writing docs that John shared with me, and this finished example to create the first version of the skill. And this is my personal skill creator skill that I built myself and I usually run it after a long project or conversation to either update an existing skill or create a new skill for a certain process. And so here I’m going to give this skill to Notion AI. I’m going to attach John’s call and all of their work examples and writing frameworks so it can create the first version of his script writing skill.

05:44 Now I want to go through this myself and later I want John to edit it because if you’re not the one who’s responsible for this process, you will never be able to create a perfect skill. Besides, if he doesn’t at least edit this final draft, he’s not going to continue using and improving this skill over time. Now, let’s open a new chat and let’s try it on another client, and let’s make sure it doesn’t read the existing script so that we can compare the results. Okay, so the skill wrote the full draft and saved it in Notion, and then I also

06:16 asked it to compare it against the actual script for this client. And look at this, it tells me that the hook that I selected was actually already under review, and the team decided to move away from this angle in the real script. And so, that’s an example of a real correction that I can now feed back into the skill so that the next writer doesn’t repeat it. But first, let me connect the skill to GitHub so the whole team can use it and benefit from these updates. So, I’ve put together the steps and the prompt in a free guide link below. So,

06:47 there’s no email, there’s no anything. You can open it and follow along with me right now. And here’s the script writing skill that we built from John’s process. I’m going to publish it in a private skills repo that I can also connect to Claude. But first, we need to create a connection under the developer tools in Represents workspace, and then we also need to make sure that this connection has read access to their skills database. Now, open Cursor or any other coding agent that you prefer, and copy the prompt from the free guide below. Make sure to replace all the links at

07:19 the top with your actual links from Notion and GitHub. Also, make sure you have the Notion API token set under the repository secrets in actions on GitHub. Now, Cursor is wiring up the scheduled sync and publishing the existing Notion skills into this repo. Cursor tells me that the sync has finished, and now if I open one of these links, I can see the skill MD file that has our exact process that we wrote in

07:51 Notion. And so now if I update the skill in Notion, it will automatically sync the skill in this GitHub repo every hour. To install these into Cloud, we need to open plugins in Cloud settings and then click add marketplace. Then here we can add the repository with our skills. Perfect. Now I can see all of the three skills from that repository that will also be automatically synced to my Cloud and to add them, I can simply click add right here. Now if I go back to Cloud, I can easily select the skill by using the

08:22 slash command. Now let’s run it on a real project and turn what we learn into an update for that same shared skill. So let’s give it this brief and see what it comes up with. Here’s the draft it produced and now I want to improve the same skill by using the latest version of the script that represent team has already reviewed. So I’m going to use the document learning skill and I’m going to tell it to adjust the same previous skill so that it improves based on that feedback. Okay,

08:53 so the lesson is now in the original Notion skill where it added this new step under the workflow section. The sync in GitHub which runs every hour now picked up this Notion change and you can see that same new step now showing up in GitHub too. Now the skill in Cloud is also refreshed automatically but if I don’t want to wait, I can also click check for updates. And if I open the contents of the skill, here’s the step number six that I have just added from another conversation. And so this lesson

09:24 came from feedback on one draft in a completely different chat and is now available to the next person using the same refreshed skill. And here’s another really cool use case that I’m personally really excited about. So we built this custom agent platform for a client who runs a course with nine agents that help the members to get through the course faster by actually writing their copy, creating the decks, and even setting up the campaigns. And before this client would always have to come back to me whenever one agent underperforms, but now he can

09:54 edit that agent’s instructions directly in Notion himself, and as soon as he edits it, the GitHub action checks it every 5 minutes, and then syncs it back into the repo. And then our platform agency, where these agents are deployed, rebuilds it and instantly deploys it to the production application. Okay, so now if you’re looking for a place to start, I think you might guess that this is a pretty cool service that you can start offering right now that almost no one else is offering yet. Most teams are still way behind on building their own skills, and a shared skills

10:25 library is how you make the feedback and the experience from every project stay inside the business, which dramatically increases the value of that business. And you can sell this as a standalone project or you can open a door with a smaller engagement like this before a full AI OS install. So, you just interview basically a couple of people like I did, pick the workflows that they already struggle with, and then build those processes into a shared skills library that the whole team can use in Cloud Code, Codex, or whichever agent they prefer. And then, once they see

10:56 those two workflows working, you can offer a much larger install with a full operating system, more agents, and more workflows. I personally think that this is a lot more effective than starting with a generic business audit, which is what everyone else seems to be using today to get clients in the door. Business owners are a lot more skeptical today from what I’ve noticed, and they’re getting pitched AI left and right. And so, the random estimate that some workflow might save them 50 hours doesn’t actually prove that AI will be able to handle this process reliably for

11:26 them. While on the other hand, for Represent, I did exactly this. I charged them $1,000 for those two workflows because the founder was skeptical about a full AI OS install, which is understandable given that they have, you know, 20 people on their team and they have been operating their own version of that for a very long time. Yet, at the same time, it was still only a few hours of my personal time because most of the work is still done by AI agents after I interviewed their employees. And then, once the founder actually sees the value that, you know, the team is actually using their skills and getting better

11:57 results, I will offer a bigger 3-month engagement for a full Notion AI OS install with more agents and more skills. So, honestly, try it. Next time you pitch a business, don’t start with a generic audit. Propose you automate two real workflows, suggest what those workflows are, especially if you’re doing cold outbound, and only then, after you provide the value, propose a bigger engagement or a full AI OS install. And if you want to take this further, I’m going to be running a live session in my school community on Thursday, October 1, about getting

12:28 clients for your business with AI in 2026. I’m going to go over how you can use this exact skills opener to get new clients through the door, plus how you can cut through all the AI slop in their mailboxes right now. We’ll go over how to build your offer, who to target, my client acquisition systems, and the Notion agents that I use to automate the boring parts of the client acquisition process. And if you want to see how I built my whole one-person AI business on top of Notion AI from scratch, check out this video next, and I’ll see you there.