Big Idea

The argument in one line.

Chaining sub-skills into a super-skill with Obsidian as the memory layer creates a self-improving research loop where each run teaches Claude Code how you like your work done.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You are already using Claude Code and want to automate repetitive research workflows with a single slash command.
You use NotebookLM manually for analysis and want Claude Code to drive it programmatically.
You use Obsidian as a second brain and want your AI assistant to read and write into it continuously.
You research the same topic area repeatedly and want outputs to accumulate into searchable, interlinked knowledge.

SKIP IF…

You have not set up Claude Code yet — this workflow requires a working install and familiarity with skills.
You want a turn-key solution; this is a DIY system with ~30 minutes of initial setup.

TL;DR

The full version, fast.

A single Claude Code slash command can chain a YouTube search skill, a NotebookLM analysis skill, and a super-skill that wraps both so one prompt triggers a full research run, offloads heavy AI compute to Google via NotebookLM at zero token cost, and deposits the results as linked markdown files in an Obsidian vault. The CLAUDE.md file in the vault acts as a persistent preference layer, so the more you run the workflow, the more the output aligns to how you actually want things done.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 01:32

01 · The Power of Three

Hook establishing the capstone premise — Skill Creator, NotebookLM, and Obsidian topics from prior videos are being synthesized into one workflow.

01:32 – 05:23

02 · The Workflow

Whiteboard diagram walkthrough: Claude Code drives a YouTube search skill into NotebookLM for analysis and deliverable generation, results land in Obsidian, and CLAUDE.md drives ongoing self-improvement.

05:23 – 09:36

03 · The Setup

Step-by-step install: Skill Creator plugin via /plugin, build YouTube search skill, install notebooklm-py via terminal, authenticate via CLI, use Skill Creator to generate the NotebookLM skill from the GitHub repo, combine both into one super-skill.

09:36 – 13:44

04 · Executing the Workflow

Live demo using /yt-pipeline to research Claude Code MCP servers, pipeline runs 6 minutes, returns a research markdown note and MCP infographic, both visible in Obsidian graph view with backlinks; CLAUDE.md updated to capture preferences.

13:44 – 14:34

05 · More Resources

Recap of the flexible template concept and CTA to Chase AI+ masterclass and free community.

Atomic Insights

Lines worth screenshotting.

NotebookLM runs entirely on Google compute, so using it via Claude Code costs zero Claude tokens for the analysis step.
The Skill Creator can read a GitHub repo URL and generate a working Claude Code skill for that library in one prompt.
A super-skill is just a skill that invokes other skills — all the complexity lives in the prompt, not in the code.
Storing research outputs in Obsidian means Claude Code can reference every prior session without you manually providing context.
Updating CLAUDE.md after each session codifies preferences Claude will apply next run — this is the self-improvement mechanism.
NotebookLM deliverables like infographic or slide deck can take up to 15 minutes; text-only analysis is a few minutes.
The YouTube layer is interchangeable — swap it for PDFs, articles, or any structured data source without changing the rest of the pipeline.
A vault of 100 research notes produces qualitatively different Claude behavior than a vault of 10 — the compound effect is nonlinear.
Running the Skill Creator built-in eval before combining into a super-skill catches bugs before they surface in the full pipeline.
The notebooklm-py library uses undocumented Google APIs, so treating this as a production dependency carries real breakage risk.

Takeaway

One command, three tools, compounding memory.

WHAT TO LEARN

Wrapping sub-skills into a super-skill lets you add complexity without adding friction, and Obsidian as the output layer turns each run into training data for the next.

Chaining Claude Code skills into a super-skill means one slash command can do the work of three, and any individual layer can be swapped without rebuilding the whole pipeline.
NotebookLM handles compute-heavy analysis and deliverable generation at no Claude token cost; Claude Code only spends tokens on orchestration.
The CLAUDE.md file in an Obsidian vault acts as a persistent preference layer that Claude reads at session start, so output style improves without re-prompting.
Storing research outputs as linked markdown in Obsidian lets Claude Code reference prior sessions automatically, removing the need to re-inject context each run.
The Skill Creator can generate a working skill from a GitHub repo URL or a natural-language description, which means adding new tools to the pipeline requires no manual code.
Running the Skill Creator eval step before combining into a super-skill isolates bugs to the sub-skill level where they are easier to fix.
The workflow is intentionally source-agnostic: the YouTube search layer is interchangeable with any data source that can be wrapped in a skill.

Glossary

Terms worth knowing.

Skill: A Claude Code extension stored as a markdown file in the .claude folder that defines a reusable slash command with its own instructions and behavior.
Skill Creator: A Claude Code plugin that generates new skills from a natural-language description, handling file creation and formatting automatically.
Super-skill: A skill whose instructions tell Claude Code to invoke other existing skills in sequence, acting as an orchestrator that chains sub-skills into a single command.
notebooklm-py: An unofficial open-source Python library that exposes Google NotebookLM features via CLI and Python API, including notebook creation, source ingestion, and deliverable generation.
CLAUDE.md: A markdown file in a project or vault root that Claude Code reads at session start as its instruction set, used here to encode preferred analysis style and output conventions.
Vault: The root folder Obsidian monitors for markdown files; all notes, links, and the knowledge graph are derived from files in this directory.

Resources

Things they pointed at.

07:07toolnotebooklm-py ↗

04:52productChase AI+ Claude Code Masterclass ↗

05:12linkFree Chase AI Community ↗

Quotables

Lines you could clip.

03:57

“This almost becomes like a self-improving loop. The more I run the workflow, the more it gets its analysis in the way I like it.”

Clean standalone statement of the core value prop→ TikTok hook↗ Tweet quote

10:40

“These are tokens you are not paying for and Claude Code does not have to use. This is all offloaded to Google.”

Counterintuitive cost angle→ IG reel cold open↗ Tweet quote

12:32

“The CLAUDE.md file is the brain within the brain.”

Tight memorable phrase, no setup needed→ newsletter pull-quote↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

metaphoranalogy

00:00If ClaudeCode plus Notebook LM is amazing and ClaudeCode plus Obsidian is free value and ClaudeCode plus the brand new Skill Creator is legitimately game changing, then what's gonna happen when we combine all these tools together in a practical yet simple to set up workflow that you can start using today in under thirty minutes?

00:20Well, is exactly what we are gonna find out in today's video as I show you step by step how to create one of the most powerful workflows inside of Claude Code. This workflow turns Claude Code into an absolute research monster.

00:34And this video is also pretty much a capstone of everything we've talked about in the last few videos because we've covered things when it comes to Claude code in Notebook LM and Claude code in Obsidian and Claude code in the new skills creator. But here's where we take all these lessons and we synthesize it into something that has practical value.

00:50And on that note, what's important isn't my exact use case. Right? This is a personal chase.ai use case, right, and how I do research for my content.

00:59But you're not a content creator. You probably have a real job. So what I want you to focus on throughout this entire lesson isn't the exact intricacies of how I'm doing my YouTube search.

01:08You should be focused on how do I swap the YouTube search for whatever use case I have and whatever source of information I need, whether that's PDFs or articles or text or whatever. Right? How can we fit in this template into your life?

01:21That's where the value lies, and that's what I want you to focus on. And it's also something this is great at. Right?

01:26This is a very flexible workflow that can adapt to your needs, and we love that. So what the heck is this workflow going to be doing?

01:35Well, like I said, this is research on steroids. So we are gonna be inside of Claude Code, and we are going to do some research via YouTube. Right?

01:43My source of data in this case is going to be YouTube videos. To do that, we will use a specific skill. From there, we are gonna send that YouTube data to NotebookLM via Claude code.

01:55NotebookLM will do analysis on those videos for us. NotebookLM will also give us any deliverable we want, whether that's a podcast or a video or an infographic or a slide deck. And then it passes all of that back to us inside of Cloud Code.

02:09All of this is executed through skills. Furthermore, we are going to combine all those sub skills into essentially one super skill.

02:19We will do this using the skill creator. Right? So that's where the skill creator comes in and obviously the notebook LM stuff will come into play here.

02:30What about Obsidian? Right? Because this is good in a vacuum, but, like, we kinda wanna supercharge this.

02:35I'm probably not just gonna run this workflow one time. Well, enter Obsidian.

02:40All this data we analyze and more so than the individual data, the way we attack the data, how we like our analysis done, what we want the deliverables to look like, how we think.

02:51All of that will be recorded by Claude code in a series of markdown files, a series of text files that Obsidian will be able to take a look at because this is all gonna happen in our vault. Now looking at Obsidian right here, the vault's great, right, for a couple reasons.

03:04For me as the human being, I have great insight into what's going on in my text files. I can click through the files. I can see how they link together, I get cool neat little graphs.

03:12But more importantly, inside of Claude code, all those markdown files are transparent to Claude code itself.

03:19It's easier when it's set up in this obsidian sort of format for Claude code to find the things it needs. Furthermore, over time, we will be able to refine how Claude code speaks to us and thinks in this manner via the claude.md file, which over time means Obsidian helps Claude code do this workflow in a manner we want.

03:41Right? With Obsidian added into this workflow, we're able to turn turn Claude code into, like, this well trained personal assistant that executes this workflow on our behalf.

03:54And that's super powerful. This almost becomes like a self improving loop. Right?

03:58Because the more I run the workflow, the more it gets its analysis in the way I like it, the more I talk to Claude Code, the moral head data is recorded, and Claude Code continues to build and build and build over time this corpus of knowledge and evidence for how I like to work. And so that's how we get this, like, awesome symbiotic relationship and all these things kind of helping one another by combining Cloud Code with the Skill Creator with NotebookLM with Obsidian.

04:23Right? And you can see how flexible this is because this sort of workflow changes whether, you know you know, know, we could take out YouTube, it could be PDFs. Right?

04:30You could even take out the notebook LMPs. You could really have any workflow here. Right?

04:35Insert whatever flow. But if you keep this template of flow, Obsidian, and improve skills via the skill creator, you have something super powerful at your fingertips, and it's not something a lot of people are doing.

04:47Now before we get into how we set this up exactly, a word from our sponsor, yours truly. Again, if you wanna learn more about Claude Code, I just released a Claude Code masterclass inside of Chase AI plus.

05:00It takes you from zero to essentially AI dev regardless of your technical background or lack thereof. Chase.ai plus is great if you're serious about AI and you're trying to make a career out of this thing.

05:10So definitely check that out. Also, there is a free Chase AI community.

05:15You can find that in the description. All the skills we talk about today as well as a number of other free resources can be found there. So there's something for everybody.

05:23So first thing we gotta do is create our skills. You will notice I am inside my vault. We have to be in whatever our vault folder is for Obsidian to pick up on this stuff.

05:31Now skill creator, skill, how to install it, get it working. Make sure you check the video above. I go in-depth, but the five second version, you're just gonna do slash plugin.

05:40You will search for the skill creator tool. You can see mine is installed right here, skill creator.

05:48Install it, exit Cloud Code, spin it back up, you're ready to go. And so if I wanna build a skill, I'm gonna do slash skill creator to make sure it actually uses the skill, then and we're just gonna describe it. In this case, I said I wanted to create a skill that searches YouTube and return structured video results.

06:03It should use the y t dash d l p to search for videos by query, return the results, blah blah blah blah blah. This This is how it is for my YouTube thing. Adjust it for what you want as your source.

06:12Again, these prompts will be available inside of my community. Once you run that, it will create the skill automatically inside of your dot Claude folder. It'll give you some descriptions about what it did with the skill creator tool.

06:22Remember, we have the ability to run tests on it as well if we want to, but we'll skip that for now. So that gives me the YouTube skill. I can now search YouTube.

06:29What about the Notebook LM side? Well, just like the last two things, I have a full video deep dive on that. Check it above, but I will give you the thirty second rundown.

06:37So Notebook LM doesn't have a public facing API. So for us to connect Claude code to Notebook LM, we are gonna be using this GitHub repo, the Notebook LM dash py. I'll put a link in the description.

06:48To install it is very easy. We're just going to run these commands inside of our terminal. So we'll just copy this.

06:54I create a new terminal. Again, I am not inside a Cloud Code at this point. This is just purely the terminal, and I will paste them in there and run the install.

07:02After I run that install, I need to log in to notebook l m authenticate. You see it here in the CLI section. I just copy that notebook l m space login, put it in the terminal, hit enter, a browser window will pop up asking me to log in.

07:16I log in, and that's it. You are done and installed, you can now use NotebookLM. However, we need to teach Claude code how to actually use it.

07:24That's where the skill comes in. Now this repo gives us a command to do it. We can run this Notebook LM skill install if we want.

07:31We also have an ability, what would probably be better now that we have the skill creator, would be to, like, just copy, you know, essentially this entire GitHub repo or just put a link to it, give that to Claude code and say, hey. Use the skill creator to create a skill for notebook l m dash pi.

07:49And you see that prompt right here. Skill creator. Create a skill so we can best use the Notebook LM skill seen here.

07:55Right? Like, this is, like, one of the best things about Claude code is it will do things that affect its, like, own use. Right?

08:02Like, it understands how skills work within its own ecosystem. And so when I do stuff like this, it sort of self improves in a way, which is great. And once you run that, you'll get the same message essentially that you saw above when we created the YouTube search skill.

08:15And when it comes specifically to the NoteLM skill, these commands allow us to do anything and more from the Claude code terminal that you could do inside of NotebookLM normally. So we have the ability to create our own notebook.

08:27We can add as many sources as we like. Well, up to 50. It could be from our drive, copy text files, YouTube, etcetera.

08:33And then like I mentioned before, we have all the deliverables that NotebookLM can give us, audio review, mind map, flashcards, infographic, etcetera, etcetera. So now we have the YouTube skill, and this graphic has just become hideous. Right?

08:45Let's clean this up. So we have the YouTube skill. We now have the Notebook LM set up.

08:50But again, I don't wanna tell Claude code one by one. Alright. Do the YouTube skill.

08:54Sick. Thumbs up. Okay.

08:55Now do the do that skill. Cool. Thumbs up.

08:57I wanna do this all at once. I just want to turn it into one skill, and that's what we'll do now. Returning our workflow into a skill.

09:04And so to create that YouTube pipeline, that workflow super skill, you can see same exact process. Skill creator, and then I just did a stream of consciousness for it to create that.

09:14Pretty much saying, hey. I want this YouTube, um, pipeline skill.

09:19I want it to use a YouTube search. I want it to send it to Notebook LM, and I want, hey. If I ask for it, some sort of deliverable, and I want it brought back.

09:26Right? That's what I said in way too many words. And at that point, it will create the skill, tell you what it did, and then ask if you wanna run any evals, which is up to you.

09:35And at that point, our workflow is essentially all set up. Right? Skills are ready to go.

09:40It's inside Obsidian. Now all we have to do is execute it. So let's do that.

09:45And in our use case, what we will ask for is we will ask for Claude code to go search up videos that have to do with Claude code and MCP. I wanna find out the top five MCP servers, so I wanted to go grab the sources, and I wanted to do analysis, not just what on the top five are, but how are those videos doing? Like, what is driving views?

10:04What are some sort of outliers? What are the gaps? And what can we do to capitalize on?

10:09And I'll also ask for it to take that analysis and create an infographic for me. And that's the exact prompt you see here. I have my YouTube pipeline skill up and loaded.

10:18I could've used natural language, but anytime you use the slash command, you know, it's gonna work a 100%. Like I said, YouTube, MCP, Clog code, analysis, and I asked for an infographic. So you can see it's starting the pipeline, calling the subskills with NotebookLM as well as YT search.

10:34And again, the great thing about this NotebookLM stuff is the fact that all of this processing by the AI is done by NotebookLM. Like, these are tokens you're not playing for and Cloudco doesn't have to use.

10:46This is all offloaded to Google. Thanks, Google. So after six minutes, the analysis is complete.

10:51Know that most of the time when you're talking about, like, just, like, text analysis and you wanna know what Notebook LM is giving back to you, that's pretty quick. The deliverables can take time. So if you're looking for a full slide deck, for example, that can sometimes take up to fifteen minutes.

11:04Right? Because it's several images it needs to create. If it's just like a one off, like an infographic, handful of minutes.

11:09So here's our infographic. Right? Talking about MCP.

11:12Cool. We didn't give it a lot of, um, guidance in terms of the visuals that we wanted to see, but solid. Right?

11:18Suba base, Context seven, Playwright. Alright. Breaks it down into autonomous coding and the essential Vibe coding stack.

11:25So what did they say? Supabase, Figma, Sentry, Posthog, Context seven, Playwright.

11:32Can't argue with that. And then up top, you can see here it gave us the full markdown file for the research. Now remember, this is inside Obsidian.

11:40So while this seems just like a normal markdown file where stuff is randomly in double brackets, it's a much more it's much more obvious and easy for us as human beings to see this in context via Obsidian. Here's the same document inside of Obsidian. Key takeaways, servers, it has the backlinks that will show me the other articles it's related to.

11:59I can see it inside of the graph. Right? Cool stuff.

12:02But that's not where the Obsidian value ends. Remember, the Obsidian value is the fact that I have, you can see it over here on the left, all these markdown files, which taken in the aggregate, pretty much show Claude code how it is I work. And if we look over here to the Claude MD file, and that's what we see right here, the Claude MD file becomes that brain within a brain.

12:23Right? If this vault is the second brain of mine where I have all these ideas, well, the Claude dot MD file is, again, the brain within the brain that tells Claude what this all means and what that means in terms of conventions of how to talk to me, how to give me deliverables, how I want things done.

12:40And so like I said, over time, this vault will grow and grow and grow and grow, but it's very easy for CloudMD to grow along with it.

12:49And, again, be trained and learn and grow alongside this corpus of knowledge. And it's as simple as telling Cloud Code, hey.

12:56Update CloudMD based on our latest conversations so these conventions are maintained and you're actually doing what I wanna do.

13:04And that's as simple as saying, can we update Cloud MD so it better reflects my work style analysis and output preferences based on our latest conversations. Right?

13:12Something as broad as that is enough for Claude to kinda, like, go nuts with it. If you wanna be more specific, you can be more specific. Right?

13:19That's the great thing about this is it's very flexible and it's up to you. And over time, that relationship between Claude code and Obsidian is what it's gonna cause it to improve its performance.

13:32Right? Doing that over the course of a week won't have too much of an effect. Doing it over a month definitely will.

13:37Doing it over a year and hundreds and hundreds of documents and conversations, that will have a huge lasting effect. So that is where I'm gonna leave you guys today.

13:46I hope you got more out of it than just this workflow in particular and, you know, a little inside view of how I do my sort of content research. Because, again, the big sell here with this is that we can take all this away. Right?

13:58And all we need is some sort of workflow in some manner that helps you, right, in whatever it is you do. And if we can take that workflow and turn it into skills and even turn a massive skills into a single skill and plug it into this sort of pipeline, well, then we get the situation where everything is helping each other.

14:16Right? So and, again, on the long term, tons of value there. So let me know in the comments what you thought.

14:24As always, if you wanna learn more about Claude code, you wanna check out the Claude code masterclass, check out Chase AI plus. There's a link to that in the comments. And as always, I'll see you around.

The Hook

The bait, then the rug-pull.

Three tools that each earned a standalone video are here combined into one. The pitch is a research pipeline that costs nothing to run, writes its own memory, and gets better the more you use it.

Frameworks