7 Free GitHub Repos That Make Claude So Good It Feels Illegal
AI Edge's host installs seven free, high-star GitHub repos live on screen, turning stock Claude Code into a social-research engine, a browser operator, an ads department, and a 279-agent company.
Posted
2 days ago
Duration
Format
Listicle
hype
Views
13.6K
350 likes
57 · 43
Big Idea
The argument in one line.
Free, community-built GitHub repos already solve most of what limits stock Claude, and stacking a handful of them, research, browser control, ads auditing, diagramming, and pre-built agent teams, turns one chat window into something closer to a working company.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You already use Claude Code or Claude Skills and want specific, installable tools rather than another general explainer of what agents are.
You run a business, agency, or content operation and want to see a plug-and-play path to delegating research, ads, or ops work to AI.
You're curious what a multi-agent 'AI company' setup actually looks like installed and running, not just described in theory.
SKIP IF…
You want a single deep technical tutorial on one tool; this covers seven repos at a fast pace with limited depth on any one.
You're not using Claude Code, several of the repos here (skills, plugin installs) are Claude-specific rather than model-agnostic.
TL;DR
The full version, fast.
The video walks through seven free GitHub repos with a combined roughly one million stars that extend Claude Code beyond its stock behavior. Last30Days crawls Reddit, X, YouTube, and Hacker News for real audience sentiment instead of news-article summaries. Playwright MCP gives Claude a real, headless browser instead of screenshot-guessing. Claude Ads audits paid accounts across eight platforms with specialist sub-agents. Archify turns a plain-English process description into an architecture diagram. Gary Tan's G-Stack replicates Y Combinator's own idea-review chain inside Claude Code. Agency-Agents installs 279 pre-trained specialist agents across whole business divisions. Buzz gives a Slack-style shared workspace where humans and agents work in the same channels. The through-line: these tools are already built and free, so the fastest path to a more capable Claude is installing them, not re-engineering the workflow from scratch.
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.
Host claims Claude has become 'orders of magnitude more powerful' since finding these repos and previews all seven, promising the full install guide free in his Skool community.
01:22 – 05:04
02 · Repo 1: Last30Days
A research skill that scans Reddit, X, YouTube, Hacker News, and the web for real social sentiment instead of just news articles; demoed comparing public opinion on Claude vs ChatGPT vs Gemini.
05:04 – 09:05
03 · Repo 2: Playwright MCP
Microsoft's official browser-automation MCP server, giving Claude a real headless browser instead of screenshot-and-guess clicking; demoed pulling and comparing YouTube channel stats against a competitor.
09:05 – 11:59
04 · Repo 3: Claude Ads
A paid-media operations tool with eleven specialist agents that audit ad accounts across eight platforms in parallel, score account health, and rank fixes by impact.
11:59 – 14:41
05 · Repo 4: Archify
A skill that turns a plain-English description of a process into an architecture, workflow, or lifecycle diagram, pitched as a way to map a business's SOPs before automating any of it.
14:41 – 18:22
06 · Repo 5: G-Stack (Gary Tan / Y Combinator)
Y Combinator CEO Gary Tan's own 23-skill Claude Code stack, a chained review process (office-hours, CEO review, engineering review, design review, ship) demoed interrogating a new paid-community launch.
18:22 – 22:40
07 · Repo 6: Agency-Agents
A repo of 279 pre-trained specialist agents across full business divisions (engineering, design, paid media, finance, sales, security); demoed spinning up a scoped three-agent marketing team from one prompt.
22:40 – 24:21
08 · Repo 7: Buzz
Jack Dorsey's Block-built Slack-style chat app where humans and AI agents share channels; demoed creating a channel and connecting an agent to it via the Claude Code CLI.
24:21 – 24:49
09 · Outro
Host closes with a promise of a follow-up video combining Agency-Agents and Buzz, and repeats the CTA to his free Skool community for the install guide.
Atomic Insights
Lines worth screenshotting.
Asking Claude to research a topic by default returns an aggregation of news articles, not real audience sentiment from Reddit, X, YouTube, and Hacker News.
Last30Days has roughly 140 contributors and describes itself as an AI-led search engine scored by upvotes, likes, and real money, not editors.
Playwright MCP reads a web page's accessibility structure directly, so Claude can click the exact element needed instead of guessing where to click from a screenshot.
Playwright MCP is Microsoft's own browser-automation framework and the standard tool professional developers already use for end-to-end web testing.
Claude Ads runs eleven specialist agents in parallel to audit ad accounts across Google, Meta, YouTube, LinkedIn, TikTok, Microsoft, Apple, and Amazon at once.
Claude Ads has 9,300 stars, ranks fixes by impact rather than just listing problems, and stays read-only by default until changes are explicitly approved.
Archify has 64,400 stars and converts a plain-English description of a business process into a visual architecture, workflow, or lifecycle diagram.
Gary Tan's G-Stack packages Y Combinator's own idea-interrogation process into 23 chained Claude Code skills, from /office-hours through /plan-ceo-review to /ship.
G-Stack has 133,000 stars, and each skill in its chain hands its output to the next skill, so a design doc feeds a CEO review which feeds an engineering review.
Agency-Agents has 152,000 stars, 124 contributors, and installs 279 pre-trained specialist agents across divisions like engineering, design, finance, sales, and security.
A single prompt describing a three-person marketing team was enough for Agency-Agents to select and configure a scoped team from its full agent roster.
Buzz, built by Jack Dorsey's Block, is a Slack-style chat app where each hired AI agent gets its own channel and can be wired to Claude Code over the CLI.
Stacking these tools compounds: research from Last30Days can feed creative direction for Claude Ads, while Playwright MCP handles the browser steps in between.
Takeaway
Free repos that scale Claude into a team
WHAT TO LEARN
Free GitHub repos already solve most of what limits stock Claude, research, browser control, ads auditing, diagramming, and pre-built agent teams, and stacking them turns one chat into a working operation.
02Repo 1: Last30Days
Asking Claude to research a topic by default only returns an aggregation of news articles; getting real audience sentiment means using a tool that also crawls Reddit, X, YouTube, and Hacker News.
Before launching a product, community, or offer, you can stress-test the idea against what real people are already saying rather than guessing at demand.
03Repo 2: Playwright MCP
A browser tool that reads a page's actual structure works faster and more reliably than one that takes a screenshot and guesses where to click.
Comparing your own content's performance against a competitor's with exact numbers side by side surfaces concrete gaps a general impression would miss.
04Repo 3: Claude Ads
Running paid ads across many platforms benefits from one audit that scores every account at once instead of checking each dashboard separately.
A tool that ranks fixes by impact, not just by what's broken, tells you what to act on first instead of leaving you with an undifferentiated list.
05Repo 4: Archify
Turning a plain-English description of a process into a visual diagram makes it easier to spot the exact step worth automating.
Understanding exactly what a task involves before handing it to an agent produces more consistent results than delegating a vague instruction.
06Repo 5: G-Stack (Gary Tan / Y Combinator)
A chain of review skills, where each one hands its output to the next, catches problems earlier than one broad review done at the end.
Getting a second, independent opinion on a plan from a different AI model or account surfaces blind spots a single reviewer would miss.
07Repo 6: Agency-Agents
A pre-built roster of specialized roles, each trained on a narrow skill, gets you a usable division of labor faster than prompting one generalist AI for every task.
Starting with a small team scoped to your actual business, rather than installing every available role, keeps a multi-agent setup usable instead of overwhelming.
08Repo 7: Buzz
Running a multi-agent setup entirely inside a terminal gets messy fast; a chat-style interface with dedicated channels makes the whole team visible and organized.
Putting each agent in its own channel, the way you'd organize human departments, lets people and AI work in the same shared space instead of separate tools.
Glossary
Terms worth knowing.
MCP (Model Context Protocol)
A standard that lets an AI model like Claude connect to external tools and data sources, such as a real browser, through a defined server interface rather than a one-off integration.
Claude Skill
A packaged, reusable set of instructions installed into Claude Code that gives it a specific trained behavior, callable by name, instead of re-explaining the task in every chat.
Headless browser
A real web browser that runs without a visible window, letting a program like Playwright load pages, click, and read content without a human watching or a screenshot being taken.
Agentic swarm
A setup where several AI agents, each with a narrow specialty, run in parallel or hand work to each other on the same task, rather than one general model doing everything sequentially.
Diarization
Identifying which speaker said which part of an audio or video recording, used here in reference to attributing transcript lines to specific hosts or guests.
Resources
Things they pointed at.
01:24toolLast30Days
05:04toolPlaywright MCP
06:36channelNate Hurk (AI channel, used as competitor comparison)
“I have almost 300 agents just sitting in Claude now that have already been trained.”
concrete, surprising number that sells the repo in one line→ IG reel cold open↗ Tweet quote
22:50
“This is the place, like Slack, but for agents to have all of your agents conversing with each other.”
one-sentence explainer for an otherwise abstract product→ TikTok hook↗ 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.
17px
metaphor
I feel like I've been using Claude Wrong for the last six months. How did I not find out about these GitHub repos sooner? Because ever since I started using these, I - I feel like my Claude has become orders of magnitude more powerful.
I went through all the major topics on GitHub, Claude code, agent skills, Claude skills, MCP servers, Claude AI, AI agents, and I found the best GitHub repos to inject into your Claude. These are repositories that you can simply download, connect to your Claude, and turn Claude into a monster. One of these repos can help you scale up your Claude into an entire 30 agent team that can run your business for you.
Another one of these repos is the exact same system that the CEO of Y Combinator uses to run their multi -billion dollar business. Another one was just launched and was created by Jack Dorsey and is the future of agentic work.
And another one is a repo that I've been using for the last week to scale and optimize my ads. So I've been putting these to the test across my real workflows. I'm going to show you the top seven, which have almost a million stars on GitHub between them.
So they're very highly rated. And I'm going to show you how to turn Claude into an absolute beast. All of the repos.
that I select today are going to be available for free in the description below. If you join my free school community under the resource section, just like every video, I'm going to have the full guide from this video. It's going to contain the seven repos and installation guide, so you can just chuck them straight into cloud code and they're going to start working pretty much straight away.
The install is actually really, really easy. So the first one is last 30 days. This is very cool because when most people are researching, they just use Claude and Claude just researches the internet.
So when they're researching a topic, they basically just get an aggregation of news articles. But what Last 30 Days does is it also scans the social side. So it'll scan Reddit, X, YouTube, Hacker News, Polymarket, and the web all at once to give you real opinions, not...
just from news articles but also from real people so you can basically ask it about any topic and it's going to come back to you with what's rising in popularity what's overrated what people are saying with examples and this becomes very valuable in many ways i'll give you some example prompts if you run a business for example you can stress test the market's opinions about a particular app you want to onboard or a particular funnel that you're thinking of running if you are wanting to use a product you can stress test whether the product is good or not if i'm researching for a youtube video i can get real research based on what people are saying so I don't just need to ask Claude and rely on Claude for researching because I know I'm covering all of the major platforms and social opinion.
So if you do anything research related, which in reality is all of us, because if you run a business, you're constantly researching. If you make content, you're constantly researching. And just in general, if you want to learn anything, you're probably researching.
Last 30 days is an amazing tool. So let me show you. I just asked, I'm running a prompt in the background to get the public sentiment and the opinions and the current vibe check on Claude versus ChatGPT versus Gemini.
What people are using. and what people are switching to, and a complaint nobody at the AI companies wants to hear. Last 30 Days basically describes themselves as an AI -led search engine scored by upvotes, likes, and real money, not editors.
So you can see what's actually moving the needle in terms of social interest and engagement, which is very important. It's great for me because in content creation, that's all you really care about. You care about social sentiment more than news rankings because at the end of the day, you want real people to watch your videos.
So you can see here the engine has done... The research. It's broken down the community sentiment.
Polarized on Claude. Loyal on output quality. Furious on limits.
Strengths. Sticks to your existing code base better than Astra. Highest ambition ceiling.
Coworkers landing with the non -coders. Weaknesses. September 14th limit cut.
Public cancellations and lawsuit. With ChatGPT. Enthusiastic on GPT -Astra.
Uneasy on privacy and ads. Strengths. What people love.
Weaknesses. Common complaints. Gemini.
Love price, reliability, and volume. Not for being the smartest. and then it summarizes all of the information here with what the agents have reported back.
So that's just an example of a broader prompt and you can see that the research outputs here in my opinion are stronger than just asking Claude, especially for social stuff. But here's where it gets interesting. If you run a business and you want data about a potential product or how to launch a product.
For example, I have a school community and I might want to know what are people saying about the current school communities? What do people complain about? What would make one potentially worth money if I were to do a paid offering?
Mine's free, but if I were, what would make it worth it? So you can ask questions like this. You could ask questions about what ads people are responding to, what products people are responding to and use a tool like this to do.
feasibility studies on a product or an offer because you're getting opinions from real people. So let's say you were running a consulting business, like a, you know, white glove AI concierge type business. You could ask what the average person is looking for.
And because it's crawling through X, Reddit, YouTube, getting real people's thoughts, it's going to aggregate that information much better. So I'm sure you guys, you know, could start to brainstorm and think of the power of a tool like this. And because it has the agents preset.
and pre -trained to sweep in a certain way and scan in a certain way the output's going to be better than just asking claude hey get a social score from x reddit or youtube so based on this prompt i'm already starting to get some insights premium buyers aren't paying for more information they're paying you to help them get something done 100%.
Low ticket trap is a math problem, not a marketing problem when you're charging. What separates the ones people stay in is call frequency and feedback on real work. Key patterns from the research and then you've got the source material down below.
So the next MCP that I've been finding very, very useful in my workflow, especially for content creation, is Playwright MCP. So this essentially... gives Claude browser automation capabilities.
So it enables an LLM like Claude or GPT to interact with web pages through structured accessibility snapshots, which bypasses the need for screenshots or visually tuned models, which means that it can be a lot faster to interact with the browser and do research than computer use. I don't know if any of you have been using Grokbot recently.
Although it's very capable, a lot of the time it's extremely slow. Whereas this bypasses everything. This is literally Microsoft's official MCP server for Playwright.
So it's the tool that real developers use. to automate their agentic workflows in their browser. So it means that Claude can essentially interact with a real browser.
So clicking, typing, researching, scanning, filling forms, reading pages, and coming back with information, which essentially gives Claude hands, right? So Claude on its own can't do anything on the browser. It can just read it.
If you put Claude in a Chrome extension or you use computer use, or if you use its own browser, technically it can take a screenshot, it guesses where to click, and then it can click on pixels, but it's very slow. and it lives in the browser. Whereas Playwright MCP opens its own browser, reads the page structure, clicks the exact button needed to interact, and it can run without you, so it's essentially headless.
So it can stay logged in, which means you can chain it onto other workflows more efficiently. Let me give you an example of something it can do. So I'm going to ask it to open a browser and go to my channel, AI Edge, and then I want it to open the videos tab, read my most 10 recent uploads, log all of the data, then do the same for one of my competitors.
I'm going to use Nate Hurk as one of the biggest AI channels for this. example put all 20 in one table sorted by views per day since upload tell me in three lines what their titles are doing that mine aren't now i can see here that it's actually loaded up a browser you can see it's going on to both of our channels it's taking the data you can see here it's already taken the titles it's loaded them it's looked at the competitor page so it's basically just interacting with the browser this is a tool that many professional developers use so it's microsoft's browser automation framework so it's become the standard for end -to -end testing and web development so if you're a front -end team if you're building landing pages uis it's very common to use playwright mcp for tweaks so if an agent writes a ui change it can open the page in playwright check that it rendered instead of the developer like alt tabbing to look and essentially you end up with a scenario where the agent can now see the result of its own code so it can use it for autonomous checking of its own work the key distinction to understand is that it actually accesses the real browser as opposed to chord or gbt which takes screenshots it slows the whole process down and it misses information and it's less accurate and it's less autonomous so you can see here it's giving me the output it's giving me the views
the titles, the day since upload, and it gives me some of the reasons why he's getting more views than me. This is a very quick test, but it's something that's very useful when it comes to YouTube research. Like, for example, I could say, research all videos on GitHub repos.
You know, pull all of those videos, go in, pull the transcripts, come back to me with the transcripts, and summarize the reason why the most viral videos went viral. Or I could say, search Claude, pick the most viral outlier topics, and help me ideate 10 content ideas based on that. So there are many ways that you can use Playwright.
It's just a bit more seamless than... you know, using the Claude browser. So if you've ever run into issues with the browser, if you do a lot of browser work, I think this is a great MCP for you.
All right. So for the next two repos, repos three and four, we're going to focus specifically on jobs. So we're going to focus on ads.
This is something that I do a lot in my business. If you're on a marketing company as well, this is going to be super useful and something that I think all of you can use as number four, which is a skill that enables you to build. diagrams and workflows from scratch.
So you can literally describe a system, a workflow, and it's going to make an amazing visual demonstration of that workflow, which you can then give to an agent. It's super cool. Then we're going to move on to company use cases in repo five and six.
I'm going to show you how you could run a 30 employee company with Claude, all thanks to a repo from Gary Tan, the CEO of Y Combinator, which literally makes Claude work like a Y Combinator company. And then number six and seven is just going to take any startup idea, any company, pretty much if you're running any business on AI.
to the next level all right so the next repo is called ads this is so sick if you're running paid ads now although a lot of creators or agencies or companies they focus on organic content one thing that is really effective especially right now is meta ads and this is a way that you can track all of your ads from meta from youtube pretty much any platform as you can see here all in one place.
You can optimize your ads. You can find out why they're working, why they aren't working. You can track versus competitors.
This is pretty much a way to have an entire ads agency within Claude. And I really wanted to speak about it today because no matter if you are working in the agency model, if you're working for clients, if you're doing AI concierge or consulting, if you're running your own business, you probably... need a repo like this.
So it has 9 .3k stars. I honestly think it's underrated because I think it's great. So this is essentially a cloud -first portable paid media operations for agencies, consultants, and in -house performance teams.
Don't you think it's just crazy now? And especially as we get to the last repos, I think it'll become obvious that we have software tools that can literally put a professional marketing team in your pocket. I was testing out some of the hedge fund stuff before and I find it mind -blowing because you can literally have a professional hedge fund using pro quant strategies in your pocket now on your cloud account.
Like it's just crazy. what you can do these days. But essentially the way Claude Ads works is you run one command, the agents fan out.
So they have 11 specialist agents, which audit every platform in parallel. It'll give you a health score of all your ads, and then it will give you an action plan. So it will rank the potential fixes by impact and what you can ship right now.
So you can audit all of your ads at once. Not only does it do audits, it can also plan campaigns, budget. You can strategize.
It can tell you when you should up ad spend in a certain area, when you should lower ad spend in a certain area, what your ROI is, if you're missing something. And if you combine it with - With the first repo, Last 30 Days, that I shared with you as well, we can get new ad creative ideas based on what people are talking about.
It's just going to put rocket fuel on top of this. And that is the real power of these repos and MCPs. It's really stacking.
For example, you could do the research on Last 30 Days. You could use this ads agent to do the grunt work. You could use the Higgs field MCP to generate the media because it has access to the latest video and photo models.
You could use Playwright MCP to actually go into a browser and do something. Like you can stack all of these on top of... each other and the last one's going to talk about you can actually sync all of this with your other ai agent employees so that is how you have to think of ai now it's tool stacking using each tool for what it's good at and building a genuine company you know you want to have your ads manager you want to have your financial officer you want to have every agent with a specific trained role and that's why these repos are good because you don't need to go through the effort of training it someone has already done it and not just someone you have in this case four contributors and these are top contributors you know literally claude and codex sometimes you have crazy amounts of contributors like let's go on to like last 30 days for example has 140 contributors so there are a lot of amazing developers who are putting you know time behind this so as you can see in the demo here you have a bunch of commands you can choose what command you want to run for example the audit command then it will spin up an orchestrator it'll run parallel audit agents and then it will
provide an output. And you could see the example output here. So that's an absolute banger one.
Let's get into Archify, which is a skill to create beautiful, verifiable architecture, workflow, sequence, data flow, and lifestyle diagrams, which are their own HTMLs, which you can put into Notion. You can export them. I'll show you exactly how to do it.
And this one has 64 .4K stars. So essentially it's a way to take... plain English and build it into architecture.
So this is especially amazing if you're doing client work because you can just describe the client's problem. Let's say you get on a discovery call with a client. They tell you their problems.
You can then determine a workflow in plain English with Claude and then you can use it to create an amazing diagram, which then you can automate or you can give to an agent to automate. And I find being able to graph engineer and visualize processes not only helps your agents, but helps you understand processes better, which in turn helps you get the results that you need out of AI.
And this was actually one of the top trending repositories. People have been loving it. As you can see, it creates these diagrams.
You can switch between dark load and light mode for different flows that you would like to create. And what I actually recommend people do is they build out SOPs for their real business. Like you've really just got to be writing everything down in your business.
Have a database like Notion, for example, have every single SOP. So how you go about creating a video, how you go about planning for an ad campaign, how you go about onboarding a client, how you go about doing your financials, whatever you do. If you're working nine to five, how you do your specific task, write down each component, draw it up into an architecture and a workflow using Archify.
And then from there. go and find each bit that you can automate because once you've mapped it out and you've graph engineered and you have a visual exemplification of what you do you can actually go and start picking individual components that you want to outsource and this is really important because before you set an agent to do something you really want to understand exactly what it's doing and i feel like this also helps you get a much better understanding of ai it genuinely makes you better at using ai because you understand why things are happening and how things are happening and you're going to get much more consistent outputs with because it's already been trained.
It's already been stress tested and it already has top developers versus trying to graph engineer yourself, which to be honest, I was doing for a long time. And as I said in the beginning of the video, before finding some of these repositories, I genuinely think I was using Claude and AI wrong in many ways. Not wrong, but I was just putting too much work into things when people have already done the grunt work for me.
Like the reality is... developers all around the world are having the same issues that you have in terms of memory in terms of ai not sounding human enough in terms of it being too inconsistent or too hard to graph engineer in terms of the web or browser capabilities not being correct and that's why developers are made workarounds and that's why these highly rated githubs exist because genuinely they're better for many of these things than just using the stock and you know plug and play chat gpt or claude now we're moving on To number five and number six, these are use cases for your company or to create a company.
And this is one of the heavy hitters of this video. This is Gary Tan, the CEO of Y Combinator's exact cloud code setup. If you don't know what Y Combinator is, they're basically venture capital.
So they invest in startups. They incubate a lot of startups. They manage an estimated $12 .2 billion in assets.
And they are at the front foot of AI. They're investing and backing in a lot of the up and coming exciting tech. So you best believe these guys are ahead of the curve.
when it comes to AI. And Gary Tan released his cloud code setup with 23 opinionated tools that serve, this is crazy, as a CEO, a designer, an engineer, a manager, a release manager, you literally have an entire company in your pocket. This has 133 ,000 stars, but...
I still don't hear many people talking about it. And even I was onto this a little bit late, to be honest. When I started testing this, I realized how crazy it actually is.
So to install it, you need to follow these steps. If you are in the school, I'm going to link all of these anyway with a direct link to each GitHub and have a one -shot prompt so you can install all of these at once. Because if you go into Cloud Code and you just drop that MD file and install everything at once, then you can just open any chat and you can just start using this stuff because it's already installed on your device.
Or you can just go through and install manually. Completely up to you if you just want to install this once. Specifically, for example, my school community is free.
So you can see that GStack has a collection of skills. So each skill feeds into the next. For example, Office Hours writes a design document that the planned CEO review skill, the CEO, needs to actually think about the problem.
Then it can hand off to the engineering manager who revites a test that the QA picks up. Review catches bugs, which means that nothing falls through the cracks because every step... knows what came before it so what i recommend is using forward slash office hours to kick off a chain for example you know let's say you want to build a daily briefing app for your calendar you go forward slash office hours then claude because it's tapped into g stack would start asking questions you give it a response it's going to challenge it it's going to give you a recommendation then you can kick off the next skill and then it's going to go through the system until you've actually you know ship the product so this is a real flow that is created by a real venture capital and incubation company but it doesn't just stop there because they also have power tools so you can use forward slash codex for a second opinion so an independent code review from open ai so let's say you're using claude to code you can get a second opinion from codex called code the other way around if you are using codex you've got careful which states the safety guard rails you've got a few really cool ones like ios qa as well
to connect to your iPhone. So basically every time you use it, you want to start with plan CO review to kick off a process. And you can also spin up multiple sub agents and really run this as its own autonomous team using the training from G -Stack.
The way to really understand G -Stack is from the lens of Gary Tan, because this is essentially the same skill stack that he runs in Claude. This is how he at YC would interrogate an idea. So if you are trying to ship something new for the company, if you are trying to launch a new product, if you are trying to make a tweak or a change you can just run it through a co review for example if i want to launch a new product which is the prompt i use here i'm going to run it through co review as long as it's tapped into my memory so if my folder has a md file and a instructions file with my actual memory and data or it's connected to my notion for example which has all the information on my company it'll be able to extract off that information as well which i think is really important for this to you know actually give applicable advice to my business it's essentially like you've hired yc to work in your company and it's constantly updated i think there was literally an update today so they're always adding new stuff which is good because some repos get stale and aren't that active but this one's super active all right while that's running the analysis let's get into the one -two punch here which is so powerful if you want to start a fully ai business like if you are hell -bent on starting a one -person ai business or even if you have a few employees and you just want to scale up your business with agents but you don't have to code agents yourself from scratch this is a plug -and -play solution probably the best
plug and play solution I've found, and not just me, it's literally got 152 ,000 stars and 124 contributors. So the market seems to think so as well. That is going to basically give you an entire AI agency or AI company from scratch.
So the Y Combinator Gary Tan stack that I just showed you is going to interrogate a product launch and it's going to run through start to finish. This is going to... be the operations side that runs your company on a day -to -day basis.
So you've got entire divisions to tap into here. And remember, each agent is actually trained on a speciality. So you've got an engineering division with a front -end developer, back -end architect, mobile app builder, AI engineer.
You can plug this into any model you want. Code reviewer. The list goes on and on and on.
I won't read them all out or you'll probably get bored. You've got a design division, a UI designer, a UX researcher. a UX architect, a brand guardian, a visual storyteller.
You've got a paid media division. And once again, you can combine this with some of the other tools that I spoke about earlier. And you can train the agents to orchestrate and delegate work to those other MCPs as well.
You've got an ad creative strategist. You've got a paid social strategist. You've got a sales division, an outbound strategist, a discovery coach, a deal strategist.
You've got a marketing division, a growth hacker, a content creator. And these are trained by real people. You've got to remember, this isn't just slop that someone's just spun up in five seconds.
This is contributed by a hundred... 124 people that have built all of these skills to act and behave and, you know, work like these employees really would in the real world.
You've got a product division. It just goes on and on. I literally can't even scroll fast enough.
I'll go through it quickly. Security division, support division, special computing division, specialized division. Look, depends what company you have, but what they've done is they've gone out and essentially built something for every single company.
So you can literally install this and I'll show you how to do that in a second. And based on whatever business you're starting, you can just pick the agents that you want. Oh, and by the way, the Gary Tan skill.
is doing an amazing breakdown in the background of how i should actually structure my launch so you can see much better than a normal chord output just the way that the skill has been refined to critically think and that makes sense because it's literally gary's workflow so let me just think of a prompt off the top of my head for agency agents i want to utilize agency agents to start a company I am a marketing company focused on marketing AI products.
I need a design specialist for my ad creatives. I need a marketing manager and I need an accountant because I'm always losing track of my finances. I want to start with a small team to begin with.
By the way, I use Whisperflow for all of my voice transcriptions. Highly, highly recommend. All right, so it's saying it has what it needs from the roster.
It's going to set up the company's own workspace with a three agent team. Remember, there's literally hundreds of agents to select from. So you could just scale it up.
It has 279 agents installed. Like, this is crazy. We just installed this for free.
I have almost 300 agents just sitting in Claude now that have already been trained. These are actual Claude skills, right? So it's not like me just asking around in Claude chat, oh, be a design strategist.
It's like, this has already been trained. Now, maybe it's not perfectly trained to how you want it. That's where you still need to make tweaks, but it's going to be a hell of a lot better than just doing it completely from scratch because there's real people, real contributors, real developers, as I showed you, 124 people that have contributed towards this.
And you can actually just scroll through if you want because it's, you know, a repository at the end of the day. And just look at the individual agents or skills under each single category. Here's finance.
Here's healthcare. Here's spatial computing. Here is scripts.
There's so much here. So you can see that I'm now speaking to the designer. I'm about to launch a new product.
What tools are you going to use to design my webpage for something super professional? You can see that it's routing to the agent. And then the agent is going to respond to me now.
So I'm speaking to, so this is really smart. It wants to use core design canvas. Then it's going to use G stack design consultation.
Then it's going to use Higgs field. And then it's going to use Versal plugin for hosting. Playwright and the in -app browser for screenshots and QA on the live page, which we installed before.
This is really smart. This is crazy. I can literally have a whole company running here.
However, something that's obvious is that running a company like this is not very efficient. I mean, you're literally in a cloud terminal. It's going to get messy.
It's just not the right UI for things. So what is the right UI for things? Well, this is where number seven comes in.
It's called Buzz. It's by Jack Dorsey. This is very recent on GitHub, but obviously they recently launched their desktop application.
And this is the place, like Slack, but for agents to have all of your agents conversing with each other. So all of those agents, the ad specialist, the designer.
the finance accountant, the CEO, all of those skills we went through, they can now all be in Slack channels. So I can set up the company like a real company. So I can create a new channel.
I can make the channel YouTube strategy, and then I can click create channel. And then I can drop my YouTube strategist in here and I can connect it to cloud code through the CLI. So I can actually tap into the agent capabilities from the agency agents repo that I just showed you.
So this is how you can actually set up a functional agentic company where you can properly. visualize things, but still tap into all of their skills through Claude. I'm going to probably have to do a separate video on this.
It deserves its own video. So if you want that, let me know in the comments below. But number seven is Buzz.
This is how you tie everything together. And all you need to do is install it from the GitHub. Once again, I'll leave my free school community below, but you could start.
developing a whole army of agents working for you, as you can see here. And you can have employees working with agents. I think that's the big difference.
Obviously, you could do it on Slack too, don't get me wrong. But Buzz, because of the CLI integration, is very AI forward. And that was why Jack Dorsey decided to do it because he wanted an AI forward solution.
So you can have real employees in Buzz working alongside agents and agents completing work when employees aren't working. So this is the future of work, in my opinion. It's not just human employees.
It's not just agent employees. It's agents and humans. humans working together.
So let me do a video to show you how to stack number six and seven on top of each other because I think it'll be crazy. But that's all I have for today. Make sure to subscribe to the channel for more content like this so I can help you get the edge in AI and my free school communities below.
If you want all of the GitHub's that I spoke about today and you can just drag and drop that MD file that I'm going to give you into cloud and it will install everything so you can go ahead and use this stuff yourself. This is how you. 10, 20, 30, 40, 50x the capabilities of the stock standard cord.
I'll see you in the next one. Have a lovely rest of your day. Peace out.
The Hook
The bait, then the rug-pull.
Before naming a single repo, the host claims he's been using Claude wrong for six months, then opens seven GitHub pages, live, and installs each one against a real workflow to prove it.
Frameworks
Named ideas worth stealing.
15:16list
G-Stack review chain
/office-hours
/plan-ceo-review
/plan-eng-review
/plan-design-review
/plan-devex-review
/ship
/reflect
Gary Tan's chained skill sequence where each skill's output feeds the next, an idea gets challenged, planned, reviewed for engineering and design, then shipped and reflected on.
Steal forpressure-testing any new product or feature idea before writing code
10:04list
Claude Ads platform coverage
Google Ads
Meta Ads
YouTube
LinkedIn Ads
TikTok Ads
Microsoft Ads
Apple Ads
Amazon Ads
The eight ad platforms Claude Ads audits in one run, split into a search/video/social segment and a commerce/retail media segment.
Steal forany agency or in-house team running ads across more than one platform
CTA Breakdown
How they asked for the click.
VERBAL ASK
01:03link
“Join my free Skool community and claim the full GitHub repos guide, under the Classroom tab.”
Stated once early (before repo 1), repeated mid-video before repo 5, and repeated again in the outro, each time pointing at the same free community link.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A tutorial arguing prompt engineering is dying now that models are smart enough to self-correct — and a walkthrough of Claude Code's /goal, /loop, and /schedule commands with a live website-audit and YouTube-monitoring demo.
Elon's new agent platform runs a team of AI employees on their own always-on computers — tested here as a CFO, an EA, a Dubai real estate scout, and a YouTube research analyst.
Two hosts run through ten GitHub repos and skills that fix one narrow AI-coding annoyance each, then argue about whether skills are already becoming obsolete.
Claude Desktop and the Codex app both quietly shipped native, multi-tab in-app browsers within days of each other — used right, one prompt now opens, drafts, and stages a dozen tabs at once.
A software architect sketches phantom-fs, the abstraction layer that lets dozens of Claude Code agents share one synced, conflict-free codebase from anywhere.
A 16-minute walkthrough of replacing a $99+/month ManyChat plan with a free, self-hosted Instagram DM automation, installed end to end by a Claude Code agent skill.
00:26
00:46
01:07
01:42
02:00
02:19
02:38
02:56
03:15
03:34
03:52
04:11
04:29
04:48
05:25
05:44
06:02
06:22
06:44
06:50
07:17
07:36
07:54
08:13
08:31
08:50
09:27
09:47
10:04
10:26
10:42
11:00
11:27
11:31
11:49
12:33
12:57
13:11
13:29
13:38
14:14
14:25
15:02
15:21
15:33
15:58
16:17
16:35
16:54
17:12
17:31
17:50
18:08
18:46
19:04
19:23
19:41
20:00
20:19
20:37
20:56
21:14
21:33
21:52
22:10
22:29
23:06
23:25
23:43
24:02
24:39
21:37
22:56
22:57
Video of the Day23:10
19:28
15:58