7 Free GitHub Repos That Push Claude to Beast Mode

The video opens with a confession that stuck with me: six months of using Claude the wrong way. AI Edge says he tested seven free GitHub packs that now make his Claude feel orders of magnitude stronger, with almost a million stars combined. Each one drops into Claude Code directly and changes how work gets done, not just what the model knows. The through line is installation simplicity with outsized operational impact, and every pack is free and ready to use.

A research engine that scores social signal

The first pick, Last 30 Days, reframes research beyond news aggregation. It scans Reddit, X, YouTube, Hacker News, Polymarket and the web at once and scores results by upvotes, likes and real money rather than editors. The promise is social signal over headline summaries. Practical uses include stress testing a market before launching an app, vetting a funnel, or grounding a YouTube brief in what people actually say. The demo in the video runs a vibe check on Claude versus ChatGPT versus Gemini, breaking down polarization, loyalty on output quality, fury on limits, and where each system shines. Built by about 140 contributors, it surfaces patterns like why premium buyers pay for outcomes not information and why call frequency matters in communities, with sources attached.

Second is Microsoft's official Playwright MCP, a browser automation layer that trades pixel guessing for structured accessibility snapshots. Instead of screenshotting and guessing where to click, it reads the page structure and drives the real browser. That makes clicks, typing, scanning, form filling and page reading faster and more accurate than generic computer use, and it can run headless while staying logged in for chaining. The demo is telling: open AI Edge's channel and a competitor like Nate Herk, pull the latest 10 uploads from each, combine all 20 entries into a single table ordered by daily views since upload and summarize in three lines what the competitor titles do differently. The same pattern extends to harvesting GitHub video lists, pulling key points, or letting an agent verify its own UI change by opening the page in Playwright instead of asking a human to tab over.

Third is Claude Ads, flagged as underrated at over 9.3k stars. It is a portable, Claude-first paid media operations skill for agencies, consultants and in-house performance teams. It covers 12 platforms including Google, Meta, YouTube, LinkedIn, TikTok, Microsoft, Apple, Amazon, Reddit, Pinterest, Snapchat and X. One command fans out 11 specialist agents that audit every platform in parallel, produce a health score and return an action plan ranked by impact and shippability. Beyond audits it plans campaigns, budgets and spend shifts and flags ROI gaps. The presenter stresses stacking here: generate ideas with Last 30 Days, do the grunt work with Claude Ads, generate media with a model like Higgsfield and verify in the browser with Playwright. Maintained by a small core of four, it is pitched as an agency in your pocket.

A browser layer that gives Claude hands and eyes

The fourth repo is Archify, described as trending with over 64k stars and built to turn plain English into verifiable visuals. It creates architecture, workflow, sequence, data flow and lifecycle diagrams as self-contained HTML files with motion and clean export, switchable between dark and light modes and embeddable in Notion. The use case is especially strong for client work where a discovery call's messy narrative needs to become a buildable workflow. The presenter's advice is to write down every SOP in the business, map it with Archify, then pick which chunks to automate. Graph engineering in this way helps both the agents and the human understand what is being automated, leading to more consistent results than hand drawing or prompting a generic model.

The fifth heavyweight is Garry Tan's personal Claude Code setup, GStack. Tan is President and CEO of Y Combinator, an incubator behind Coinbase, Instacart and Rippling with about 12.2 billion dollars under management, and his setup ships as 23 opinionated tools that act as CEO, designer, engineering manager, release manager, doc engineer and QA. With over 133k stars and near daily updates, the design is a chain rather than a single chat. Office hours writes a design doc, plan CO review rethinks it as a CEO would, engineering management locks architecture, and QA catches what falls through. Power tools add a second opinion via Codeex, safety rails via Careful and iPhone bridging via iOS QA. The presenter shows a prompt for a daily briefing app flowing through this chain, made far more relevant when the repo reads memory files or a connected Notion workspace.

A skill that makes thinking visual

The sixth pick is the agency agents catalog with about 152k stars, 124 contributors and 279 agents installed after setup. It is pitched as the operations side that runs a company day to day. Engineering division holds frontend, backend, mobile, AI engineer and reviewer; design holds UI designer, UX researcher, brand guardian and visual storyteller; paid media, sales, marketing, product, security and support each have their own specialists. The point is that these are not empty role prompts but trained skills shaped by many contributors. The live example chooses a three person core for an AI marketing company, designer plus marketing manager plus accountant, and spins up a workspace in minutes. When asked how to design a professional web page, the designer routes through Claw design canvas, GStack consultation, Higgsfield generation and Vercel hosting with Playwright QA, showing how the stack orchestrates.

At this point the presenter's main argument lands: real power is tool stacking with narrow specialists, not a single mega prompt. Research from Last 30 Days, execution from Claude Ads, visuals from a generator, verification from Playwright, daily ops from the 279 agent roster, all sharing the same Claude Code memory. Each piece does what it is best at and the sum is larger than swapping models alone. There is also an honest friction noted: running an entire company inside a Claude terminal gets messy and is not the right UI. That friction sets up the final piece.

Stacking the pieces and fixing the interface

The last repo is Buzz, Block's open source answer associated with Jack Dorsey. It is a self hosted, Slack like collaboration surface built on a Nostr relay with a new desktop app. The idea is humans and agents working together in shared channels rather than isolated chats. Specialists from the previous catalog like ad strategist, designer, finance and CEO can live in channels and be invoked through Claude Code's CLI. The demo creates a YouTube strategy channel and drops the strategist inside. The presenter frames Buzz not as a Slack replacement but as an AI forward complement where employees and agents interleave and agents keep work moving after hours. He flags that stacking the agency roster on top of Buzz deserves its own deep dive.

Installation is presented as surprisingly simple. Every repo is linked in the description and a full guide lives in the free school community; a single markdown file can be dropped into Claude Code to install everything at once or each repo can be added individually. Once installed the skills sit on the device and are callable from any chat, with memory files or Notion providing company context for grounded answers. The presenter says he validated the flow on real workflows from ads to content research and the setup was quick rather than fiddly.

Inside GStack the workflow philosophy matters more than the tool count. Instead of one long chat, skills feed the next: a design doc becomes a CEO review, which becomes an architecture lock, which becomes a test, which becomes a QA pass. Nothing falls through because each step knows what came before. The second opinion pattern is concrete: have Claude write code and get an independent review from Codeex, or vice versa. Updating almost daily, GStack avoids the staleness that kills many repos and simulates how YC would interrogate a new product, tweak or launch.

For the agency roster the scale is an asset if scoped wisely. You do not need all 279 agents for every company; start with a small team and expand as needed. Because each division is trained separately, a three person starter team already behaves like distinct professionals rather than the same chat wearing different hats. On quality, the presenter leans on the 124 contributors who shaped the skills from real client work rather than synthetic demos, which is why the skills arrive opinionated and practical. Tweaks for your own context are still needed, but the starting point is far ahead of training from scratch.

Stepping back, the picture is a Claude that stops being a chat window and starts being an operating surface. Research, browser, ads, diagrams, management and collaboration each matured in separate repos and each now acts like a professional teammate. The presenter's claim of 10 to 50 times improvement on stock Claude will read as hype to some, yet the task level wins feel credible when a narrow skill handles a narrow job. The hedge fund analogy used in the video captures it: just as you can now carry pro quant strategies in your pocket, you can carry a marketing and product team as well.

The closing call is to think in stacks, not singles. Grab the guide from the free community, drop the markdown file into Claude Code and run the same seven pack in your own account within minutes. The channel ask fits the same frame: staying ahead in AI now means connecting tools to each other rather than betting on one. The presenter promises a follow up on stacking the agency roster with Buzz, inviting viewers to try the current stack and share what they build.