Top 5 Claude Code Skills... 100,000+ github stars
Jack Roberts demonstrates five Claude Code tools that reduce costs and improve AI development workflows: Karpathy's Graphify for codebase knowledge graphs, Firecrawl for clean web scraping, NotebookLM integration for research, Awesome Design for pre-built design systems, and Claude Code Router for routing tasks to cheaper AI models.
Summary
The video presents five GitHub repositories and tools that enhance Claude Code workflows for developers and AI entrepreneurs. The host, Jack Roberts, frames each tool around cost savings, speed improvements, and practical business applications.
The first tool is Karpathy's Graphify, inspired by Andrej Karpathy's approach to LM knowledge bases. It converts any codebase into a queryable knowledge graph, supporting over 25 programming languages and multimodal inputs including PDFs and audio. The key benefit is token efficiency — instead of Claude reading every file from scratch each session, it navigates the graph like subway lines to reach relevant code directly. Roberts claims this can be up to 70 times cheaper for codebase queries, though he notes it works best for repositories with 500+ files.
The second tool is Firecrawl, an AI-powered web scraping service that converts messy HTML into clean, structured data. Roberts integrates it as a custom MCP connector within Claude, then demonstrates using it to scrape 20 business leads in the pool cleaning industry in Austin, Texas — pulling names, emails, websites, and interesting facts — outputting the results as an interactive HTML document. He notes the free tier provides around 500 credits, with a paid hobby plan at $16/month.
The third tool is a NotebookLM skill that connects Claude Code to Google's NotebookLM via browser cookies. This allows Claude to programmatically create and query research notebooks based on user preferences and memory systems. Roberts demonstrates asking Claude to build a notebook from 20 YouTube videos about Instagram growth, then querying it for actionable advice. He highlights that querying NotebookLM this way costs nothing extra, since the processing happens on Google's side.
The fourth tool is Awesome Design, a GitHub repository containing design systems from 68 brand identities including Apple, Claude, and Lamborghini. It provides typography, fonts, and design frameworks that Claude Code can reference to produce visually polished websites without requiring a professional designer. Roberts demonstrates building a Claude-styled website by simply passing the design markdown to Claude Code, producing a branded result quickly. He cautions that it is a library of predefined templates rather than infinitely flexible, and recommends generating multiple versions simultaneously to find the best result.
The fifth tool is Claude Code Router, a local proxy that intercepts Claude Code's API calls and reroutes them to cheaper models via OpenRouter when full Claude Opus capability is not needed. Roberts compares it to swapping a Formula 1 engine for a Fiat Punto engine when simply parking the car. He notes that models like Kimi K2.6 can be up to 88% cheaper. However, he warns that MCP tool calling relies on Anthropic-specific formatting, some models like Gemini can behave unreliably, and there is added latency compared to native Claude.
Key Insights
- Roberts claims that using Karpathy's Graphify knowledge graph to query a codebase can be up to 70 times cheaper on tokens compared to standard Claude Code sessions, because it navigates directly to relevant code rather than scanning files sequentially — though he notes the overhead only pays off on repos larger than 500 files.
- Roberts demonstrates that Firecrawl can be connected to Claude as a custom MCP connector for free using an API key, and used to scrape structured business lead data — including emails, websites, and company facts — from the web, with the free tier providing approximately 500 credits.
- Roberts argues that querying NotebookLM through Claude Code costs nothing on the user's end because the question is sent to Google's NotebookLM and only the answer is returned, avoiding the token cost of processing large text corpora locally in Claude.
- Roberts describes Awesome Design as a library of 68 pre-coded brand identity systems — including Apple and Lamborghini — that Claude Code can use as design references, arguing that AI is strong at coding but does not inherently understand what visual excellence looks like without explicit examples.
- Roberts warns that before canceling a Claude Opus subscription in favor of Claude Code Router, users should know that MCP skills rely on Anthropic-specific tool call formatting and can behave unreliably with non-Claude backends like Gemini, and that tool calling quality drops significantly below Kimi K2.6 for multi-file refactoring tasks.
Topics
Transcript
[0:00] In this video, I'll share the top five new cla code skills that will make your business cheaper, faster, and better so you stop wasting time and make more money even if you are a complete beginner. And if you don't know who I am, my name is Jack Roberts. I built and saw my last tech startup with 60,000 customers. Now I scale my own AI businesses and tell you all the stuff that actually works. So if you haven't already, grab that coffee. Let's dive straight in. Starting with Andre Karpathy's Graphify. What you're looking at on screen here is a graph of one of my projects that I have that I'm using [0:30] include code and…
Full transcript available for MurmurCast members
Sign Up to AccessMore from Jack Roberts
100 hours of Hermes Agent lessons in 23 minutes
Jack walks through advanced features of Hermes Agent, an AI personal assistant, covering memory systems, background tasks, scheduled cron jobs, model switching, and integration with tools like Obsidian, GitHub, and various AI models. The video aims to help users unlock capabilities beyond basic chatbot usage. Key themes include connecting external memory systems, delegating tasks to specialized AI models, and building a persistent, context-aware personal assistant.
Claude Code = $10,000 Beautiful Slides
Jack demonstrates a GitHub-based system using Claude Code to generate professional slide decks from any website URL by codifying 20 universal design principles. The system uses Firecrawl to extract brand DNA from websites and can integrate AI-generated images via APIs like Krea.ai. The result is polished, brand-consistent HTML presentations created in minutes with minimal input.
The most valuable thing I got from working at McDonald's
The speaker reflects on their first job at McDonald's at age 17, earning £5 an hour. The most valuable takeaway was the realization that they never wanted to work such a job again, reinforced by calculating how long it would take to earn a million pounds at that wage.
Claude Code Memory System = CHEAT CODE
Jack Roberts presents a three-tier Claude memory system designed to give AI tools persistent context across all platforms and sessions. The system consists of short-term identity memory, mid-term project memory via structured folders and claude.md files, and long-term memory using either Obsidian or Pinecone for archiving conversations and expert knowledge. The goal is to eliminate the 'amnesia' problem where AI loses context between chats.
ChatGPT ads aren't actually a bad thing. Here's when I want them.
The speaker argues that ads in ChatGPT are not inherently bad, particularly when users are actively seeking product recommendations. They distinguish this as 'pull advertising,' where the user initiates the request and ChatGPT synthesizes results based on provided context.