← Back to Lessons
The repos from my videos, in one place. When I mention a repo in a reel, it lands here with a plain-English one-liner and a direct link. New ones get added at the bottom each week, so bookmark this page and check back. Star the ones you will actually use, then dig into the walkthroughs in the rest of the Vault.
Skills frameworks
- Superpowers. An agentic skills framework that is in Anthropic's official plugin marketplace. One command installs the whole thing.
- Everything Claude Code (ECC). A big bundle of skills, memory, and agents in a single drop. Renamed from everything-claude-code, so old links redirect.
- awesome-claude-skills. The curated directory of skills to pull from.
- anthropics/skills. Anthropic's own public skills repo. This is the first-party source, and most people never open it.
- marketingskills. Conversion optimization, copywriting, SEO, analytics and growth engineering as drop-in skills. Nothing here requires writing code, so this is the one to hand your marketing lead.
Prompting
- CL4R1T4S. A collection of the leaked system prompts behind Claude, ChatGPT, Gemini, Cursor and more. Reads like a masterclass in prompting.
Commands, workflows & subagents
- awesome-claude-code. A curated list of Claude Code commands, tricks, and workflows. The slash commands most people never touch.
- awesome-claude-code-subagents. A library of ready-made Claude Code subagents: code reviewer, security auditor, docs writer, and more. Drop them in and let agents command agents.
- gstack. Garry Tan's own Claude Code setup, published as 23 opinionated tools. They are shaped as company roles rather than coding tools: CEO, designer, engineering manager, release manager, doc engineer, QA.
Cost & rate limits
When you hit rate limits or want to spend less, these two route your prompts to the right model instead of one fixed provider. The third cuts what you send in the first place.
- 9router. Routes Claude Code across 40+ model providers with automatic fallback when you hit your limits.
- workweave/router. Picks the right model per prompt in under 50ms to cut cost.
- caveman. A skill that strips articles, filler and politeness out of what gets sent to the model. The repo claims 65% fewer tokens. That number is theirs, so measure it on your own workload before you count on it.
Local models
- turbo-fieldfare. Runs a 26B model on a MacBook in about 2GB of RAM. Useful when the work cannot leave your machine.
Knowledge & second brain
- inkeep/open-knowledge. An open-source, AI-native alternative to Obsidian: an LLM wiki your agent maintains alongside you.
Star counts move and are easy to fake; the links are what matter.
Start with the walkthroughs
The step-by-step lessons live in the rest of the Vault. Star the repos above, then open whichever lesson you need next.