6 GitHub Repos That Turn Claude Code Into a Full AI Team

Quick answer: The six repos to stack on Claude Code are Superpowers (planning and discipline), Everything Claude Code (agents, skills, and a lower token bill), Claude-Flow (agent swarms), Open Design (premium design output), Obsidian Skills (your second brain), and Karpathy Skills (cleaner CLAUDE.md rules). Install Superpowers and Everything Claude Code first — the rest you add as you grow.

Want to go deeper? I run a live Claude Code workshop — check the latest dates and see if you can still grab a seat.

Most people install Claude Code and get a fancy autocomplete. The people who actually ship with it have stacked a handful of GitHub repos on top — and the difference is night and day.

Here’s the thing. Out of the box, Claude Code is good. Stack the right repos on it and it stops acting like a tool that tells you “done” and starts acting like a disciplined AI team that plans, builds, reviews its own work, and ships while you sleep.

These are the six I install on every project, what each one does, why you need it, and the order to add them. Build along with me.

Why stack repos on Claude Code at all?

A bare Claude Code install knows how to write code. What it doesn’t have by default is discipline — a process that forces it to think before it builds, work in parallel, check its own work, and not wander off “improving” things you never asked for.

That’s what these repos add. Each one is a one-command install (either through or by cloning the repo and asking Claude to install it). Most auto-activate the moment they’re in — you don’t babysit them.

The six repos, in install order

1. Superpowers — my number-one pick. This is a full development methodology, not a prompt pack. It ships skills, agents, hooks (rules), and config files. Before Claude writes a single line, it forces a conversation about what you’re actually building, then it writes the spec, plans for a few minutes, runs a red-green-refactor test loop, dispatches sub-agents to work in parallel, reviews its own code, and finishes the branch. Anthropic liked it enough to pull it into their official marketplace. Install this first — when I teach my Claude Code class, this is the one every student plugs in for an instant head start.

2. Everything Claude Code — the best beginner head start. The creator won Anthropic’s hackathon, and it shows: 48 specialized agents, 182 skills, 68 commands, token optimization that can cut your bill roughly in half, and memory that survives between sessions. It also ships real safety — CVE scanning and prompt-injection blocking so malicious code can’t sneak in — and it works across Claude Code, Codex, Cursor, and Gemini. Massive documentation, its own site. You’ll outgrow some of it as you level up, but on day one nothing gets you further, faster.

3. Claude-Flow — graduate into this one. When one instance of Claude isn’t enough, this is where you go: 100 specialized agents that self-organize into swarms, shared memory across every session, and federated comms so agents on different machines talk securely without leaking data. It’s advanced — I run it on a separate machine for heavy, multi-project, team builds. Don’t start here. Add it when you’re a power user pushing big jobs.

4. Open Design — install this early if you build anything visual. It’s the open-source answer to Claude’s Cloud Design: 13 coding agents, 31 design skills, and 72 branded design systems baked in (Linear, Stripe, Vercel, Airbnb, Apple, Notion). It picks a direction, builds, critiques its own work, and ships a real design artifact you can see. It’s also lighter on tokens than Cloud Design, which matters if you’re working off a subscription. This is one of my favorites — I run it consistently and it pairs perfectly with a front-end skill.

5. Obsidian Skills — the bridge to your second brain. Built by a design lead at Obsidian, this gives Claude native Obsidian skills: wikilinks, embeds, callouts, properties, bases, JSON canvas, the Obsidian CLI, and defuddle for stripping web clutter to save tokens. If you live in Obsidian for notes and research like I do, this stops the endless copy-paste between your notes and your AI — they finally talk to each other. Only worth it if you actually run a second brain.

6. Karpathy Skills — a tiny file that sharpens everything. Andrej Karpathy posted a list of how language models write bad code: wrong assumptions, over-complication, drive-by refactoring, “improving” what you didn’t ask for. Someone turned the fix into a single CLAUDE.md file with four principles — think before you code, build the simplest thing, make surgical changes, and don’t touch what wasn’t asked — plus a rule to define success criteria so the agent loops until it actually works. It’s tiny, it installs in seconds, and it makes every CLAUDE.md file you already have behave better.

One honest gotcha

Don’t install all six at once on day one. Superpowers and Everything Claude Code overlap a lot, and Claude-Flow is genuinely heavy — if you bolt the swarm system on before you understand the basics, you’ll drown in agents you don’t know how to steer. Add Superpowers, get comfortable, add Everything Claude Code, then reach for the others only when a real need shows up. Six repos, one ecosystem — not six things to babysit.

Why this matters

This is the gap between using Claude Code and operating with it. A bare install helps you type faster. A stacked install plans the work, builds it with discipline, designs it well, remembers what you did last week, and corrects its own mistakes — which is exactly what a team does.

I run five businesses on AI agents doing the heavy lifting. The repos above are the difference between an assistant that talks and a team that ships. Start with two, grow into the rest.

FAQ

What’s the best repo to install on Claude Code first?Superpowers. It’s a full methodology that forces Claude to spec, plan, test, and review before it builds — so you get disciplined output instead of a tool that just says “done.” Anthropic pulled it into their official marketplace, and it’s the first thing I have students install.

Do these Claude Code repos cost money?No. All six are open-source GitHub repos. You install them with a single command — either through Claude Code’s menu or by cloning the repo and asking Claude to install it. The only cost is your normal Claude Code usage, and a couple of these (like Everything Claude Code) actually lower your token bill.

Will installing multiple repos slow Claude Code down or conflict?Some overlap — Superpowers and Everything Claude Code cover similar ground — so add them in order rather than all at once. Claude-Flow is the heavy one; only add it when you’re running big, multi-agent jobs. Start with two repos, get comfortable, then grow.

Do I need to know how to code to use these?No. Each repo is a one-command install and most auto-activate on their own. You’re adding skills and rules to Claude Code, not writing software. If you can paste an install command and follow a short setup, you can run all six.

Want this stacked and configured into your business the right way — without trial and error? This is exactly the kind of AI ecosystem I architect for seven-figure operators. Book a call with Nuno M. Tavares at https://nunomtavares.com/booking/

Written by Nuno M. Tavares — AI automation consultant based in New Jersey, USA; founder of Automated Marketer and Rapid Active Marketing. (Not the Portuguese footballer of the same name.)