Claude · Setup

ECC: the Claude Code config from an Anthropic hackathon winner (and how to install it)

The open-source Claude Code config built by an Anthropic hackathon winner (more than 225,000 GitHub stars): 271 skills, 67 agents and 92 commands, free, and how to install it on your tool in two minutes.

QQuentin Megevand
July 1, 2026 · 5 min read

Everyone spends hours building their Claude Code config brick by brick: a skill here, an agent there, a few homemade commands. Meanwhile, the winner of an Anthropic hackathon open-sourced his, for free, and it already contains everything you are trying to piece together. It is called ECC, and it has passed 225,000 stars on GitHub. One command, and Claude Code goes from an assistant that answers to a pro-level assistant.

ECC is an open-source repository under the MIT license (so free, reusable) that loads 271 ready-to-use skills, 67 specialized agents and 92 commands into your tool in one shot. All maintained and updated every week. This guide shows you what is inside and how to install it in two minutes, without drowning in it.

Claude AI Lab

The Claude AI Lab is my Skool community where I share my Claude systems and the more advanced modules. Entry is free.

Join the Lab →
What you need before you start
1
Your AI tool. ECC works on seven tools: Claude, Claude Code, Cursor, Codex, Gemini, Opencode and Zed. You keep yours, you change nothing.
2
Node installed. The simplest install runs through an npx command. If you have already run npx, you are ready.
3
The right mindset. You do not need all 271 skills on day one. You install the recommended profile, you explore, you use what serves you.
The thing to understand

ECC is not a product to buy. It is an open-source config you load into the tool you already use, in one command, and that you can remove just as fast.

1

What is inside ECC

🔗 github.com/affaan-m/ECC

Three main building blocks, plus a system that holds it all together.

🧠
271 skills, by family
Ready-to-use skill packs: languages and frameworks (Python, React, Rust, Go, Next.js...), security and compliance, research and data, design and motion, DevOps, content and marketing, cost tracking. Real know-how per domain, not generic prompts.
🤖
67 specialized agents
Sub-agents with a precise job: per-language code reviewers, build fixers that resolve compile errors on their own, architect, code reviewer, performance optimizer, security agent.
⌨️
92 commands
Action shortcuts: orchestration (plan, build-mvp, add-feature), project management, session save and resume, code review, security scan, cost report.
🔒
A security layer, AgentShield
A dedicated guard, runtime hooks, a RULES.md file and a SOUL.md that frame what the agent is allowed to do. Useful the moment you let it act on its own.
💾
Instincts and memory
A continuous learning system (the agent builds up instincts across sessions) and a memory that saves and resumes from one time to the next.
Good to know

You will not see all 271 skills at once. Your assistant picks the right one for the task. You get the result without managing the machinery.

2

Installing it in two minutes

🔗 ecc.tools

One command. You just swap the target for your tool (cursor, codex or gemini instead of claude).

npx ecc-install --profile core --target claude
🎚️
Three profiles for your need
minimal loads the essentials, core is the recommended profile, full installs everything. Start with core: complete enough without being overwhelming.
🔁
On another tool
Same command, you change the target: --target cursor, --target codex or --target gemini. The config adapts to the tool.
📦
Or locally
If you prefer to pull everything onto your machine: git clone the repo, then ./install.sh --profile core --target claude. Same result, more control.
The trap to avoid

Do not stack two install methods. If you already installed through the plugin, do not rerun the full installer on top: you would create duplicate skills and behaviors. One method, that is it.

3

Where to start once installed

💬 in your assistant

Do not go exploring all 271 skills by hand. Ask your assistant to show you what it gained, then test one agent on a real need.

🧾
Take inventory
"List the new skills and agents you now have thanks to ECC." You see at a glance what you have on hand.
🔒
Test security
"Run a security review of my code with the dedicated agent." A concrete case, a result you can judge right away.
🔎
Test research
"Use the research skill to give me a state of the art on [topic]." You swap the topic, and you see how deep it goes.
Where to start

One agent tested on a real need beats ten skills you will never open. Pick the task that weighs on you most this week and run the agent that covers it.

How to get value without spreading yourself thin

With 271 skills and 67 agents, the temptation is to try everything on the first night. That is the surest way to close the tool without getting anything out of it. Install the core profile, take inventory, and pick one thing to test: a security review, a build to fix, a state of the art to produce.

Once that use is part of your routine, you add another. That is how a huge config becomes useful: not by admiring it, but by using one agent at a time, on tasks you actually had to do.

Worth remembering

ECC gives you in one command what most people spend months assembling. The value is not in the number of skills, it is in the two or three you actually use.

Want to go further?

And day-to-day, I post one reel a day on Instagram: @quentin_iamarketing