If you checked GitHub Trending this week, one project stood out: Everything Claude Code — or ECC for short. Nearly 200,000 stars. Thirty thousand forks. And it all started with a hackathon.
The author, Affaan Mustafa, is a San Francisco-based developer. Last September, Anthropic and Forum Ventures co-hosted a competition: "Build a company from zero to one using AI Agents." The rules were brutal — compress weeks of work into a single day. Customer discovery, product scoping, prototyping, development — all of it.
Affaan and his teammate won. Their product, zenith.chat, is an AI-powered customer research platform. And here's the wild part: Affaan says he barely wrote any code. Claude Code did almost everything.
What made the difference? A workflow he'd been refining for over ten months. After the hackathon, he open-sourced it. That workflow became ECC.
What's Actually Inside ECC?
Three months and many iterations later, ECC ships with:
- 63 specialized Agents — each with a distinct role: architecture design, test writing, code review, security auditing
- 249 Skills — reusable workflows and domain knowledge modules
- 79 Command shims — legacy shortcuts for quick operations
When you install ECC on Claude Code, you're not getting a config tweak. You're booting up a full AI development team.
The Secret Sauce: On-Demand Loading
Two hundred and forty-nine skills sounds like context-window suicide, right? That was my first thought too.
ECC solves this with lazy loading. Skills aren't dumped into context all at once. If you're working on a TypeScript project, the TypeScript review agent activates. When you start writing Python tests, the TDD agent kicks in. Everything else stays dormant.
This is how they fit a massive skill library inside Claude without blowing up the token budget.
AgentShield: Security That Ships With Your Code
Here's something that deserves its own spotlight. We've all had that moment of panic — did I just commit an API key?
ECC includes AgentShield, a security auditor that scans your codebase in milliseconds. It catches credential leaks, hook injection risks, and MCP server misconfigurations before they reach your repo.
Flip on --opus mode, and things get even more interesting:
- A red-team Agent hunts for vulnerabilities
- A blue-team Agent patches them
- An auditor Agent compiles the full report
It's a miniature security operations center, running inside your terminal.
Installation: Two Lines
Installing ECC as a plugin is dead simple:
# Add the marketplace
/plugin marketplace add https://github.com/affaan-m/ECC
# Install the plugin
/plugin install ecc@ecc
Manual installation is also available if you want granular control over which agents and skills you pull in. One catch: rules files must be copied manually — the plugin system doesn't distribute them automatically.
ECC isn't locked to Claude Code either. It works with Codex, Cursor, OpenCode, Gemini CLI, and several other AI coding harnesses.
The 200K-Star Question
ECC isn't without its critics. The most common complaint: "There's too much stuff I'll never use." Affaan is upfront about this — the config is tailored to his workflow. His advice: pick what fits, delete what doesn't, build your own.
And that, honestly, is the point.
AI coding tools have evolved from autocomplete engines into orchestratable multi-agent systems. Anthropic's own 2026 Agentic Coding Trends Report predicts that single agents will give way to coordinated teams — exactly the architecture ECC embodies right now.
The configuration layer — skills, rules, memory, security — is becoming just as important as the model itself. ECC's trajectory from a hackathon project to 198K stars proves this isn't a niche opinion. It's where the industry is heading.
Maybe the real differentiator, once models reach a certain threshold, won't be the model at all. It'll be how we tune them.
GitHub: github.com/affaan-m/ECC
What's your AI coding setup look like? Drop a comment — I'd love to hear what's working (and what's not).
Top comments (0)