Affaan Mustafa Open-Sources His Claude Code Setup (ECC)
A LinkedIn post by Stanislav Beliaev.
The guy who won Anthropic’s hackathon open-sourced his entire Claude Code setup 🔥
Affaan Mustafa shipped a customer discovery platform that lets founders validate product-market fit through synthetic AI personas in 8 hours, with Claude Code. Then he open-sourced the exact system that made it possible.
The repo is called ECC — and it already hit 210K stars.
Most Claude Code setups are a CLAUDE.md file and a few slash commands. This is something else entirely.
What ECC ships
- 64 specialized subagents — planner, architect, security reviewer, ML pipeline reviewer, language-specific reviewers per stack
- 261 skills across backend, frontend, ML, security, and deployment workflows
- 84 slash commands that work across every supported harness
- Hooks that save session context on shutdown and reload it on start, so Claude doesn’t reset between sessions
- Continuous learning — the system extracts patterns from your sessions and turns them into reusable skills automatically
- AgentShield, a security scanner with 1,282 tests and 102 static analysis rules, scanning your
CLAUDE.md, hooks, MCP configs, and agent definitions for injection risks
Architecture
The architecture is what makes it different. A main session delegates to the right specialist based on the task — the way an actual engineering team splits work — with verification loops and eval checkpoints built in. It works across Claude Code, Cursor, Codex, OpenCode, and GitHub Copilot.
Repo: https://github.com/affaan-m/ecc
Key Points
- Affaan Mustafa won Anthropic’s hackathon building a customer-discovery platform in 8 hours with Claude Code, then open-sourced the setup as ECC (210K+ stars)
- Combines 64 subagents, 261 skills, and 84 cross-harness slash commands
- Sessions persist context across restarts via shutdown/start hooks; continuous learning auto-extracts reusable skills
- AgentShield scans agent configuration files for injection risks
- Commenters note the design mirrors real engineering teams and want guardrails on auto-extracted skills