affaan-m/ECC

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

What it solves

ECC is a "harness-native operator system" designed to standardize and optimize how AI agents (like Claude Code, Cursor, and GitHub Copilot) operate within development environments. It solves the problem of inconsistent agent behavior, token waste, and lack ofof persistence across different AI coding assistants by providing a reusable layer of skills, memory optimization, and security scanning.

How it works

ECC acts as a substrate for AI agents, providing a set of pre-defined agents, skills, hooks, and rules. It integrates across multiple "harnesses" (the AI tools themselves) through a plugin system or manual installation. Key technical components include:

  • Control Plane: A system (with a Rust-based prototype in v2.0) that manages sessions, status, and daemon processes.
  • Hooks & Rules: Runtime enforcement and behavioral guidelines that ensure agents follow specific patterns (e.g., security or language-specific rules).
  • State Store: A SQLite-based system for recording session history and managing state.
  • Skills & Instincts: Reusable, evolving patterns that agents can extract from sessions to improve their performance over time.
  • AgentShield: An integrated security scanner to identify attack vectors and CVEs.

Who it’s for

Developers and engineering teams who use AI agents for coding and want to production-ready workflows, better token efficiency, and cross-platform consistency across different AI harnesses.

Highlights

  • Cross-Harness Compatibility: Works with Claude Code, Cursor, OpenCode, Codex, Gemini, Zed, and GitHub Copilot.
  • Extensive Library: Includes hundreds of specialized skills and agents for various languages (TypeScript, Python, Go, Java, etc.) and domains.
  • Memory & Token Optimization: Features hooks for automatic context saving/loading and system prompt slimming.
  • Security Integration: Built-in security scanning via AgentShield.
  • Multi-Language Support: Rules and patterns for over 12 different language ecosystems.