ECC: The Agent Harness Performance Optimization System for LLMs

Summary

ECC is an advanced agent harness performance optimization system, providing a coordinated engineering system and toolbox for LLM development. It enables agents to plan, test, implement, review, and learn effectively, enhancing productivity and security for AI-driven coding workflows across platforms like Claude Code, Codex, and Cursor.

Use at your own risk

OSRepos shares public repositories for knowledge and discovery only. Any installation, execution, configuration, or use of code from these repositories is the user's own responsibility. Always review the repository, source code, dependencies, licenses, and security implications before running or installing anything. OSRepos is not responsible for issues, damages, or losses resulting from third-party repositories.

Introduction

ECC is a comprehensive agent harness performance optimization system, designed to elevate the capabilities of large language models in coding environments. It provides a coordinated engineering system and toolbox, integrating skills, instincts, memory, and robust security features for a research-first development approach. ECC supports a wide range of platforms, including Claude Code, Codex, Opencode, Cursor, and other agent harnesses. This MIT-licensed open-source project, primarily written in JavaScript, offers a rich ecosystem of 68 specialized agents, 286 reusable skills, 94 command shims, and powerful hooks for runtime automation, continuous learning, and AgentShield security scanning.

Installation

Installing ECC is streamlined, but it is crucial to choose only one installation path per harness to avoid conflicts.

For Claude Code, the recommended approach is to use the guided package setup:

npx ecc-universal setup

Alternatively, you can use the native Claude plugin commands:

/plugin marketplace add https://github.com/affaan-m/ECC
/plugin install ecc@ecc

ECC 2.2 also offers guided setup for Codex and Kimi Code via npx ecc-universal install --guided. For other agents and editors like Cursor, OpenCode, Gemini, Zed, and GitHub Copilot, specific selective installers or instruction layers are available, as detailed in the official documentation. Remember, do not stack installation methods for the same harness.

Examples

ECC empowers your agent with structured workflows for common development tasks:

Building a Feature: Start with /ecc:plan "describe the feature" to create an implementation blueprint, then activate the tdd-workflow skill to enforce a test-driven approach.
Fixing a Bug: Reproduce the bug with a failing test using the tdd-workflow skill, implement the fix, and verify the test passes.
Reviewing New Code: Use /code-review for a fresh-context review, catching regressions and blind spots.
Repairing a Build: The /build-fix command helps systematically resolve build errors.
Finding Security Vulnerabilities: Run /security-scan to perform an OWASP Top 10 audit using AgentShield.

Why Use ECC?

ECC transforms unstructured agent interactions into a coordinated engineering process, offering significant advantages:

Structured Planning: Plans become editable artifacts before implementation, rather than disappearing into chat history.
Enforced TDD: Test-Driven Development is a gated RED -> GREEN -> REFACTOR workflow with clear evidence, not just an instruction the model might forget.
Fresh Context Review: A dedicated, fresh-context reviewer identifies regressions and blind spots, improving code quality.
Intelligent Memory: Sessions are distilled into summaries, instincts, and reusable skills, moving beyond enormous transcripts.
Automated Quality Checks: Hooks enforce deterministic checks outside the prompt, ensuring consistent quality.
Agent Configuration Security: AgentShield scans the harness itself as an attack surface, enhancing security.

Furthermore, ECC's Memory Vault allows sharing context between different harnesses, providing a local, inspectable Markdown format for durable context and handoffs across Claude, Codex, Hermes, OpenClaw, and Kimi.