Frequently Asked Questions

ECC (The Agent Harness Performance Optimization System) is a comprehensive toolkit that transforms AI coding assistants into disciplined engineering partners. It provides a structured system where agents plan before building, test their implementations, review their own work from fresh contexts, and learn from repeated successes to create reusable skills and workflows.

Instead of rebuilding the engineering process in every prompt, you install ECC once and make it an integral part of how your agent works. It optimizes the context window while persisting everything else that matters.

ECC is designed for developers, development teams, AI/ML practitioners, and researchers who want to leverage AI coding assistants while maintaining engineering discipline and code quality.

Specifically, it's ideal for:

• Software Engineers who want to build better code faster with TDD workflows and automated quality gates.
• Development Teams that need to maintain consistent standards across projects.
• AI/ML Developers who require specialized workflows for data contracts, evals, and deployment.
• Researchers who want to accelerate their work with first development practices.

ECC works best with Claude Code today and has supported paths for other popular AI coding assistants:

• Claude Code - Full feature support with native plugin integration
• Codex - Native plugin support with marketplace integration
• Cursor - Project-local adapter with agent definitions
• OpenCode - Plugin payload build and full install support
• Kimi Code - Project-local configuration and skills
• Gemini CLI - Project-local adapter
• Zed - Project-local adapter
• GitHub Copilot - Built-in support through instruction files
• Others - Antigravity, Qwen, Hermes, OpenClaw, CodeBuddy, JoyCode

Feature parity varies by platform. Please check the support status before assuming full compatibility.

ECC offers several installation methods depending on your AI coding assistant:

Recommended for Claude Code:

npx ecc-universal@2.2.2 setup

This guided setup wizard will help you configure ECC in your preferred scope (user, project, or local).

For other assistants:

• Codex: codex plugin marketplace add affaan-m/ECC
• Cursor: ./install.sh --profile minimal --target cursor
• Kimi Code: ./install.sh --profile minimal --target kimi

Choose only one installation method per assistant to avoid conflicts. Do not stack multiple installations.

ECC offers several installation profiles to match different needs:

• core - The default profile with essential agents, skills, commands, and optional hooks
• minimal - Low-context install with no hook runtime, only rules and core components
• full - Complete installation with all components including hooks and MCP servers

You can also customize your installation by selecting specific skills or components:

./install.sh --target claude --skills tdd-workflow,security-review

Start with the workflow you need. Here are some common starting points:

• Building a feature: /ecc:plan "describe the feature", then use the tdd-workflow skill
• Fixing a bug: Reproduce it with a failing test, then use tdd-workflow
• Reviewing code: /code-review for a fresh-context review
• Repairing a build: /build-fix
• Checking security: /security-scan

Skills are the primary workflow surface. Commands remain convenient entry points and compatibility shims.

ECC includes 68 specialized subagents for delegation, including:

• planner - Creates implementation blueprints for features
• architect - Handles system design decisions
• tdd-guide - Enforces test-driven development
• code-reviewer - Reviews code for quality and best practices
• security-reviewer - Analyzes code for vulnerabilities
• build-error-resolver - Fixes compilation and build errors
• refactor-cleaner - Identifies and removes dead code
• doc-updater - Synchronizes documentation
• Language-specific reviewers (Go, Python, TypeScript, Java, etc.)
• Domain-specific agents (ML, database, HarmonyOS, etc.)

ECC provides 292 reusable workflows loaded on demand, covering:

• TDD Workflow - Write tests first, implement, refactor
• Security Review - OWASP Top 10 audit checklist
• Search-First - Research-before-coding workflow
• Continuous Learning - Pattern extraction from sessions
• E2E Testing - Playwright end-to-end testing patterns
• ML Engineering - Data contracts, evals, deployment
• Backend Patterns - API, database, caching patterns
• Frontend Patterns - React, Next.js patterns
• Language-specific - Go, Python, TypeScript, Java, etc.
• Framework-specific - Django, Rails, Spring Boot, etc.

AgentShield is ECC's integrated security scanning system that helps ensure secure development practices. It scans for:

• Prompt injection attempts
• Malicious hook configurations
• Unsecure MCP server connections
• Excessive permission grants
• Secrets and sensitive data in code
• Potentially dangerous agent behaviors

You can run security scans with /security-scan or use the standalone agentshield scan --path . command.

ECC's memory and learning system allows it to adapt to your coding patterns over time. Key features include:

• Session Summaries - Automatic saving of key insights from each session
• Instincts - Pattern extraction with confidence scoring
• Continuous Learning - Automatic skill creation from repeated successes
• Context Controls - Management of long-term context

You can view learned instincts with /instinct-status, export them with /instinct-export, and import them with /instinct-import.

Hooks are trigger-based automations that fire on specific events. For example:

• Pre-Tool Use - Checks before executing code edits
• Post-Tool Use - Verifications after actions
• Session End - Saves state and learns patterns
• Stop - Handles session interruptions

Hooks can enforce coding standards, suggest improvements, or perform automated tasks. They're configured through the hooks runtime, which can be enabled during installation.

Rules are always-follow guidelines organized into language-agnostic and language-specific directories:

• common/ - Universal principles (always install)
• typescript/ - TypeScript/JS specific patterns
• python/ - Python specific patterns
• golang/ - Go specific patterns
• swift/ - Swift specific patterns
• php/ - PHP specific patterns
• arkts/ - HarmonyOS / ArkTS patterns

Rules are automatically loaded and provide consistent standards across your projects. You can install them globally or project-specifically.

Yes! ECC is MIT-licensed open source and remains free for personal and commercial use. The core functionality is available at no cost through our self-hosted installation methods.

We also offer ECC Pro, a hosted GitHub App that provides additional features for private repositories:

• Private repository support
• Advanced security scanning
• Priority support
• Collaboration features

ECC Pro is available for private repositories starting at $19 per seat per month. Pricing scales based on the number of seats and features required.

You can install the free GitHub App for private repositories to explore basic features before upgrading to Pro.

For detailed pricing information, please visit our pricing page.

ECC Pro accepts all major credit cards including Visa, MasterCard, American Express, and Discover. We also support payment through GitHub Sponsors for open-source contributions.

For enterprise customers, we offer invoice-based billing with NET 30 terms. Contact our sales team for custom pricing and enterprise agreements.

You can support ECC development in several ways:

• GitHub Sponsors - Direct financial support for the maintainer
• ECC Pro - Upgrade to the paid plan for additional features
• Community Contributions - Help with documentation, testing, or feature development
• Feedback - Report issues and suggest improvements

All sponsors are recognized on our website and in the project documentation. Your support helps us continue developing and maintaining ECC.

ECC offers several support channels:

• Documentation - Comprehensive guides and API references
• Discord Community - Active community for questions and discussions
• GitHub Issues - Bug reports and feature requests
• Email Support - For ECC Pro customers
• GitHub Discussions - Community Q&A

ECC Pro customers receive priority support with guaranteed response times.

ECC is designed to respect your privacy. The self-hosted version collects minimal data:

• Anonymous usage statistics (optional)
• Session summaries (with your explicit permission)
• Error reports (for improving the product)

No sensitive code, proprietary information, or personal data is ever transmitted without your explicit consent. All data processing happens locally whenever possible.

For the self-hosted version of ECC, all data remains on your local machine:

• Configuration files in ~/.claude/ (Claude Code)
• Skills and agents in your project directory
• Memory and learning data locally stored
• Session summaries saved locally unless explicitly shared

ECC Pro customers can choose to store certain data in the cloud, but all data is encrypted at rest and in transit.

ECC follows industry-standard security practices:

• Regular security audits of the codebase
• Vulnerability disclosure program for researchers
• Timely patches for discovered issues
• Secure-by-default configuration
• AgentShield scanning for potential threats

We take security seriously and encourage responsible disclosure of any vulnerabilities. Please see our security policy for details.

ECC is designed with security in mind and follows best practices:

• No hardcoded secrets or keys
• Input validation and sanitization
• Secure configuration management
• Regular dependency updates
• Minimal required permissions

While we don't currently have formal third-party certifications, we maintain high security standards and are working toward compliance with relevant frameworks.

ECC has minimal system requirements:

• Operating System - Windows 10+, macOS 10.14+, or Linux
• Node.js - Version 18 or newer (for the universal package)
• Memory - 4GB RAM minimum (8GB recommended)
• Storage - 500MB free space for installation
• Internet - Required for initial setup and updates

After installation, ECC can work offline for most operations.

No, ECC does not require GPU acceleration. It runs entirely on your CPU and works with any AI coding assistant that has internet access to the provider's API.

If you want to use self-hosted models, you'll need appropriate GPU capacity for those models, but ECC itself doesn't require GPU resources.

Yes! ECC works through each harness's normal configuration, so you can use custom AI models or endpoints. For example:

• Self-hosted models through compatible gateways
• Custom API endpoints with your own authentication
• Open-source models running on your infrastructure

Simply configure your AI coding assistant to use your preferred model or endpoint, then install ECC as usual. ECC's workflows and skills are model-agnostic.

ECC can be installed alongside multiple AI coding assistants, but you should use separate installation methods for each:

• Claude Code: Use the plugin installation
• Codex: Use the marketplace plugin
• Cursor: Use the project-local adapter
• Kimi Code: Use the project-local configuration

Choose one installation method per assistant. Do not stack multiple installations for the same assistant, as this can cause conflicts.