ECC Tools

ECC Tools (Everything Claude Code) is an agent harness engineering toolkit — a system for building the infrastructure that guides, controls, and optimizes AI coding agents so they produce reliable output without constant human supervision. Rather than treating AI code editors as standalone tools, ECC treats them as agents that need a harness: strict architecture enforcement, fast feedback loops, managed context, and learned conventions. It analyzes repository git history using AI to extract coding patterns and workflows, then generates structured skill files that teach Claude Code how your team works.

Harness engineering is the emerging practice of constructing scaffolding around AI agents so they perform coding tasks accurately and consistently. ECC Tools is one of the most complete implementations of this approach, consisting of two complementary products: an automated GitHub App that creates pull requests with extracted skills and instincts, and a large open-source repository of production-ready agents, skills, hooks, commands, rules, and MCP configurations. Together they form the harness that keeps AI agents on track.

The core problem harness coding solves is that every codebase contains implicit knowledge — architectural decisions, naming conventions, testing patterns, deployment workflows — buried in commit history. Without a harness, each AI coding session starts from scratch. ECC Tools makes this knowledge explicit and machine-readable, giving agents computed context at session start so they follow your conventions from the first line of code.

• Harness-First Architecture - Designed around harness engineering principles: strict rules enforced by linters, hooks for fast feedback loops, context management to prevent agent drift, and verification loops that maintain quality gates
• Automated Skill Generation - Install the GitHub App, comment /ecc-tools analyze on any issue, and receive a PR with SKILL.md and instincts extracted from up to 5,000 commits of git history
• Continuous Learning v2 - An instinct-based learning system that observes your workflow and creates atomic, confidence-weighted behaviors that evolve over time, allowing the harness to improve itself as your codebase grows
• Specialized Subagents - Pre-built agents for delegated harness coding tasks including code review, security analysis, test-driven development, build error resolution, architecture planning, and documentation syncing
• Cross-Platform Harness Support - Works with Claude Code, Cursor, OpenCode, and Codex from a single configuration source, providing consistent harness behavior across all major AI coding tools
• Hook System - Trigger-based automations forming the fast feedback loop layer of the harness, including memory persistence across sessions, strategic compaction suggestions, and session evaluation for pattern extraction
• Slash Commands - Quick-execution commands like /tdd, /plan, /code-review, /e2e, and /security-review for common development workflows
• AgentShield Integration - Built-in security scanning with 102 rules and over 1,200 tests that audit AI agent configurations for prompt injection vulnerabilities and misconfigured guardrails
• Pattern Detection - Analyzes commit history to detect coding conventions, architectural patterns, folder structures, and recurring processes that should be encoded into the harness
• Instinct Evolution - The /evolve command clusters related instincts into full skills, letting the harness progressively build domain expertise from observed patterns
• Language-Specific Skills - Pre-built harness skill sets for TypeScript, Go, Python/Django, Java Spring Boot, and Swift covering coding standards, backend patterns, and testing workflows
• Context Window Management - Strategic compaction skills and model selection guidance to optimize token usage, a critical harness engineering concern that prevents agent context overflow
• Rules System - Always-enforced guidelines for security, testing, coding style, and git workflow that form the strict architecture layer of the harness
• Verification Loops - Checkpoint and continuous evaluation systems implementing the fast feedback loop principle central to harness engineering
• Multi-Agent Orchestration - PM2 process management and parallel execution using git worktrees for complex multi-service harness coding workflows
• MCP Configurations - Pre-configured server setups for GitHub, Supabase, Vercel, Railway, and other common development services

To get started, install globally with npm i -g ecc-universal, or add the plugin in Claude Code with /plugin marketplace add affaan-m/everything-claude-code. The GitHub App can be installed at github.com/apps/ecc-tools for automated repository analysis.