ECC installs an engineering workflow into a coding agent instead of leaving it to be re-explained in every prompt. The sequence it enforces is plan, test, implement, review, verify, remember, improve: plans become editable artifacts, test-driven development becomes a gated red-green-refactor loop with captured evidence, and a fresh-context reviewer inspects the result before verification runs.
The distribution is a set of components you install per harness: 68 agents specialized for planning, review, build repair, security, architecture, and domain work; 286 skills covering areas such as test-driven development, research, documentation, frontend, data, and operations; 94 maintained command shims; selectable language rules; and hooks that run deterministic checks outside the model context. Skills load when a task needs them, rules are always loaded and therefore installed selectively, and agents keep their own context and tool permissions.
ECC works with Claude Code, has a supported Codex sync path, and provides capability-limited adapters for Cursor, OpenCode, Gemini, Zed, GitHub Copilot, Antigravity, and Qwen, with a support matrix that records which features exist where. Installation goes through a guided package command or the Claude Code plugin, and stacking two install methods in the same harness is called out as a source of duplicated skills, commands, and hooks.
AgentShield, the bundled security scanner, treats the agent's own configuration as an attack surface. It scans CLAUDE.md, settings files, MCP configuration, hooks, agent definitions, and skills against 102 static analysis rules across secret detection, permission auditing, hook injection analysis, MCP server risk profiling, and agent config review, and reports to the terminal, JSON, Markdown, or HTML with exit code 2 on critical findings for build gates.
Two further pieces round out the system: a Memory Vault that stores project, team, and user memories as inspectable Markdown so several harnesses can hand work to each other, and a skill creator that analyses a repository's git history locally to generate SKILL.md files and instinct collections, with a GitHub App path for larger histories and pull-request automation.
Features
- Gated TDD workflow: red-green-refactor with captured failing-test evidence before implementation
- Fresh-context review: a separate reviewer agent inspects work the implementing context produced
- Agent library: 68 scoped subagents for planning, review, build repair, security, and language-specific work
- Skill library: 286 on-demand workflows spanning testing, research, security, docs, data, and operations
- Selective rules: always-loaded standards split into common and per-language directories
- Hooks: event-triggered scripts that enforce checks outside the model context
- AgentShield: 102-rule scanner for prompts, hooks, MCP config, permissions, secrets, and agent files
- Memory Vault: shared Markdown memory and handoffs across harnesses, scoped to project, team, or user
- Skill creator: generates skills and instincts from git history, locally or through a GitHub App
- Cross-harness install: Claude Code plugin, Codex sync, and adapters for several other harnesses
Runs in
Claude Code
Terminal-based agentic coding tool that reads your codebase and runs tasks through natural-language commands
Codex
Lightweight coding agent that runs in your terminal
Hermes
Self-improving AI agent with a learning loop that creates and refines skills, recalls past sessions, and runs across chat platforms
OpenClaw
Self-hosted single-operator assistant runtime that links models, tools, and messaging channels through one local Gateway
