ECC Gives Claude Code and Codex a Ready-Made Workflow: 72 Agents and 328 Skills in One Install
Coding agents like Claude Code have to be reminded of the process in every new chat: plan first, then tests, then review. The ECC kit installs once and brings that process pre-built: 72 narrow agents (planning, review, security) and 328 skills, which are ready-made instructions such as TDD. The repository gained 1,183 GitHub stars in a single day, and its MIT license is permissive. If you copy the rules into prompts, test the kit in a sandbox repo first.
AI-processed from GitHub Trending AI; edited by Hamidun News
If you work with Claude Code, Codex, or Cursor, you know the routine: in every new chat, you have to remind the agent that it needs a plan first, then tests, and that it shouldn't touch code it wasn't asked to change. The affaan-m/ECC project aims to eliminate that repetition. It contains 72 agents (narrow "specialists" in planning, review, security, and other tasks) and 328 skills (ready-made instructions such as "how to write code test-first" or "how to write documentation").
You install it once, and the process becomes part of the agent rather than part of every prompt. In one day, the repository climbed to 1,183 stars on GitHub Trending, and the code is released under the MIT license, so you can use and modify it freely.
What's inside ECC
The authors describe the project as an optimization system for the agent harness. The harness is everything around the model: rules, tools, memory, and checks. ECC bundles these parts into a single kit. Besides the 72 agents and 328 skills, it includes 99 legacy command shims (old short commands kept for compatibility), hooks, rules, memory, and continuous learning.
The README summarizes the working cycle in one line: plan, test, implement, review, verify, remember, improve. In everyday terms, you ask for a field to be added to an API. The agent writes a plan first, then tests, then code. After that, it reviews its own work from a clean context, meaning without the misconceptions that have built up in the chat. At the end, it saves useful information to memory and turns repeatedly successful solutions into reusable skills.
AgentShield deserves a separate mention. It is a security scanner that checks prompts, hooks, MCP configuration, access permissions, secrets, and agent files. If you connect many third-party tools, that kind of verification has been overdue.
Supported environments
ECC works best with Claude Code. There are native installation paths for Codex and Qoder. For Cursor, OpenCode, Gemini, Zed, GitHub Copilot, Antigravity, Qwen, and other environments, there are adapters with limited capabilities. The authors themselves warn that before counting on the full feature set, you should check the compatibility table. The project is maintained by a single maintainer, who ships updates weekly for seven environments.
Version 2.2 added a step-by-step installation wizard for Claude Code, Codex, and Kimi Code. You need Node.js 18 or newer. The Claude plugin additionally requires Git and Claude Code 2.1 or newer. The basic command is npx [email protected] setup. For multiple agents at once, there is an install --guided mode: it shows the channels and installation paths and asks for one final confirmation. Before the real installation, you can do a dry run with the --dry-run flag.
Pitfalls and money
The project's main rule is one installation method per environment. If you layer the Claude plugin on top of a full manual installation, skills, commands, and hooks can end up duplicated. Installing into several different environments at once is safe. For a lightweight setup, there is a minimal profile. Hooks require explicit consent, and that is the right call: they execute code on your machine.
There is also a security warning. Install ECC only from the official repository. The project does not verify third-party re-uploads or unofficial mirrors, and they may contain malicious code. Pinning a version is not the same as an audit: you need to verify the release sources and the package integrity in the registry yourself.
On licensing, things are simple. The repository is MIT-licensed. For private repositories, there is a paid GitHub app, ECC Pro, at $19 per seat per month. The authors say this money funds the open part of the project.
What it means in practice
ECC is neither a new model nor yet another agent. It is a process scaffold that many teams are currently assembling by hand from files containing rules, skills, and hooks. If you already have a well-tuned setup, ECC may be superfluous. Hundreds of skills and dozens of agents can easily turn into noise in the context, and rules that load all the time consume the context window. The authors themselves advise starting with general rules and a single package for your stack.
If you are only now building discipline around coding agents, the project gives you a ready-made starting point. Install it in a test repository, do a dry run, keep only the skills you need, and check whether you have fewer manual reminders in your prompts.
- Nov 21, 2026Anthropic назначила цену IPO в $83 за акцию: объём размещения около $81 млрд
- Nov 21, 2026Anthropic to Unveil Claude for Factories: 17 Ready-Made Skills and 4 Integrations with Shop-Floor Software
- Nov 21, 2026AI Coding Assistant Under $10: Cursor at $8, Xiaomi MiMo at $6 a Month
- Nov 21, 2026Anthropic goes public and raises a record $107.3 billion
Key Questions Answered
What's inside ECC?
The authors describe the project as an optimization system for the agent harness. The harness is everything around the model: rules, tools, memory, and checks. ECC bundles these parts into a single kit. Besides the 72 agents and 328 skills, it includes 99 legacy command shims (old short commands kept for compatibility), hooks, rules, memory, and continuous learning. The README summarizes the working cycle in one line: plan, test, implement, review, verify, remember, improve.
What are the key details on Supported environments?
ECC works best with Claude Code. There are native installation paths for Codex and Qoder. For Cursor, OpenCode, Gemini, Zed, GitHub Copilot, Antigravity, Qwen, and other environments, there are adapters with limited capabilities. The authors themselves warn that before counting on the full feature set, you should check the compatibility table. The project is maintained by a single maintainer, who ships updates weekly for seven environments.
What are the key details on Pitfalls and money?
The project's main rule is one installation method per environment. If you layer the Claude plugin on top of a full manual installation, skills, commands, and hooks can end up duplicated. Installing into several different environments at once is safe. For a lightweight setup, there is a minimal profile. Hooks require explicit consent, and that is the right call: they execute code on your machine. There is also a security warning. Install ECC only from the official repository.
Build & Ship Production AI Agents Without Boilerplate Code
Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.