ECC for Claude Code: what it is, what it ships and whether you need it
ECC (affaan-m/ECC) is an MIT-licensed open source pack that turns Claude Code, Codex, Cursor and other coding agents into a team with a method: it ships 72 subagents, 328 skills, 99 commands, per-language rules, memory hooks and the AgentShield security scanner. It installs as the ecc@ecc plugin, and its core loop is plan, test, implement, review, verify and remember.
This is part one of an eight-part guide. Here you will see what is inside the repository, how it is split up, which install profiles exist and who it is for. The next parts go through each piece with tested commands.
What ECC is and where it comes from
ECC started as its author Affaan Mustafa's personal Claude Code setup and now describes itself as an "agent harness operating system". In practice it is one repository with three public names that are easy to mix up:
| Identifier | What it is |
|---|---|
| `affaan-m/ECC` | The GitHub repository (source code, MIT) |
| `ecc@ecc` | The plugin for the Claude Code marketplace |
| `ecc-universal` | The npm package with the guided installer and the `ecc` CLI |
We tested v2.2.3 (commit 4adeee6, November 19, 2026) by cloning the repository and running its CLI in a clean environment. The license is MIT, Copyright (c) 2026 Affaan Mustafa, so you can use and modify it in commercial projects.
The author also sells ECC Pro, a paid GitHub App for private repositories. Everything in this guide is the free part of the repository.
What is inside: the real inventory
We counted the folders in the tested commit:
| Folder | Count | What it is for |
|---|---|---|
| `agents/` | 72 subagents | Planning, per-language code review, build repair, security, architecture |
| `skills/` | 328 skills | TDD, research, security, frontend, data, ML, content, operations |
| `commands/` | 99 commands | Shortcuts like `/plan`, `/code-review`, `/build-fix`, `/save-session` |
| `rules/` | 122 files | Standards loaded always (common) or per file type (TypeScript, Python, Go…) |
| `hooks/` | `hooks.json` + scripts | Automation on session start and end, before compaction, on file edits |
| `mcp-configs/` | 1 default connector | Only `chrome-devtools`; everything else is a skill wrapping a CLI or API |
There are also adapters for other harnesses (.codex/, .opencode/, .cursor/), a Python desktop dashboard (ecc_dashboard.py), three long Markdown guides (shortform, longform and security) and an examples/ folder with real CLAUDE.md files for a Next.js SaaS, a Go microservice, a Django API, a Laravel API and a Rust API.
Of the 72 agents, 62 use the sonnet model, 6 use haiku and only 4 use opus (we read the frontmatter of every file). That is a deliberate cost choice: the expensive model is kept for planning and architecture.
The loop it proposes: from idea to memory
The README sums it up in one line: plan -> test -> implement -> review -> verify -> remember -> improve. Each step has its own piece:
- Plan with
/ecc:plan: the agent restates your request, looks for patterns in your code, splits the work into phases and waits for your confirmation before touching anything. - Test first with the
tdd-workflowskill and thetdd-guideagent: failing test, minimal code, refactor. - Review with
/code-review: a reviewer with a fresh context that is not the one that wrote the code. - Verify with
/quality-gate,e2e-testing, or/build-fixwhen something does not compile. - Remember with
/save-session,/resume-sessionand continuous learning, which turns repeated patterns into "instincts" and later into skills.
Part 3 of this guide walks through this loop on a sample project. If you have used other skill packs, the idea is close to Superpowers, but ECC is much bigger and adds hooks, rules and its own CLI.
Install profiles: you do not need everything
The most expensive mistake with ECC is installing all 328 skills and 72 agents without thinking: the catalog is advertised to the model and takes up context. That is why the installer ships seven profiles:
| Profile | Modules | When to pick it |
|---|---|---|
| `minimal` | 5 | You want rules, agents and commands, no hooks |
| `core` | 6 | Minimal plus hooks |
| `developer` | 9 | The default profile for app work |
| `security` | 7 | Projects where security review comes first |
| `research` | 9 | Research, synthesis and content publishing |
| `full` | 36 | The whole catalog; only if you know what you are doing |
If you are not sure what you need, ecc consult "what you do" recommends components from a plain-language query. We test it in part 4 on the agents, and the result is worth checking by hand.
Which agents it works with
ECC works best with Claude Code, its main target and the only one marked "stable". The README publishes an honest support table:
| Harness | Status |
|---|---|
| Claude Code | Stable (plugin or selective installer) |
| Codex | Supported native plugin |
| Qoder | Supported native plugin |
| Cursor, OpenCode, Copilot CLI | Beta |
| GitHub Copilot Chat | Instructions only, no hooks |
| Gemini, Zed, Qwen, Kimi and others | Experimental |
On native Windows the core works, but continuous learning v2 has open bugs; under WSL it follows the Linux paths.
Who it is for (and who should skip it)
It is a good fit if:
- You already use Claude Code every day and want a repeatable method (plan, test, review) without writing it into every prompt.
- You work in several languages: there are reviewers and build resolvers for TypeScript, Python, Go, Rust, Java, Kotlin, C++, Swift, PHP, Ruby, Dart and more.
- You care about the security of your own agent setup (hooks, MCP, permissions): AgentShield audits it.
Skip it if:
- You are new to Claude Code. Learn what skills are and how to install them first; ECC has so many pieces that it is hard to know what each one does.
- You already run another large skill pack. Mixing two full methods (for example ECC and Superpowers) duplicates instructions and confuses the agent.
- Your token budget is tight and you will not prune the catalog. Start with
minimaland add what you miss.
The 8 parts of this guide
- What ECC is and what it ships (this article).
- Install ECC in Claude Code without duplicates: plugin, npx, Windows, profiles, doctor and uninstall.
- The daily ECC workflow: plan, TDD, code review and build-fix.
- ECC's 72 agents: which to use and when.
- Skills and rules: install only what you need.
- Hooks, memory and continuous learning.
- Saving tokens and context with ECC.
- Securing your agents with AgentShield.
FAQ
Is ECC free? Yes. The repository and the ecc-universal package are MIT. The only paid part is ECC Pro, the GitHub App for private repositories.
Is it the same as "Everything Claude Code"? Same project. Some older posts use the long marketplace identifier; the plugin is now ecc@ecc and the author treats the old name as a legacy alias.
Do I need Node.js? For the guided installer, yes: Node.js 18 or newer, Git and Claude Code 2.1 or newer on your PATH.
Where do I install it safely? Only from the official sources listed in the README: the affaan-m/ECC repository, the npm packages ecc-universal and ecc-agentshield, the ecc@ecc plugin and the ecc.tools website. The author warns about third-party copies carrying malware.
Conclusion
ECC is the most complete pack we have seen for giving a coding agent a method, and also the easiest one to install badly. Keep three ideas: pick a small profile, do not mix install methods and run /ecc:plan before asking for code. Next up is part 2: installing ECC without duplicates.
// source code
Help us grow
codigoagente is free, with no paywalls. If this article helped you, any of these keeps the blog going.
- Share it with someone who'll find it usefulWhatsAppLinkedInXFacebook
- Meet our sponsorSafe Build Labs
More on AI agents

- Skills for your agents
- Claude Code

guide · Marketing psychology guide for AI agents · #3
- Skills for your agents
- Claude Code

guide · Marketing psychology guide for AI agents · #2
- Skills for your agents
- Claude Code