Everything Claude Code (ECC): what it is, how to install it and when not to use it

Everything Claude Code, now called ECC, is a free layer you install on top of Claude Code that imposes a seven-step way of working: plan, test, implement, review, verify, remember and improve. It was created by Affaan Mustafa after winning an Anthropic hackathon and now has 274 thousand stars on GitHub. Here is what it includes, how to install it and when not to use it.

If you're interested in changing Claude Code's rules another way, read the analysis of Claude Code mods.

Where Everything Claude Code comes from

In September 2025, Anthropic and Forum Ventures held a hackathon in New York with more than a hundred teams. According to the project's accounts, Affaan Mustafa built an app, zenith.chat, in eight hours without writing the code by hand. He won, and the prize was 15 thousand dollars in API credit. What mattered came next: instead of keeping his method to himself, he published the setup he had spent ten months refining with Claude Code. It was called Everything Claude Code; today it is called ECC.

What ECC is and what pieces it includes

ECC is not a new model or a separate app: it is a layer on top of Claude Code. To apply its seven steps it brings four kinds of pieces:

• Agents: specialists; one plans, another reviews, another audits security.
• Skills: instructions Claude loads only when it needs them.
• Hooks: automatic actions that run before or after each tool.
• Rules: what must always be followed.

When it launched it had 30 agents, 135 skills and 60 commands. Today it has 68 agents, 293 skills and 94 commands. Stars went from 117 thousand in the first weeks to 274 thousand, with more than 40 thousand forks. The current version is 2.2.3, from October 1, under the MIT license.

How to install ECC in two minutes

Inside Claude Code you add the plugin marketplace using the official repository's address, install the plugin called ECC, check that it shows up in the list and start with the planning command, typing in quotes what you want to build. You need Claude Code 2.1 or higher, because from there hooks load on their own. There is another route: a wizard you run with npm that detects whether you use Claude Code, Codex or another tool. Pick one route, not both, because you duplicate everything.

A real task: adding a sign-up form

First the planning command: instead of writing code, the planner agent returns a plan with files to touch, risks and order of steps. You read it and correct it before a single line of code exists; if you don't like it, it redoes it in seconds. Then the test-first skill comes in, writing tests that fail on purpose, and the implementation continues until they pass. At the end you call the code review, and if something doesn't compile, the build-error agent fixes it. You stop typing and start approving.

The smartest piece: a reviewer with fresh eyes

When an AI reviews its own work in the same conversation, it carries all its earlier decisions and tends to justify them. ECC launches the reviewer in a clean session, without that history, like a colleague arriving in the morning. It also saves a summary of each session on your computer and detects patterns you repeat as instincts with a confidence level. What is learned without review does not become a rule on its own: you approve it first.

The hidden cost: the context window

Claude has a 200 thousand token window. The project's own documentation warns that with all connectors turned on at once it drops to about 70 thousand: 130 thousand tokens lost, 65% before writing a line. That is why the project recommends Sonnet instead of Opus, which it says cuts 60% of the cost, and lowering thinking from 32 thousand to 10 thousand tokens, 70% less. These are the project's figures, not measured by WAiBOT.

Do 274 thousand stars prove it works?

No. A star is a bookmark. As far as we found, nobody has published an independent measurement showing Claude writes better code with ECC. What is defensible is something else: planning before building, testing first and reviewing with fresh eyes are practices that already work without this tool. ECC packages and automates them: if you already do them by hand, it saves you work; if not, it forces you to do them.

When it breaks and what the risks are

• Installing everything: 293 skills at once bloat the context; install selectively.
• Mixing install methods: you duplicate hooks and skills, and the authors themselves report it as a cause of incidents.
• Expecting parity everywhere: in Claude Code it is stable, in Cursor it is beta and in GitHub Copilot it is instructions only, with no hooks or agents. One main maintainer supports seven tools.
• Fake copies: hooks are code that runs on your machine, and the Cloud Security Alliance documented 7,600 fake repositories posing as skills and tool servers. Install only from Affaan Mustafa's official GitHub repository, the official npm package or the ECC.tools page. The project includes AgentShield, with 102 rules, which looks for secrets and traps in your setup, but it reviews, it doesn't isolate.

How to apply it in your business?

If you use Claude Code almost daily, start with the minimal profile: three or four skills, the planner and the review. Use it for a week on a real project and budget an hour to install it, trim what you don't use and understand what each agent does. If you notice it saves you redoing work, add more; if not, uninstall it with the project's safe uninstall. The alternative is your own instructions file with three rules of yours, which takes ten minutes and bloats nothing. It wins if your workflow is simple; ECC wins when you work in a team or already repeat the same method in every project. It isn't for you if you don't use Claude Code almost daily, if your work is twenty-line scripts or if you don't pay for a plan with headroom.

Frequently asked questions

What is Everything Claude Code (ECC)?

It is a layer of agents, skills, hooks and rules installed on top of Claude Code that imposes a seven-step method. It is not a model or a separate app.

Is it free?

The repository is free and MIT-licensed. The project charges 19 dollars per seat per month for a version for private repositories.

Which Claude Code version do I need?

2.1 or higher, because from there hooks load on their own.

Does it make Claude write better code?

There is no independent measurement showing it. What it automates are practices that already work: planning, testing first and reviewing with fresh eyes.

Where do I install it safely?

Only from the official GitHub repository, the official npm package or ECC.tools.

Conclusion

Everything Claude Code is a method, not magic: plan first, test first and review with fresh eyes, already packaged and free. What is valuable is not the 274 thousand stars but that it forces you to work like a good team. If you install it, do it only from the official repository and start small. Tell me in the video comments where your AI workflow fails most: planning, testing or reviewing.