68 AI Agents. 292 Skills. Two Commands to Install.

One developer built what the major AI labs have not shipped: a free, open-source plugin that installs 68 specialized AI agents and 292 skills directly into your code editor. It crossed 257,000 GitHub stars in under a year.

The project is called ECC. It is free. It works in Claude Code, Cursor, Codex, OpenCode, and Zed. And the people who built Claude Code, Codex, or Cursor did not give it to you.

What Is a Buying Signal Worth?

Before getting to the install, here is why this matters for a business owner rather than just a developer.

Every software-reliant business runs some version of this cycle: something breaks, someone fixes it, the fix creates a new problem, and the cycle repeats. Each round takes time and cost.

ECC puts a specialist in the loop for each step. A planner. A builder. A reviewer. A security auditor. A test runner. Each one handles one phase and hands off to the next. The cycle runs faster, with fewer breaks.

What ECC Does

ECC describes itself as an agent harness operating system. The idea is that AI agents are more effective when they operate in a defined loop with memory, verification, and specialization built in.

The workflow it enforces: plan, then test, then implement, then review, then verify, then remember the lessons, then improve. Not one model doing all of it in one shot. A coordinated sequence.

Representative agents in the system:

• planner: Designs the feature architecture before any code is written
• architect: Makes system design decisions
• code-reviewer: Checks quality and catches errors
• security-reviewer: Runs a vulnerability analysis on the output
• tdd-guide: Enforces test-driven development
• build-error-resolver: Diagnoses and fixes compilation failures
• e2e-runner: Runs end-to-end tests via Playwright
• refactor-cleaner: Removes dead code

Language-specific reviewers cover Go, Python, TypeScript, JavaScript, Java, Kotlin, C++, F#, and Rust.

In use, the commands are conversational. "Plan this feature" and the planner designs the architecture. "Build it" and the builder writes the code. "Secure it" and the security reviewer audits for vulnerabilities. "Test it until 80% coverage" and the test runner runs tests until the threshold is met.

How to Install It

For Claude Code:

/plugin marketplace add https://github.com/affaan-m/ECC
/plugin install ecc@ecc

Universal installer (works across all supported editors):

npx ecc-universal@2.2.1 setup

This is the recommended path if you are not sure which editor you will use or if you want it available in multiple environments. The setup script handles the configuration for each editor automatically.

For Codex:

codex plugin marketplace add affaan-m/ECC && codex plugin add ecc@ecc

System requirements: Node.js 18 or later, Git, and Claude Code 2.1 or later if you are using the Claude Code plugin path.

Done. The agents are available immediately after install.

What You Get Back

1. 68 agents covering planning, building, review, security, testing, and refactoring
2. 292 skills spanning TDD, research, security, documentation, frontend, data, machine learning, and operations
3. 94 commands for quick access to the most common agent actions
4. Language-specific reviewers for nine programming languages
5. MIT license: free for open-source and commercial use

The Commands in Practice

After install, open any project in your editor and run:

plan this feature: [describe the feature]

The planner agent returns an architecture breakdown before any code is written. Review it, adjust the scope if needed, then:

build it

The builder implements the plan. When it finishes:

secure it

The security reviewer audits the implementation and reports vulnerabilities. Then:

test it until 80% coverage

The test runner writes and runs tests until the coverage threshold is met.

The whole loop runs inside your editor. The output of each step is the input to the next.

Honest: What This Does Not Replace

ECC coordinates agents better than a single model prompt would. It does not replace your judgment about what to build.

If you hand it a vague feature description, the planner produces a vague architecture. If you hand it a specific, constrained spec, the output is specific and constrained. The quality of the input still determines the quality of the output.

The test runner also works only as well as your test suite setup allows. If your project has no testing infrastructure, the tdd-guide helps establish it, but there is setup work before the "test until 80%" command does what you expect.

Install Only From the Official Repository

The README notes this explicitly and it is worth repeating: install ECC only from the verified sources.

• GitHub: github.com/affaan-m/ECC
• npm: ecc-universal or ecc-agentshield

Third-party mirrors of popular open-source packages are a known attack vector. Do not install from any other source.

One Action This Week

Install ECC. Open a feature or bug you have been putting off. Run plan this feature with a two-sentence description of the problem.

Read the architecture the planner returns. If it identified an approach you had not considered, or a dependency you would have missed, the five-minute install paid for itself.

The GitHub repository is linked at the bottom of this page.