ECC: The 291K-Star Toolkit That Supercharges Claude Code

Claude Code already writes code. The problem it does not solve is discipline: every session starts with you re-explaining that tests come first, that changes need a review, and that secrets stay out of commits. ECC — short for Everything Claude Code — is the open-source project that turns those reminders into a resident system. It has passed 291,000 stars on GitHub, making it the most-starred Claude Code configuration project on the platform, according to coverage of its rise. This is what it actually installs, what it costs, and where the catches are.

ECC is not a model and not a rival to Claude Code. It is a layer on top of it: 71 specialised agents, 311 skills, 97 legacy command shims, hooks, rules, a memory system and a security scanner called AgentShield, all wired into the coding agent you already run. The author, Affaan Mustafa, built the collection in eight hours as his entry to the Cerebral Valley x Anthropic hackathon in February 2026, and won it. He had already won first place at an earlier Anthropic hackathon with a project called Zenith Chat. The repository was created in January 2026, renamed from Everything Claude Code to ECC on 1 October 2026, and shipped its current release, v2.2.3, the same day. It is MIT-licensed, and the README states plainly that the open-source project stays free forever.

What ECC actually does

The whole system compresses into one loop: plan, then test, then implement, then review, then verify, then remember, then improve. Install it once and every session follows that order instead of you re-typing it. A plan becomes an editable artifact before any code is written. Test-driven work becomes a gated red-green-refactor cycle with evidence captured before implementation, not after. The review step runs from a fresh context, so the same context that wrote the code is not the one judging it. Sessions are distilled into summaries, instincts and reusable skills rather than saved as enormous transcripts. And hooks can enforce deterministic checks — formatting, type checks, a dev-server guard — outside the prompt, where the model cannot talk its way around them.

Underpinning that is a method the project calls eval-driven development. Traditional test-driven development asks whether the code produces the right output for given inputs. Eval-driven development asks a different question first: how good is this output, and how will you know? Before writing code, you define how success will be measured across dimensions tests do not capture — readability, maintainability, security posture, performance, alignment with the architecture. The hackathon coverage credits the approach with 65% faster feature completion and 75% fewer first-pass review issues. Treat those numbers as the author's own reported results, not independent benchmarks: they come from coverage of the hackathon entry, and no third party has reproduced them.

What is inside the install

Two numbers in that table deserve a second look. The skill count has grown from 261 in the v2.0 release to 311 in v2.2.3 — fifty new skills in about two months — so this is a project still iterating fast, not a finished artefact. And AgentShield, the security scanner, is listed in the README with 102 static analysis rules, 1,282 tests and 98% coverage. Those are the project's own figures, reported by independent roundups that checked the README; they describe the scanner's test suite, not a guarantee that it catches every misconfiguration.

How to install it

The recommended path is the guided setup through npm. It needs Node.js 18 or newer, Git, and Claude Code 2.1 or newer on your PATH:

npx [email protected] setup

The wizard inventories the official marketplace and every native Claude install scope before it writes anything, then installs, updates or safely moves the ecc@ecc plugin to the scope you choose — user, project or local. Rerun the same command to update ECC, change scope or change the hook profile. A multi-harness wizard configures Claude Code, Codex and Kimi Code in one reviewed flow:

npx [email protected] install --guided

There is also a native plugin path inside Claude Code itself:

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

Pick one path per harness and stop there. The README is explicit: do not stack install methods. Installing ECC twice into the same harness can duplicate skills, commands, hooks or configuration. If you already layered installs and things look duplicated, the project points you to its reset and uninstall section rather than asking you to untangle it by hand. A minimal, no-hooks install is available for anyone who wants the rules, agents and workflows without the hook runtime:

npx [email protected] install --profile minimal --target claude

That last option matters more than it looks. Hooks are the part of ECC that runs code automatically when events fire in your session. The project requires an explicit decision — an --enable-hooks flag — before any install materialises the hook runtime. Start without hooks, read what the skills do, and add the runtime only when you know what you are handing it.

Which coding agents it works with

Read that table before assuming feature parity. ECC works best with Claude Code today — that is the project's own wording — and the adapters are honest about their limits. The Copilot target is the clearest example: it installs skills and agents only, its hook automations are written against Claude Code's hook event model and are not ported, and Copilot Chat in VS Code does not read the .github/skills/ or .github/agents/ paths at all. The instruction and prompt layer is what carries ECC's approach into Copilot Chat sessions. If your daily driver is not Claude Code, check the platform support matrix in the README for your harness before installing.

What it costs — and what it does not include

The toolkit itself costs nothing. The repo is MIT-licensed, the npm packages ecc-universal and ecc-agentshield are free, and the README's line is unambiguous: OSS stays free. What costs money is the hosted GitHub App, ECC Pro, which brings the system to private repositories:

Two pieces of fine print. First, yearly billing on Pro saves about 17% against the monthly rate, which works out to roughly $15.77 per seat per month — the site states the percentage, the dollar figure is arithmetic from it. Second, and more important: ECC does not include Claude Code. The toolkit is free; the agent it supercharges is not. Claude Code ships with Claude Pro at $20 a month, Max from $100, or pay-as-you-go API billing, and there is no free tier. Budget for both, or ECC is a very good steering wheel with no car attached.

The catches, stated plainly

First, the star count is real but it is not a quality certificate. ECC gained roughly 15,000 stars between 11 October and 28 October 2026 — 276,490 on the 11th, 284,128 on the 20th, 291,250 when this article was checked — which is about 870 a day. That is genuine momentum, and it is also the kind of momentum that attracts copycats. The README opens with a warning in bold: install ECC only from verified channels — the GitHub repository, the ecc-universal and ecc-agentshield npm packages, the GitHub App, the ecc@ecc plugin slug, and ecc.tools. Third-party re-uploads and unofficial mirrors are not maintained or reviewed by the project and may contain malware. For a tool that installs hooks into your coding agent, that warning is not boilerplate. Verify the source before you run the installer.

Second, 311 skills is a lot of surface area to hand an agent you have not audited. One independent guide puts the trap plainly: install everything, never open a guide, and assume more is better, and you have handed Claude a toolbox you have not looked inside. The sensible path is the one the project itself suggests — start with the core profile, add only the rule packs for your stack, and read the skills you actually invoke. The memory system, the hooks and AgentShield all run with the privileges of your coding session; convenience that deep deserves a look before it is switched on.

Third, this is largely a one-maintainer project. The README says so itself: a single maintainer ships weekly across seven harnesses, funded by sponsors and Pro subscribers. That is a strength — the project moves fast and one person holds the vision — and a concentration risk. If the maintainer stops, 291,000 stars do not keep the releases coming. The MIT licence means anyone could fork it, but a fork is not a roadmap.

Fourth, the quality claims are the author's. The 65% faster feature completion and 75% fewer first-pass review issues come from coverage of the hackathon win, not from an independent benchmark. The AgentShield figures — 102 rules, 1,282 tests, 98% coverage — describe the scanner's own test suite. ECC is a configuration and workflow system, not a model, so there is no benchmark table to hide behind; judge it on whether your own sessions get more predictable, not on a number in the README.

Pros and cons

Pros: free and MIT-licensed with no paid tier for the toolkit itself; a complete system — agents, skills, hooks, memory and security scanning — rather than a single trick; works across Claude Code, Codex, OpenCode, Cursor and more; the install wizard preflights and asks for confirmation before writing; AgentShield treats your agent configuration as an attack surface, which most setups never do.

Cons: Claude Code itself is a separate paid product, so the real cost starts at $20 a month; 311 skills can overwhelm a new user who installs everything blind; hooks and memory run with your session's privileges and deserve review first; feature parity drops off sharply outside Claude Code; a single maintainer is a concentration risk; the headline performance numbers are author-reported, not independently benchmarked.

FAQ

Is ECC really free?

Yes, for the toolkit. The repository is MIT-licensed and the README states the open-source project stays free forever. You pay only if you want the hosted GitHub App on private repositories — ECC Pro at $19 per seat per month — or for Claude Code itself, which starts at $20 a month for Pro. For the full picture on Claude Code's pricing, see our Claude Code explainer.

Does ECC replace Claude Code?

No. ECC is a layer on top of Claude Code: skills, agents, hooks, rules and memory that give the agent a repeatable engineering workflow. Without Claude Code (or another supported harness), there is nothing for ECC to plug into.

Will ECC work with Cursor, OpenCode or Codex?

Yes, with limits. Codex has a native plugin path. OpenCode, Cursor, Gemini CLI, Zed, Copilot and others have adapters, but feature parity varies — Copilot, for example, gets skills and agents only, with no hooks, rules or commands. Check the platform support matrix in the README for your harness before installing. Our roundup of the 7 best AI coding agents in 2026 compares the harnesses themselves on price and limits.

Is it safe to install 311 skills and hooks?

Reasonably safe if you install from official sources and review what you enable. The README warns that third-party re-uploads may contain malware, and the installer requires an explicit --enable-hooks flag before the hook runtime is installed. Start with the minimal or core profile, add only the rule packs for your stack, and read the skills you invoke.

What does ECC Pro add over the free version?

Pro is the hosted GitHub App for private repositories: private-repo analysis, PR-triggered audits, 50 analyses per active seat pooled across the installation, 1,000 commits per run, AgentShield-backed scanning and automatic checks on large pushes. The free tier covers public repositories only, at 10 analyses a month and 200 commits per run.

Key takeaways

• ECC is a free, MIT-licensed system of 71 agents, 311 skills, hooks, rules and memory that gives Claude Code a repeatable engineering workflow: plan, test, implement, review, verify, remember, improve.
• It has passed 291,000 GitHub stars with about 870 new stars a day in late October 2026, and it is the most-starred Claude Code configuration project on the platform.
• Install it with npx [email protected] setup, pick one install path per harness, and never stack install methods — the README warns that doubles skills, hooks and configuration.
• The toolkit is free, but Claude Code itself starts at $20 a month, and private-repo automation through ECC Pro costs $19 per seat per month.
• Install only from official sources, start without hooks, and read the skills you hand your agent — 311 of them is a lot of surface area to grant blind.

ECC's pitch is not that your agent gets smarter. It is that your agent gets a process it cannot forget, a reviewer that starts from a fresh context, and a memory that outlives the session. For a free, MIT-licensed toolkit, that is a serious offer — as long as you install it from the real source, budget for the Claude Code subscription underneath it, and treat the 311 skills as a library to curate rather than a pile to swallow whole. Checked and written on 28 October 2026, when the repository stood at 291,250 stars.

Official links: ECC on GitHub · ecc.tools · ECC pricing · ecc-universal on npm