Built With Claude Code: How ECC Hit 253K GitHub Stars

Most "built with Claude Code" stories end at a launch tweet and a few hundred stars. ECC — short for a hackathon project originally named "Everything Claude Code" — kept going past that point.

As of this week it sits at #6 on GitHub's overall daily trending page, picked up 1,897 stars in a single day, and its repo now shows roughly 253,000 stars and 37,900 forks total, per GitHub's own trending page and the repo itself. The account behind almost all of it is one person.

What ECC actually is

ECC describes itself, in its GitHub repo description, as "the agent harness performance optimization system" — a layer that sits on top of AI coding agents like Claude Code, Codex, Cursor, and OpenCode and gives them, per its own README, "a coordinated engineering system and toolbox: it plans before it builds, verifies changes with tests, reviews its own work from a fresh context, remembers what matters, and turns repeated wins into reusable skills and workflows." Concretely, that's packaged as "68 agents, 286 skills, and 94 legacy command shims, plus hooks, rules, memory, continuous learning, and AgentShield security scanning" — installed as a plugin or npm package on top of whichever coding agent a developer already uses.

It's not a rewrite of Claude Code. It's closer to an opinionated operating layer that tells an existing agent how to work — plan first, write tests, review its own diff before calling a task done — rather than what to build. That's the same instinct behind measuring how much of a follow-up task an agent actually finishes on its own once the first workflow ships.

The origin: built with Claude Code at a weekend hackathon

The part worth taking seriously is where this started. ECC's own README credits a specific event for its security-scanning module: "Built at the Claude Code Hackathon (Cerebral Valley x Anthropic, Feb 2026). 1282 tests, 98% coverage, 102 static analysis rules." The same hackathon credit shows up independently in the standalone AgentShield repo — a separate, MIT-licensed project by the same account that ECC's security scanning is built on. Two repos, one author, the same origin story told twice.

The repo's original name backs this up structurally, not just narratively: its old URL, affaan-m/everything-claude-code, still resolves — it now lands on the current affaan-m/ECC repo, the standard GitHub behavior when a repo gets renamed. A tool that started as "everything Claude Code" outgrew its own name once it started supporting Codex, Cursor, Gemini, and other agent harnesses, and the acronym stuck.

The numbers, and where they come from

Everything quantitative here is publicly countable on GitHub itself. That's a higher bar than our look at Aider could clear for its self-reported install and token metrics — these ECC numbers are platform-displayed counts, checked directly on the repo page, not a badge the maintainer wrote themselves:

• ~253,000 stars, ~37,900 forks — displayed on the repo page.
• 2,667 commits on the main branch.
• 43 open issues, 124 open pull requests — a real, active backlog, not a dormant repo coasting on old stars.
• #6 on GitHub's daily trending list, with 1,897 stars gained in one day — direct from github.com/trending.
• The npm package ecc-universal is live: the public npm registry shows the latest published version, 2.2.0, went out August 27, 2026 — one version behind the 2.2.1 already declared in the repo's own package.json on the main branch.

That gap is worth naming rather than glossing over: it means the repo's version file is running ahead of what's actually published to npm, a normal and minor lag rather than evidence of anything wrong — but it's a reason to check the registry directly instead of taking a package.json number at face value. That's the same kind of gap that showed up between GitHub's tagged releases and PyPI in the Aider piece — and a different discipline than the browser and OS products racing to build for AI agents instead of humans, which mostly report their own traction rather than a platform's.

One name, MIT license, a Pro tier

The LICENSE file is short and specific: MIT License, with the line "Copyright (c) 2026 Affaan Mustafa." That's the same name on the package.json author field and the only name attached to every release. His GitHub profile lists a bio of "💫 everything is tokens," 10.3k followers, and affiliations with two small GitHub organizations, Ito-Markets and ECC-Tools — the latter of which currently shows just one public repository of its own, meaning the actual ECC codebase lives under his personal account, not a company org.

The business model sits directly in the README: "OSS stays free. This repo is MIT-licensed forever." — and next to it, the paid layer: "ECC Pro is the hosted GitHub App for private repos," with "[p]rivate repos from $19/seat/mo" linked out to the project's own pricing page. That pricing line comes from the project's own README, not a third-party writeup. We weren't able to independently load the marketing site itself in this pass, so treat the $19/seat figure as accurate to what the maintainer has published — not independently re-verified against the live checkout page. No revenue numbers are published anywhere in the repo, so none are claimed here.

The part that's easy to romanticize — and the part that isn't

It's tempting to read "253K stars, one GitHub account" as a pure solo-hacker fairy tale. Two things complicate that.

First, 124 open pull requests is a lot of outside contribution to sort through for one maintainer — the same bottleneck Aider runs into with its own backlog. A single name on the license doesn't mean a single person is reading and merging everything that shows up.

Second, the project's own CHANGELOG shows the kind of engineering discipline that doesn't usually survive a hackathon demo. The latest entry, v2.2.0, documents a deliberate cut: the default set of MCP connectors was reduced from six down to one (chrome-devtools) "after the June 2026 audit," with the rationale spelled out in a separate MCP-CONNECTOR-POLICY.md doc — a policy that requires any new default to be "Universal" and that "MCP beats a CLI/API wrapped in a skill," explicitly rejecting "Popular" as a justification on its own. That's a written governance process — the kind most startups don't bother drafting until years in — on a project that started as a hackathon entry earlier the same year.

What this generalizes to

The transferable lesson isn't "win a hackathon and hope." It's that ECC's growth is legible in a way most viral GitHub projects aren't: every number in this piece — stars, forks, commits, the npm version, the license, the hackathon credit — sits in a file or a platform page you can open yourself right now. The project didn't just get lucky with a demo; per its own changelog, it kept writing down why it says no to things (six MCP connectors cut to one, with a documented rule for what earns a spot back) even while the star count kept climbing. For a solo builder shipping fast with Claude Code, Cursor, or anything similar, that's the harder habit to keep than the initial hackathon build itself: writing the rule for what you won't add, at the exact moment your project has the least reason to say no to anything.