# Your coding agent can already write code. This is the system around it.
The repo's own pitch line says it plainly: "Your agent can write code, but ECC gives it a coordinated engineering system and toolbox." That's the honest framing: a process wrapped around the coding agent you already use, no model swap required.
Everything Claude Code (ECC) is a free, MIT-licensed plugin at github.com/affaan-m/ECC. 260,928 stars, 39,042 forks, created eight months ago, still being pushed the same day this was checked.
The loop
Say "plan," and a planner agent writes the blueprint before a line of code exists. Then it writes tests first and keeps building until they pass. A second agent reviews the work from a fresh context. /code-review dispatches a real separate subagent (agents/code-reviewer.md, its own Read/Grep/Glob/Bash tools) — a different Claude Code conversation reading the diff cold, not the one that wrote it grading itself. And it remembers: hooks watch your sessions, extract atomic "instincts" with confidence scores, and the next session starts with the ones that matter.
Getting it running
Two commands inside Claude Code, no account, no key:
/plugin marketplace add https://github.com/affaan-m/ECC
/plugin install ecc@ecc
The native install brings skills, agents, commands and plugin-managed hooks. Rules don't ship through the plugin path, so add them separately, and start narrow:
git clone https://github.com/affaan-m/ECC.git
cd ECC
mkdir -p ~/.claude/rules/ecc
cp -R rules/common ~/.claude/rules/ecc/
cp -R rules/typescript ~/.claude/rules/ecc/ # swap for your own stack
The README says it outright: start with rules/common plus one language pack you actually use, not all 23.
Watching it remember
/instinct-status shows what it's learned so far, with confidence scores attached. /evolve clusters related instincts into a reusable skill or command once you've got enough of them. Nothing gets injected below a 0.7 confidence floor, capped at six per session. Don't expect a payoff on session one; this pays off after real use, not after install.
What to know before you lean on it
The rules mention an 80% test-coverage target. That's advisory prose in rules/common/testing.md — no hook in the repo actually enforces it, so treat it as a target you hold yourself to, not a gate the tool holds you to.
Codex is fully supported. Cursor has an adapter, but the README lists it as beta, not full parity — check its own lifecycle notes before you depend on it there.
You may see this credited with winning Anthropic's own hackathon. That claim belongs to one sub-tool, AgentShield, built at the Claude Code Hackathon (Cerebral Valley x Anthropic, Feb 2026) — not the whole repo, and not this install.
AgentShield itself — the security scanner for prompts, hooks, and permissions — is a separate, unaudited-by-ECC install (ecc-agentshield on npm). Still free, just not bundled.
The paid tier
The plugin and the repo stay MIT, free forever — "OSS stays free," in the README's own words. ECC Pro, $19 per seat per month, is a separate hosted GitHub App that runs PR-triggered audits on private repos with pooled team usage. You don't need it for anything above.
The checker
harness_check.sh in the kit reads your own ~/.claude/plugins state and tells you plainly whether ECC is actually installed, rather than asking you to take a video's word for it. Reads only — installs nothing, sends nothing.
Comment HARNESS and I'll send you the setup.