Make Claude Code actually reliable

ECC is a free, MIT-licensed set of rules, skills, agents, memory and hooks for Claude Code, Codex and Cursor. It's config files in folders that your assistant reads — no model change, no magic.

1. The problems it fixes

If you use an AI coding assistant daily, this list will feel familiar:

• It forgets everything between sessions. Every morning you re-explain your architecture.
• You retype the same prompts — "write tests," "follow our conventions" — forever.
• Inconsistent style. Every file looks like a different developer wrote it, because effectively one did.
• Weak verification. Code that looks right and was never run.
• No project standards the assistant reliably respects.
• Security as an afterthought — nobody audits the config files the agent itself reads.

2. The core idea

Most people chasing better output reach for a bigger prompt. A longer CLAUDE.md. More instructions crammed into context. It works for about a session and a half.

The model doesn't change. Everything around it does — and that's the whole pitch. Here's roughly what handles what:

3. Installing it (real commands)

ECC ships as a Claude Code plugin. Two commands, run inside Claude Code.

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

/plugin install ecc@ecc

Requirement: Claude Code CLI v2.1.0 or later — earlier versions handle plugin hooks differently. Check with claude --version.

Want a smaller footprint?

There's a script install with profiles. minimal skips hooks entirely.

./install.sh --profile minimal --target claude

.\install.ps1 --profile minimal --target claude

Not sure which pieces you need? Describe the job in plain English and it recommends components:

npx ecc consult "security reviews" --target claude

Three identifiers that aren't interchangeable

4. Two things that break setups

Both are silent. Neither throws an obvious error. Fix them at install time and you avoid the two failure modes that make people give up on this.

git clone https://github.com/affaan-m/ECC.git

mkdir -p ~/.claude/rules/ecc
cp -R rules/common ~/.claude/rules/ecc/
cp -R rules/typescript ~/.claude/rules/ecc/

One more if you install components manually: Claude Code loads skills only from direct children of ~/.claude/skills. Nest them under ~/.claude/skills/ecc/ and they silently never load.

If something's already broken

node scripts/ecc.js doctor
node scripts/ecc.js repair

node scripts/uninstall.js --dry-run

Worth knowing before you commit: uninstall is a documented, dry-runnable path. This isn't a one-way door.

5. What's actually inside

6. The security scan people skip

Your assistant reads config files — CLAUDE.md, settings.json, MCP configs, hook definitions, agent definitions, skills. Every one is an input, and inputs can be poisoned.

AgentShield scans exactly that surface:

npx ecc-agentshield scan

npx ecc-agentshield scan --fix

Five categories: secrets detection, permission auditing, hook injection analysis, MCP server risk profiling, agent config review. Reports come as terminal, JSON, Markdown or HTML, and it exits with code 2 on critical findings so you can gate a CI build on it.

7. The three official guides

The repo is code; the reasoning lives in three companion guides. Bookmark these — they're worth more than any summary:

• The Shorthand Guide — setup, foundations, philosophy. Start here.
• The Longform Guide — token optimization, memory persistence, evals, parallelization.
• The Security Guide — attack vectors, sandboxing, sanitization, AgentShield.

8. Is it worth it?

Worth it if you use an AI coding assistant most days, work in a real codebase with standards worth enforcing, and you're tired of re-explaining your project every session. Memory and rules alone carry it.

Skip it if you use AI coding occasionally or for one-off scripts. There's a lot of it, and you'll spend longer configuring than you save.

Three things to know before you commit

• Context cost is real. Rules and memory load into your context every session. That's the tradeoff for persistence — the Longform Guide covers token optimization, and it's worth reading before you install everything.
• It installs to ~/.claude, not your repo. This is your machine's setup, not something your team inherits by cloning. Teammates each install their own.
• It's a fast-moving, single-maintainer project. MIT-licensed and free, which is great for adoption — just know that before a team workflow depends on it.

The underlying idea holds even if you never install it: most bad AI coding output is a context problem, not a model problem. Fix what surrounds the model and the same model does noticeably better work.

Repo: github.com/affaan-m/ECC — free, MIT licensed, by @affaan-m. Not affiliated — I just think it's the right way to think about this.