ECC by affaan-m

What it is

ECC is Affaan Mustafa’s set of agents, skills, commands, hooks and rules for AI coding agents. It gives an agent an engineering routine: it plans before it builds, proves changes with tests and reviews its own work from a fresh context. Hooks keep what matters between sessions, and repeated wins turn into reusable skills.

1. Plan the change
2. Write the test first
3. Implement
4. Review from a fresh context
5. Verify
6. Remember what worked and improve

Who it suits

Developers who run coding agents on long projects and want a fixed routine with memory between sessions. Teams that want every agent change tested and reviewed before merge get the most from it.

If you want a light setup, the full catalog and its hooks can be more than you need, and the README offers a minimal path with no hooks.

Where to use it

Support differs by agent, and the README says how far each one goes.

• Operating systems: the core CLI runs on Windows, macOS and Linux
• Optional features: some need Bash or Python
• Other agents: Zed, Qwen, Hermes, OpenClaw, Kimi Code, CodeBuddy and JoyCode install from a clone with the README’s script

How to set it up

Get it saves a zip of Mindket’s setup notes for each agent. ECC itself installs from affaan-m’s repository and the ecc-universal npm package.

The README’s Windows walkthrough is the clearest first install:

1. Open PowerShell and check that Node, Git and Claude Code each answer.
2. Run the guided setup.
3. For a personal setup, choose Global user and Standard hooks, then confirm.
4. Start a new Claude Code session and run /plugin list to check that ecc@ecc is on.

node --version
git --version
claude --version

npx ecc-universal@2.2.3 setup

Rerun the same setup line to update ECC, change its scope or change its hook profile. Codex takes ECC as a native plugin from the same repository:

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

The low-context route installs ECC into Claude Code with no hooks. It is a route of its own, taken instead of the setup above.

npx ecc-universal@2.2.3 install --profile minimal --target claude

Pick one route per agent. Installing ECC twice into the same agent can duplicate skills, commands, hooks or configuration.

How to use it

A feature starts with /ecc:plan and a short description, and the rest follows from there.

Manual installs may expose shorter forms, such as /plan.

Example

In Claude Code you plan a feature before any code changes.

/ecc:plan "Add user authentication with OAuth"

The planner agent drafts an implementation blueprint for you to confirm or edit before coding starts. Then the tdd-workflow skill has the tdd-guide agent write a failing test before the change, and /code-review hands the result to the code-reviewer agent.

For newcomers and experts

If you are new. ECC wraps your coding agent in a routine, so a change starts with a plan you approve and a test that fails before any code is written. You drive it with commands such as /ecc:plan. Install the plugin in Claude Code and plan one small feature with it first.

For experts. Much of its value lives in hooks that run local scripts at set points of a session, so the routine holds even when the model forgets. ECC_HOOK_PROFILE sets how strict they are, standard by default, and ECC_DISABLED_HOOKS turns single hooks off:

export ECC_HOOK_PROFILE=standard
export ECC_DISABLED_HOOKS="pre:bash:tmux-reminder,post:edit:typecheck"

Skills are the main surface now, and the commands stay as entry points while ECC moves to skills first.

The plugin shows that whole catalog to the model, so pick a selective install when context matters. The cost is overhead on every turn and more moving parts to check when an agent acts oddly.

Good to know

The README asks you to treat hooks and MCP servers as executable configuration.

• Hooks: plugin hooks run local scripts on each prompt and tool call, at session start and end, and before compaction
• GateGuard: checks destructive shell commands, such as rm, before they run
• Stored data: session summaries and learned skills stay under ~/.claude by default
• Rules: plugins cannot ship rules, so you copy the rule packs you want by hand
• OpenAI: /astra-review sends your diff to OpenAI through your Codex CLI login, and only when you run it
• Diagnostics: ECC uploads none on its own
• Removal: ECC tracks its install state, so its uninstall takes out only the files it owns
• Official sources: install only from them, since the maker warns that unofficial copies may carry malware
• License: MIT