What is in it
ECC describes itself as “the agent harness performance optimization system”. The idea is that you install a way of working once, instead of repeating it in every prompt. The README sums up the loop as plan, test, implement, review, verify, remember, improve. Per the README in November 2026:
| Part | Count | What it is for |
|---|---|---|
| Agents | 72 | Specialized subagents for planning, review, build repair, security and architecture |
| Skills | 327 | TDD, research, security, docs, frontend, data, ML, operations and more |
| Commands | 98 | Older slash-command entry points, while ECC moves to skills first |
| Hooks and memory | Runtime | Enforcement, session summaries, “continuous learning” and “instincts” |
| Rules | Selective | Standards you choose per language or project, added by hand |
| AgentShield | Included | Scans prompts, hooks, MCP config, permissions, secrets and agent files |
The repository has over 300,000 stars on GitHub, which makes it one of the most visible Claude Code add-ons. It works best with Claude Code, has native plugin paths for Codex and Qoder, and “capability-limited adapters” for Cursor, OpenCode, Gemini, Zed, GitHub Copilot and others.
Installing it sensibly
- Use the guided setup:
npx [email protected] setup. It checks existing installs before it changes anything. - Choose the scope on purpose.
useraffects every project on your machine;projectwrites into the repo, so your colleagues get it too;localis for you in one project. Start withlocaloruserwhile you evaluate. - Choose a hook profile. Hooks run on every tool call or session event. The README suggests Standard for a typical personal setup.
- Add rules selectively. Only the language and project packs you actually need.
- Install only from official channels. The README warns explicitly about re-uploads and mirrors.
The common misunderstanding: “more skills means a better agent”
327 skills and 72 agents sound like 327 improvements. In practice, every skill description, rule and hook output competes for the same context window, and every hook adds a step to every action. Claude Code loads skills on demand, which helps, but a team that installs everything at once will struggle to tell which part changed the agent’s behavior. ECC itself offers a minimal profile, and the README tagline is “Optimize the context window. Persist everything else.” Start small, add what solves a problem you have, and remove what you don’t use. Background on the building blocks: Claude Code skills, hooks and plugins and memory.
Claude Code’s own documentation puts a mechanism behind that cost. Skill descriptions are loaded into context so Claude knows what is available, and the docs say every skill in the listing adds to your context on every turn, whether or not Claude ever uses it. The combined description and when_to_use text is cut at 1,536 characters per skill in that listing, and the skill body loads only when the skill is used. Setting disable-model-invocation: true keeps a skill’s description out of context, so it runs only when you invoke it.
A second point: ECC is code that runs on your machine with your permissions. Hooks execute shell commands. Review what you install, as you would any dependency, and read is Claude Code safe for company code.
What this means for a team
ECC gives one developer’s agent a strong default process, and with project scope the whole repo can share it. Its memory features are local: the Memory Vault keeps project memories in .ecc/memory/ and personal ones in ~/.ecc/memory/, and the README itself calls memory “unreviewed context, not executable policy”. That is honest, and it means team knowledge still has to be reviewed and kept somewhere everyone, and every tool, can reach. See shared memory across coding agents.
Where Bombos fits
ECC standardizes how one agent works; Bombos gives every agent on the team the same memory of what the team knows and has agreed. It is a plugin underneath Claude Code, Codex, Cursor, Copilot, Grok and your own agents that learns how your team works, draws on repos, documents and mail, and shows with every answer what it rests on. It runs in the cloud in Europe or standalone on your own server, and one developer can try it first with their own agents.
Related questions
Is Everything Claude Code an official Anthropic project?
No. It is maintained by an independent developer on GitHub, under the MIT license. The repository moved from affaan-m/everything-claude-code to affaan-m/ECC.
How do I install ECC in Claude Code?
The README recommends the guided setup: npx [email protected] setup (Node.js 18 or newer, Git and Claude Code 2.1 or newer). It installs the plugin ecc@ecc in the user, project or local scope you choose. Do not also do a manual install on top.
Is ECC free?
The repository is MIT-licensed and stays free, according to the README. ECC Pro is a separate hosted GitHub App for private repositories, from $19 per seat per month.
Does ECC install rules automatically?
No. Claude Code plugins cannot distribute rules, so you add only the rule packs you want by hand.
Where should I download it from?
Only from the official channels the README lists: github.com/affaan-m/ECC, the npm packages ecc-universal and ecc-agentshield, the ECC Tools GitHub App, the plugin slug ecc@ecc and ecc.tools. The project warns that unofficial mirrors may contain malware.
