Everything Claude Code: what's inside and how to install it
The repository now lives at affaan-m/ECC: 71 agents, 314 skills, 98 command shims, 22 rule packs and a hook runtime for Claude Code. Here is the inventory, the verbatim install commands, and which parts are worth switching on.
TL;DR
- “Everything Claude Code” is a community collection of Claude Code configuration — agents, skills, commands, rules, hooks and MCP configs — maintained by Affaan Mustafa. It is not an Anthropic product.
- The repository was renamed: github.com/affaan-m/everything-claude-code now redirects to affaan-m/ECC, and the plugin slug is ecc@ecc, not everything-claude-code@everything-claude-code.
- At commit d2ef2680 (2026-11-06) the tree holds 71 agents, 314 SKILL.md folders, 98 command files, 22 rule packs, 68 scripts and 8 hook events.
- Install once per harness — the guided wizard (npx ecc-universal@2.2.3 setup) or the native /plugin commands. Stacking both is the documented cause of duplicated skills and double-firing hooks.
Repository state checked 2026-11-12: 301,569 stars on GitHub, latest release v2.2.3 published 2026-10-01, latest commit d2ef2680 on 2026-11-06. Star counts move daily — read it as a snapshot. Directory counts, version numbers and install commands on this page were read from the repository tree and manifests, not paraphrased from blog posts.
What the collection actually is
It is a plugin, not a model and not a fork of Claude Code. The repository ships a `.claude-plugin/` manifest plus five kinds of plain-text assets — subagent definitions, skills, slash commands, rule files and hook scripts — that Claude Code loads from your config directory.
The project describes itself as configs from an Anthropic hackathon winner: the author won the Anthropic x Forum Ventures hackathon in September 2025 and says the collection grew out of daily use. It is MIT-licensed, third-party and community-maintained; Anthropic does not publish or review it.
Since the 2.x line the same root directory is also mapped to other harnesses — Codex, OpenCode, Cursor, Gemini CLI, Copilot CLI, Qwen, Kimi Code and others — through adapter folders such as `.codex/`, `.opencode/` and `.cursor/`. Feature parity differs per harness, which is why the README keeps a support matrix.
- Version
- v2.2.3 · 2026-10-01
- Latest commit
- d2ef2680
- License
- MIT
- Stars (snapshot)
- 301,569
What's inside the repository
The counts below come from the repository tree at commit d2ef2680 — 6,449 entries, not truncated — rather than from the README's summary table. Folder names are the identifiers Claude Code matches against, so they are left untranslated.
| Directory | Entries | What it holds |
|---|---|---|
agents/ | 71 | Subagent definitions as markdown with frontmatter: planner, architect, code-reviewer, security-reviewer, per-language reviewers (Go, Python, Rust, Java, Kotlin, TypeScript, Ruby, F#) and per-language build-error resolvers. |
skills/ | 314 | One folder per skill, each with a SKILL.md: TDD workflow, security review, verification loop, eval harness, strategic compaction, language and framework patterns, plus domain skills for ML, data and operations. |
commands/ | 98 | Slash-command files. The project is moving to a skills-first surface, so these are described as maintained compatibility shims rather than the primary entry point. |
rules/ | 22 | Always-loaded standards split into `common/` (language-agnostic) and one folder per stack — typescript, python, golang, rust, java, kotlin, swift, php, csharp, cpp, ruby, perl, react, react-native, vue, nuxt, angular, dart, fsharp, arkts, web. 122 markdown files in total. |
hooks/ | 7 | hooks.json plus the memory-persistence scripts. Eight events are wired: PreToolUse (10 matchers), UserPromptSubmit, PreCompact, SessionStart (2), PostToolUse (2), PostToolUseFailure (2), Stop (8) and SessionEnd. |
scripts/ | 68 | Cross-platform Node.js tooling: the installer, doctor, repair, uninstall, session inspection, memory, eval harness and the Codex and OpenCode adapters. |
contexts/ | 3 | dev.md, review.md and research.md — alternative system-prompt contexts you can inject for a session. |
mcp-configs/ | 1 | One `mcp-servers.json` with MCP server definitions to copy into `~/.claude.json` after replacing the placeholder keys. |
legacy-command-shims/ | 12 | Retired short-name commands kept for opt-in, such as /tdd, /e2e, /eval, /verify and /orchestrate. Each one points at the skill that replaced it. |
Rules and skills are opt-in by design: a plugin install cannot ship rules, because Claude Code plugins have no rules channel. That is why the README asks you to copy the rule folders by hand.
How to install it in Claude Code
Three routes, all copied verbatim from the README. Pick one per harness — the README's own warning is that stacking two install methods duplicates skills, commands and hooks.
Route 1 — the guided wizard (recommended)
Installs, updates or moves the `ecc@ecc` plugin into the scope you choose (user, project or local) and sets a hook profile. It needs Node.js 18 or newer and does not require cloning the repository. The pnpm, Yarn 2+ and Bun equivalents are `pnpm dlx` / `yarn dlx` / `bunx` with the same arguments.
npx ecc-universal@2.2.3 setupRoute 2 — the native plugin commands
Run these inside a Claude Code session (2.1 or newer). This installs skills, agents, commands and the plugin-managed hooks, and stops there — do not run a full manual install afterwards.
/plugin marketplace add https://github.com/affaan-m/ECC/plugin install ecc@eccRoute 3 — the rules, which no plugin can install
Claude Code plugins cannot distribute rules, so the README asks you to clone the repository and copy the packs you want. Start with `rules/common` plus one language pack; the second copy line in the block is the one to swap for your stack.
git clone https://github.com/affaan-m/ECC.gitcd ECCmkdir -p ~/.claude/rules/ecccp -R rules/common ~/.claude/rules/ecc/cp -R rules/typescript ~/.claude/rules/ecc/ # replace with your stackCheck that it worked
Inside Claude Code, `/plugin list ecc@ecc` should show the plugin as enabled, and plugin commands appear namespaced — `/ecc:plan`, not `/plan`. If you layered installs and see duplicates, the README's fix is to uninstall and install once.
What you need first
- 1Claude Code 2.1 or newer on PATH. The README names v2.1.0 as the minimum because of how the plugin system loads hooks.
- 2Node.js 18 or newer for the guided wizard and for the hook scripts, which were rewritten in Node.js so they run on Windows, macOS and Linux.
- 3Git, plus a reviewed checkout if you plan to copy rules or read the hook scripts before enabling them.
Which parts are worth using
The repository is a catalogue, not a product, and the author says so: “start with what resonates, modify for your stack, remove what you don't use.” On that basis these four parts carry most of the value.
- 1
The subagent definitions
A review or planning role is cheap to evaluate: open agents/code-reviewer.md or agents/planner.md, read the prompt, and decide whether you want that voice in your sessions. Copying one file into ~/.claude/agents/ costs nothing and needs no plugin.
- 2
The TDD and verification skills
tdd-workflow, verification-loop and eval-harness are the workflow pieces the guides are built around: they encode write-the-failing-test-first and check-before-claiming-done as steps rather than as advice. Read them before installing — they change how sessions run.
- 3
The rules packs
rules/common is the most portable artefact in the repository: short, always-loaded standards for testing, security, git workflow and coding style. Because they are plain markdown, they work with any agent that reads a rules or memory file, and they are the easiest thing to trim down to your own house style.
- 4
The hook runtime — selectively
The hooks are where the collection actually enforces something: a Stop hook that runs format and type checks, a PreToolUse gate for destructive shell commands, session start and end memory scripts. They are also the part with the most surface area, so enable them one by one and read the script first. ECC_HOOK_PROFILE (minimal / standard / strict) and ECC_DISABLED_HOOKS exist for exactly this.
Caveats before you install
These are trade-offs the README and the repository structure make visible — not hypotheticals.
- It is third-party and unofficial. MIT-licensed community work, not an Anthropic product. The README itself warns that only the GitHub repo, the ecc-universal and ecc-agentshield npm packages, the ecc@ecc slug and ecc.tools are maintained channels, and that mirrors may contain malware.
- Context cost is real. Skills load on demand, but always-loaded rules and SessionStart context are charged to every turn: the README caps SessionStart additional context at 8,000 characters by default and exposes ECC_SESSION_START_MAX_CHARS and ECC_SESSION_START_CONTEXT=off to lower it. The same README warns that too many MCP tools can shrink a 200k window to about 70k.
- It can collide with the config you already have. Rules go into ~/.claude/rules/ecc/, agents into ~/.claude/agents/, hooks into the plugin's own hooks.json — if you already keep a CLAUDE.md convention or your own hooks, expect to merge rather than adopt.
- The identifiers changed. Older posts install everything-claude-code@everything-claude-code from affaan-m/everything-claude-code; both now resolve to affaan-m/ECC and ecc@ecc. The README calls the long form a legacy alias.
- The hooks are executable configuration. They run shell commands on your machine on tool events. Read hooks/hooks.json and the scripts under scripts/hooks/ before enabling them, and do not copy hooks.json into settings.json on top of a plugin install — the README documents that as the cause of duplicate-hook errors.
If you only take one thing
Take the rules and one or two skills. The value of this collection is not the 314-skill catalogue — nobody reviews 314 skills — it is the handful of artefacts that encode a workflow you already believe in, plus the reminder that always-loaded context has a price. Everything else can wait until a session actually goes wrong.