Search results show two names, Everything Claude Code and ECC, but there is only one project. The old address, github.com/affaan-m/everything-claude-code, still works: it answers with a 301 redirect to affaan-m/ECC. The README tells you when the rename happened. The v1.9.0 release of 21 March 2026 still carries the old name, while v2.1.0, published on 27 July, says ECC. Affaan Mustafa (affaan-m on GitHub) put a ready-made setup for Claude Code in the repo: subagents, skills, slash commands, hooks and rules.
Whether to take the whole bundle is a separate question. The short answer: install it in parts, read the hooks beforehand, and measure what it adds to context. The reasons and the commands follow.
What the repository contains
The figures below come from the README and the repo files, as of 10 November 2026.
| Component | Count | Location |
|---|---|---|
| Subagents | 71 | agents/ |
| Skills | 314 | skills/ |
| Commands | 98 (the README calls them legacy command shims) | commands/ |
| Hooks | 27 entries across 8 events | hooks/hooks.json |
| Rules | a common pack plus packs per language | rules/ |
| MCP | one default connector, chrome-devtools, plus an opt-in catalog | .mcp.json, mcp-configs/ |
The project ships under MIT and has around 300 thousand stars. The latest release, 2.2.3, is dated 1 October, though commits were still landing on 9 November. Claude Code is the focus, but the README also describes setup for Codex, Cursor and OpenCode. Next to it the authors sell ECC Pro, a paid GitHub App for private repos (the README puts the price at $19 per seat per month and up). The plugin install steps never mention it.
Stars measure attention, not review. The README itself calls ECC a one-maintainer project. The history says otherwise: the last 100 commits to skills/ as of 10 November belong to 13 people and the owner is not one of them, while hooks/hooks.json has been edited by 30 authors over its life. That is one more reason to read the hooks yourself. The README adds its own warning: take ECC only from its official channels, because re-uploads by third parties may carry malware.
What a Claude Code plugin is made of
Anthropic's documentation describes a plugin as a directory with a manifest at .claude-plugin/plugin.json and a set of components. Claude Code installs and enables them together.
| Component | Purpose | File in the plugin |
|---|---|---|
| Skills | instructions Claude pulls in when the description matches | skills/<name>/SKILL.md |
| Agents | subagents Claude can hand part of the work to | agents/<name>.md |
| Hooks | commands fired on lifecycle events: before or after a tool call, at session start | hooks/hooks.json |
| MCP servers | tool servers that connect while the plugin is on | .mcp.json |
| Mod | hooks written as JavaScript functions that can draw panes and add commands | a hooks module |
If you searched for "claude code mods", the last row answers it. A mod is simply a plugin with a hooks module, and no separate product hides behind the word.
Plugin catalogs are called marketplaces. Technically a marketplace is a repo with a .claude-plugin/marketplace.json file. The official claude-plugins-official connects by itself on first launch, and every other one you add with /plugin marketplace add. The installer asks for a scope: you alone across all projects, everyone on the repository, or you alone in one repository.
Plugins are optional. Drop a skill into ~/.claude/skills/ and it works in every project (see the skills guide). A plugin pays off when you want a whole set with updates.
Installing ECC
You need Node.js 18 or later, Git and Claude Code 2.1 or later. There are two routes. Pick one.
Route one, native, from inside Claude Code:
/plugin marketplace add https://github.com/affaan-m/ECC
/plugin install ecc@ecc
Inside a session, /plugin install does not install the plugin right away. First a panel opens with the plugin card and a scope choice. You can confirm the install in two places: the command list gains /ecc:... skills, and claude plugin list in a terminal prints the version and scope.
Route two, through the wizard:
npx ecc-universal@2.2.3 setup
The wizard first looks at what is already installed, then installs ecc@ecc into the scope you pick and asks for a hook profile. The version is written into the command. The README adds at once that a pinned version is no substitute for reading the source.
The README gives two prohibitions. First, do not stack install methods. A plugin plus a manual install.sh into one Claude Code setup means duplicated skills and hooks. Second, do not paste hooks/hooks.json into your own settings.json. Claude Code loads plugin hooks on its own, and a second copy makes every hook fire twice.
Rules do not travel with the plugin, and the README says so bluntly: Claude Code plugins have no mechanism for distributing rules. Copy them by hand into ~/.claude/rules/ecc/, starting with rules/common and adding one pack for your language. Rules sit in context permanently, so extras cost the most here. To compare them with a project instruction file, see the piece on AGENTS.md.
The plugin comes off with claude plugin uninstall ecc@ecc. Rules you copied yourself stay.
Hooks: what to read before installing
The security part of Anthropic's documentation opens with a warning: a plugin you install can run any code on your computer as your user. Claude Code's permissions and sandbox police the tool calls the model initiates. What a plugin starts on its own is outside their reach: hooks, monitors and MCP server processes bypass the sandbox. Anthropic adds that it does not control what plugins contain and cannot confirm that they work as promised or stay unchanged.
The documentation's review order:
- Which catalog each plugin came from is shown by
claude plugin marketplace list. - In the
/pluginpanel, the Will install block says hooks exist but not what they run. - Then open
hooks/hooks.json,.mcp.jsonand thebin/directory (if there is one) right in the repo. ECC has nobin/: its hooks call scripts fromscripts/hooks/, and.mcp.jsonlists a single server, chrome-devtools. - Finally, clone the repo and run
claude --plugin-dir <plugin directory> plugin details ecc. Without starting a session, the command prints skills, agents, hooks with their events and MCP servers.
ECC has 27 hooks spread over 8 events. PreToolUse holds 10 and Stop holds 8, while PostToolUseFailure, PostToolUse and SessionStart get two each. UserPromptSubmit, PreCompact and SessionEnd have one apiece. They all start Node scripts from scripts/hooks. The README in the hooks directory gives examples: a block on npm run dev outside tmux; a check before git commit that scans staged files for secrets, debugger and console.log and stops the commit on critical findings; Prettier and tsc --noEmit after edits. Some hooks also write session state to disk.
GateGuard raises the most questions. Its skill description says it refuses the first edit to a file and the first shell command until the model lays out facts: who imports this module, what schema the data has. Only then is a retry allowed. The CHANGELOG clarifies that read-only commands (ls, cat, rg, git status) do not spend the shell gate, so it stops the first command that is not on that list. The standard and strict profiles turn it on, and minimal does not. In the Unreleased section of the CHANGELOG (edition of 10 November) the authors write that their pilot over 72 sessions put the gate's price at about 45% more turns and did not assess the benefit. Turns cost tokens. No bad intent shows. Even so, a blocked everyday command is maddening until you know the reason.
What you can configure:
hooks_enabledinplugin.jsonis on by default, and next to it sitshook_profilewith the valuesminimal,standardandstrict(defaultstandard).- Strictness is steered by
ECC_HOOK_PROFILE, whileECC_DISABLED_HOOKStakes comma-separated hook IDs and switches them off. - A manual
install.shstops and waits until you pass--enable-hooksor--no-hooks. - The plugin does not activate its bundled MCP servers on its own. If the term is new to you, the MCP servers guide is the place to start.
One command deserves singling out: /astra-review. The README says it launches your local Codex CLI and sends your diff to OpenAI, and it treats running the command as your consent. Code that must not leave your network should not go through it.
You control updates. By default Claude Code does not auto-update plugins from third-party catalogs, so files you have read cannot be swapped quietly. After you update yourself, re-read the hooks diff.
Context: what ECC costs
This part surprises people. By Anthropic's documentation, while a plugin is on, the names and descriptions of its skills, agents and commands sit in the context of every request. Those tokens are billed and shrink the context window even if no component of the plugin ever ran. A skill loads in full only when it is called.
The calculation comes from the repo files (commit of 6 November): name plus description add up to 102,234 characters for 314 skills, 15,915 for 71 agents and 15,397 for 98 commands. That makes 133,546 characters. At four characters per token, it is around 33 thousand tokens. Treat the figure as approximate and more likely a ceiling: by default Claude Code truncates a skill's combined description and when_to_use at 1,536 characters (the limit moves with skillListingMaxDescChars), and a skill with disable-model-invocation: true stays out of the list. The real contribution shows in claude plugin details ecc (the Always-on line), in /context and in /skill-doctor. ECC also has a command of its own for measuring, /context-budget.
Anyone paying per token sees this background load on the bill every day. For ways to cut it, see the piece on trimming a Claude Code bill.
Where to start: picking plugins
Start from the task, since the catalog is secondary. Typical needs are covered like this.
| You need | Take | From |
|---|---|---|
| Commits, push, PRs | commit-commands | official catalog |
| Pull request review | code-review | official catalog |
| Feature work step by step | feature-dev | official catalog |
| Types and code navigation | typescript-lsp, pyright-lsp, gopls-lsp | official catalog |
| Your own hooks without someone else's code | hookify | official catalog |
| A process: brainstorm, plans, subagents with review | superpowers | official catalog |
| Everything at once, 314 skills | ecc | third-party catalog ecc |
The official catalog is addressed the same way every time: /plugin install commit-commands@claude-plugins-official. The list was checked against marketplace.json in the anthropics/claude-plugins-official repo on 10 November. It holds 317 plugins and ecc is not among them. The README of superpowers shows the same command form: /plugin install superpowers@claude-plugins-official.
If you do need ECC, take it in parts. The README has a lighter profile with no hooks:
npx ecc-universal@2.2.3 install --profile minimal --target claude
The profile installs rules, agents, commands and core workflows and leaves hooks-runtime out. It is a standalone install route, and nobody combines it with the plugin.
Connecting a codexkey key
Plugins run inside Claude Code on your computer. The gateway sees ordinary model requests. The pairing of ECC with the codexkey gateway was not tested separately. Claude Code itself is set up for a codexkey key following the setup guide, and for that you need an address, a key and model aliases.
export ANTHROPIC_BASE_URL="https://api.codexkey.online"
export ANTHROPIC_AUTH_TOKEN="sk_cdx_yourkeyhere"
export ANTHROPIC_DEFAULT_OPUS_MODEL="claude-opus-5-5[1m]"
export ANTHROPIC_DEFAULT_SONNET_MODEL="claude-sonnet-5[1m]"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="claude-sonnet-5"
Aliases are mandatory. ECC's savings tips name the models sonnet and haiku, and the gateway resolves them only when these variables are set. Otherwise Claude Code asks for its own default models and returns a model-selection error. One more detail from the ECC README: tool names in the plugin are kept short on purpose (ecc@ecc) because of strict checks on third-party gateways. Top-ups start at $1. You can pay by Russian card, SBP (the Russian Faster Payments System) or crypto. The cost of a plugin set is easiest to measure on pay-as-you-go billing, because the dashboard shows spend per request.
FAQ
What is Everything Claude Code?
It is the affaan-m/ECC repo on GitHub: agents, skills, commands, hooks and rules for Claude Code. It carried the name Everything Claude Code until summer 2026 and was then renamed ECC. The license is MIT.
How do I install Everything Claude Code?
In Claude Code type /plugin marketplace add https://github.com/affaan-m/ECC, then /plugin install ecc@ecc. The other option is npx ecc-universal@2.2.3 setup. Stick to one of the two and do not combine them.
Which Claude Code plugins should I install?
From the official catalog take commit-commands, code-review and a language-server plugin for your language. superpowers adds a development process. Install ECC whole once you are ready to go through its hooks.
Are Claude Code plugins safe?
A plugin runs code as you, and Anthropic does not vet third-party plugins. Open hooks/hooks.json, .mcp.json and bin/, switch off hooks you do not need and keep auto-update off for third-party catalogs.
What is the difference between a plugin and a skill?
A skill is a folder with SKILL.md. A plugin packs skills, agents, hooks and MCP servers together with a manifest and an update mechanism. A lone skill moves by copying its folder.
What are mods in Claude Code?
A mod is a plugin that includes a hooks module. Its hooks are written as JavaScript functions, and the mod can draw panes and add commands. It has the same rights as any other plugin.