Add a marketplace with /plugin marketplace add, install a named package with /plugin install name@marketplace, pick a scope, then reload plugins when asked. The taxonomy: a plugin is the package, a marketplace is the catalog, a skill is reusable guidance, a slash command is an invocation surface, and a subagent is a delegated worker with its own context window.
- Adding a marketplace installs nothing. It only fetches the catalog.
- The install form is
/plugin install plugin-name@marketplace-name, and the suffix is not optional. - Anthropic runs two marketplaces,
claude-plugins-officialandclaude-plugins-community, plus a separateanthropics/skillsrepository. - Custom commands and skills now run on one system. Plugins remain the packaging and distribution layer above them.
- A plugin can run code and open network connections. Read its source, hooks, and MCP config before you trust it.
Install a Claude Code plugin in four steps
Anthropic's official marketplace is already available after your first interactive startup. Add it explicitly anyway when you are documenting a clean setup or repairing a marketplace someone removed. Adding a marketplace fetches its catalog and nothing else. No package in it becomes active until you install it.
/plugin marketplace add anthropics/claude-plugins-officialOpen the plugin manager, choose Discover, read a package, install it. The syntax below is the documented form. Swap the placeholder for a real plugin name from the catalog you selected, because the name in the repository title is often not the name of the package.
/plugin
/plugin install plugin-name@claude-plugins-official
/reload-pluginsAdd or select a marketplace
Use the Marketplaces tab in /plugin or run /plugin marketplace add with a GitHub owner/repository value, a Git URL, or another documented source.
Inspect the package
Read its description and source. Check whether it adds hooks, MCP servers, executable tools, subagents, or skills.
Install at the right scope
Choose user for your account, project for a shared repository setup, or local for a project-specific install that should not be committed.
Reload and invoke
Run /reload-plugins when the interface asks for it, then use the package's documented namespaced skill or command.
What is in claude-plugins-official?
The anthropics/claude-plugins-official repository is public, and its manifest is a curated directory Anthropic manages. Packages under plugins are built and maintained by Anthropic. Entries under external_plugins are third-party integrations Anthropic selected for discovery, still owned and maintained by their publishers. The distinction is in the repository structure, not in the install command.
That structure also shows why a plugin is a bigger thing than a command. One package can ship a manifest plus MCP configuration, commands, agents, skills, hooks, settings, language-server configuration, monitoring configuration, and executable helpers. Most use only two or three of those.
Verified Anthropic plugin and marketplace sources
| Source | What it is | How to add or use it |
|---|---|---|
| claude-plugins-official | Anthropic's curated official marketplace, including Anthropic and selected external packages | /plugin marketplace add anthropics/claude-plugins-official |
| claude-plugins-community | Anthropic's separate community marketplace | /plugin marketplace add anthropics/claude-plugins-community |
| anthropics/skills | A public skills repository with document skills, example skills, templates, and the skill-creator skill | /plugin marketplace add anthropics/skills |
| anthropics/claude-code | The demo marketplace used in Anthropic's plugin documentation | /plugin marketplace add anthropics/claude-code |
Anthropic's own documentation installs commit-commands from the demo marketplace using the fully qualified name below. The suffix is what stops two catalogs publishing the same short name from colliding.
/plugin install commit-commands@claude-code-pluginsPlugins vs skills vs commands vs subagents
The terms blur because one plugin can contain all the others. Separate them by job instead: distribution, invocation, instructions, execution context.
Claude Code extension taxonomy
| Term | Primary job | Context behavior | Typical example |
|---|---|---|---|
| Plugin | Package and distribute a collection of capabilities | Depends on the components inside it | A repository package containing skills, agents, hooks, or MCP configuration |
| Marketplace | List installable plugins and their source locations | No runtime context of its own | claude-plugins-official |
| Skill | Provide reusable instructions, knowledge, scripts, and supporting files | Runs in the current conversation by default and can optionally fork | A SKILL.md that teaches a deployment workflow |
| Slash command | Give the user an explicit /name invocation | Built-in commands run coded behavior; custom commands use the skill system | /compact or a project-defined /deploy |
| Subagent | Delegate a task to a specialized worker | Uses a separate context window and returns a result to the caller | A code-review agent defined in an agents directory |
The rule that survives contact: you install a plugin, you invoke a skill or command, you delegate to a subagent. A skill in .claude/skills needs no marketplace. A subagent local to one project needs no plugin.
Custom slash commands now run on the skills system. Both .claude/commands/deploy.md and .claude/skills/deploy/SKILL.md expose /deploy, and Anthropic points new work at the skills directory because it also gets frontmatter, supporting files, and automatic discovery. Go deeper in Claude Code skills, custom commands, and subagents.
Choose the right install scope
The installer offers three scopes. User follows you across every project. Project records the setup in the repository for your collaborators. Local applies to you in this project only and never reaches version control.
| Scope | Use it when | Main tradeoff |
|---|---|---|
| User | You want the plugin in every repository you open | Convenient, but its behavior can affect unrelated projects |
| Project | The whole team needs the same repository tooling | Reproducible, but teammates must trust and approve the package |
| Local | You are evaluating a plugin or need a private per-project setup | Safest for trials, but other collaborators do not inherit it |
Reserve project scope for packages that are genuinely part of the repository's workflow. Your personal formatter, an experimental reviewer, a connector to a service only you use: those belong in local or user scope, where they cannot surprise a teammate. Open /plugin and read the Installed tab whenever you need to know what is actually active and at which scope.
Use the Claude skills GitHub library and skill-creator
The anthropics/skills repository is a library and a set of reference implementations at once: document-oriented skills, example skills, a template, and skill-creator. It separates source-available reference skills from examples meant for learning, so read the license inside each skill directory before you redistribute anything from it.
Add that marketplace and install one of its published collections. Each collection is a plugin that delivers a group of skills, which is the package-versus-content distinction in its most literal form.
/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills
/plugin install example-skills@anthropic-agent-skillsThe skill-creator skill creates, revises, tests, and benchmarks other skills. It ships inside the example-skills collection, so the install target is example-skills@anthropic-agent-skills. There is no package named after the individual skill, and guessing one is the most common way this install fails.
A third-party example: OpenAI Codex
OpenAI publishes openai/codex-plugin-cc, a Claude Code plugin that hands review and rescue work to the Codex CLI. Note the three different names in play: the repository is openai/codex-plugin-cc, the marketplace is openai-codex, and the plugin is codex. A vendor publishing its own catalog is the normal pattern, and reading all three names off the repository is how you avoid an invented one.
/plugin marketplace add openai/codex-plugin-cc
/plugin install codex@openai-codex
/reload-plugins
/codex:setupIt exposes namespaced commands such as /codex:review and /codex:adversarial-review. The namespace prevents collisions and tells you at a glance which installed plugin owns the behavior, which is exactly what you want when a command misbehaves and you have to work out what to remove.
Troubleshoot marketplace and plugin installs
Open /plugin before you reinstall anything. Marketplaces lists your configured catalogs, Installed lists active packages and their scopes, and Errors surfaces validation and load failures that otherwise look like the plugin simply not existing. Refresh the marketplace when a newly published version is missing, then reload plugins after any install or update.
/plugin
/reload-pluginsWhen a namespaced command does not appear, check all three identifiers separately: the marketplace source, the installed plugin name, and the command or skill name that version actually exports. Never infer one from the repository title. Repository, marketplace, plugin, and command are four names, and they are routinely four different names.
Questions people ask
Plugins are distributable packages that extend Claude Code. A package can include skills, subagents, hooks, MCP servers, language-server settings, monitoring configuration, and executable helpers.
Run /plugin marketplace add anthropics/claude-plugins-official for Anthropic's official catalog. Claude Code also makes that marketplace available automatically after the first interactive startup.
Use /plugin install plugin-name@marketplace-name. For an official-marketplace package, the suffix is @claude-plugins-official. Inspect the catalog first to get the exact plugin name.
A plugin is the package; a skill is the content. A plugin distributes and versions things, and it can contain one skill, many skills, or none at all. A skill dropped straight into .claude/skills works without any plugin around it.
Built-in slash commands trigger Claude Code features. Custom commands are now handled by the skills system: both .claude/commands/name.md and .claude/skills/name/SKILL.md can expose /name, with the skills form recommended for new work.
Yes. anthropics/claude-plugins-official is Anthropic's public curated marketplace. It separates Anthropic-maintained packages from selected external integrations in its repository structure.
Anthropic publishes anthropics/skills on GitHub. You can browse the skill directories directly or add anthropics/skills as a plugin marketplace and install its published collections.
Yes. OpenAI publishes openai/codex-plugin-cc. Its documented flow adds that repository as a marketplace, installs codex@openai-codex, reloads plugins, and runs /codex:setup.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.
- Claude Code: discover and install plugins the marketplace commands, install syntax, and scopes
- anthropics/claude-plugins-official the manifest, and how plugins differ from external_plugins
- anthropics/skills the skill collections, templates, and skill-creator
- openai/codex-plugin-cc a third-party marketplace and its namespaced commands