Plugin System Overview
A plugin packages Skills, Agents, Hooks, and MCP servers into a single directory. It allows one-command installation and automatic updates via marketplaces, as well as easy sharing across teams and communities.
플러그인이 필요한 시점
Skills, Agents, Hooks, and MCP servers work as independent files without plugins. For example, skills placed in ~/.claude/skills/ are available across all projects.
Plugins are beneficial when:
- Bundling multiple components to share with team members
- Applying the same configuration across multiple projects
- Creating distributable artifacts requiring versioning and updates
For personal workflows or single-project configurations, standalone files under .claude/ are sufficient.
플러그인 구조
my-plugin/
├── .claude-plugin/
│ └── plugin.json # Manifest (plugin name and metadata)
├── skills/ # Skills (SKILL.md files)
│ └── review/
│ └── SKILL.md
├── agents/ # Custom subagents (.md files)
│ └── reviewer.md
├── hooks/
│ └── hooks.json # Hook handler definitions
└── .mcp.json # MCP server configuration
Only plugin.json goes inside .claude-plugin/. Component directories like skills/, agents/, and hooks/ must reside at the plugin root. Components placed inside .claude-plugin/ will not be loaded.
활성화된 플러그인이 세션에 미치는 영향
When a plugin is enabled, it is loaded in every session, not just when explicitly needed. Keep the following in mind:
- Context Cost: Names and descriptions of skills and agents that Claude can call automatically are included in the context on every turn. The full bodies are loaded only on invocation.
- Processes: Plugin-defined MCP servers launch on every session.
- Permissions: Plugin code executes with your user permissions.
Check estimated context costs in the /plugin panel before installing official marketplace plugins. Disable or remove unused plugins under /plugin → Installed → Not used recently.
마켓플레이스 구조
A marketplace is a catalog containing a list of plugins and their sources. When you first launch Claude Code, Anthropic's official marketplace (claude-plugins-official) is registered automatically.
flowchart LR
MKT["Marketplace\n(Catalog)"]
PKG["Plugin\n(Directory)"]
CC["Claude Code\n(Session)"]
MKT -->|"Discover"| PKG
PKG -->|"/plugin install"| CC
CC -->|"Skills / Hooks / MCP"| CC
Anthropic provides three marketplace tiers:
| Marketplace | Name | Notes |
|---|---|---|
| Official | claude-plugins-official | Auto-registered on first Claude Code launch |
| Community | claude-community | Manual registration required |
| Demo | claude-code-plugins | Example plugins |
설치 범위 (Scope)
When installing a plugin, choose an applicable scope:
| Scope | Location | Target |
|---|---|---|
| User scope | ~/.claude/settings.json | All projects on this machine |
| Project scope | .claude/settings.json (git-committable) | All collaborators in this repo |
| Local scope | .claude/settings.local.json | Current user in this repo only |
If the same plugin is configured across multiple scopes, the precedence is Local > Project > User.
공식 플러그인 — Claude Security
A flagship plugin available on the official Anthropic marketplace.
- Features: Multi-agent codebase vulnerability scanning — architecture mapping, threat modeling, vulnerability detection, and independent review report generation.
- Installation:
/plugin install claude-security@claude-plugins-official
/claude-security
For detailed scanning options, see Security Considerations.
다음 단계
이 챕터를 완료하셨나요?
학습 진도를 체크하여 나의 로드맵 달성률을 높여보세요.