Claude Code plugins, reviewed before you install

What are Claude Code plugins?

A Claude Code plugin is a directory of skills, agents, hooks, MCP servers or other components that Claude Code installs and loads as one unit, according to the Claude Code plugin docs. You add one with /plugin install name@marketplace, and it comes from a marketplace, which is a catalog that lists plugins and where to fetch each one. Treat every plugin as code you are about to run, because it can execute commands on your machine with your user privileges.

The table above lists the 121 repositories in our review set that ship a Claude Code plugin manifest. Of those, 53 got an “install” verdict, 66 “caveats” and 2 “skip”. Only 3 have a clean security status, and 43 are flagged for broad access. Below: what you are installing, and how to choose.

Plugins vs skills: what is the difference?

A skill is a folder with a SKILL.md file that Claude loads when a task matches. A plugin is the packaging around one or more skills and other parts. Per the docs, a plugin can hold:

• Skills: SKILL.md instructions Claude loads when relevant, which you can also run as a command.
• Agents: subagent definitions Claude can delegate to.
• Hooks: commands Claude Code runs at points in its lifecycle, such as after every edit.
• MCP servers: tool servers Claude Code connects to while the plugin is enabled.

The manifest is a JSON file at .claude-plugin/plugin.json that gives the plugin its name. Skills, subagents, hooks and MCP servers all work on their own without a plugin; a skill saved in ~/.claude/skills/ is available in every project on your machine. Reach for a plugin when you want several pieces packaged as one unit, installed with one command and updated from its marketplace. Skills that come from a plugin are namespaced as /plugin-name:skill-name, so they do not collide with your own.

If you only want one skill, see how to install Claude skills. For how skills differ from MCP servers, see skills vs MCP.

The official Claude plugins marketplace

Anthropic publishes two general-purpose marketplaces, each a GitHub repository with a .claude-plugin/marketplace.json file (Anthropic’s marketplaces):

	Official	Community
Repository	anthropics/claude-plugins-official	anthropics/claude-plugins-community
Marketplace name	claude-plugins-official	claude-community
Contents	Plugins Anthropic maintains, plus ones from partners and other authors	Third-party plugins that authors submitted to Anthropic
How you get it	Claude Code adds it the first time you start an interactive terminal session	You add it yourself

Most of what claude-plugins-official lists comes from partners and other authors, not Anthropic. Anthropic’s own set includes commit-commands, code-review and feature-dev. Browse it from the Discover tab of /plugin, or on the web at claude.com/marketplace. We reviewed the repository at anthropics/claude-plugins-official. It is a trustworthy source for Anthropic’s own plugins, but it is a catalog, so each plugin needs its own look.

Many popular plugins sit in no Anthropic marketplace at all. They live in their authors’ repositories, and Anthropic does not review those marketplaces.

How to add a marketplace and install a plugin

These steps come from the official install guide.

1. Add a marketplace (skip this for the official one). In a session, run /plugin marketplace add owner/repo. A full git URL, a local path or a hosted marketplace.json URL also work. Add #ref to pin a branch or tag.
2. Open the plugin. Run /plugin install plugin-name@marketplace-name. In a session this opens the plugin’s details instead of installing at once, so you can review it first. Plain /plugin opens the Discover tab, which you can search.
3. Read the details pane. The Will install section lists the commands, agents, skills, hooks and MCP and LSP servers the plugin adds.
4. Pick a scope. User scope enables it for you in every project, project scope for everyone in the repository through .claude/settings.json, and local scope for you in this repository only.
5. Check it loaded. Type / and look for /<plugin>:<skill>, or run claude plugin list in your shell.

From a script, claude plugin install <name>@<marketplace> does the same without a session. Later you can disable, update or remove a plugin from the Installed tab or with claude plugin disable, claude plugin update and claude plugin uninstall. Auto-update is on by default for the official marketplace and off for third-party ones.

Are Claude Code plugins safe?

Not by default. The plugin security page says a plugin you install can execute arbitrary code on your machine with your user privileges. Anthropic’s trust warning adds that it does not control what MCP servers, files or other software plugins include and cannot verify they will not change.

What that means in practice:

• Hooks and monitors run as shell commands. Claude Code’s permission rules and sandbox cover the tool calls Claude makes, not the code a hook runs by itself.
• MCP servers from the plugin run as processes on your machine.
• A plugin’s bin/ directory is added to the PATH of the Bash tool’s shell.
• Skills, commands and agents enter Claude’s context as instructions, so a badly written or malicious one can steer what Claude does with tools it already has.
• Updates can change files you already reviewed when auto-update is on for that marketplace.

Our reviews work the same way. A scanner’s hit count is a starting point; what matters is what the files can reach. The 147 hits on the official catalog were almost all examples inside plugin-development guides. The real findings were smaller and more specific: scripts that launch claude -p subprocesses, and chat-channel plugins that store a bot token in a local .env file. See how we review and the security overview.

How to choose a plugin

Before you click install:

1. Read what it adds. Open hooks/hooks.json, .mcp.json and anything in bin/. The details pane shows that a hook exists but not what it runs.
2. Check cost. Every enabled plugin puts the name and description of each skill, agent and command into context on every turn. Plugins in the official marketplace show a context-cost estimate, and claude plugin details <name> shows an installed plugin’s footprint.
3. Install one plugin, not the catalog. Large marketplaces such as wshobson/agents (94 plugins) and daymade/claude-code-skills are collections. Both got a “caveats” verdict from us, and our advice for both is to pick single plugins.
4. Look at the license and the maintainer. Several repos here have no license file or an unclear one.
5. Prefer plugins with evidence. Evals, tests and recent commits count for more than star totals.

Best Claude Code plugins? Real ones from the reviewed set

There is no single ranking of the best Claude Code plugins, because the right one depends on your stack. These six all ship a plugin manifest, and the verdicts are ours, from reading the repositories.

• obra/superpowers: a 15-skill methodology that makes the agent clarify the goal, plan and verify. Verdict: install. Its only network call is an optional logo that reveals the version, and it can be disabled.
• everyinc/compound-engineering-plugin: a brainstorm-plan-build-review loop. Verdict: install, with a warning that the optional /lfg pipeline pushes branches and opens PRs, so set git permissions deliberately.
• chromedevtools/chrome-devtools-mcp: Google’s browser debugging toolkit. Usage statistics are on by default, and the agent gets full access to the browser it controls.
• microsoft/azure-skills: 94 Azure skills plus the Azure MCP Server. Use a least-privilege Azure identity, because the agent acts on live resources.
• phuryn/pm-skills: nine product-management plugins that are instruction text only, with one of our three clean security statuses.
• anthropics/financial-services: Anthropic’s finance reference plugins. Credible, but connectors depend on paid data providers, and output is draft-for-review.

For a ranked shortlist beyond plugins, see the best Claude skills; for single skills see the skills directory. Browse by topic in categories such as dev workflow.

Plugins in Cowork and claude.ai

The same plugin format installs on claude.ai and in Cowork, but a different set of components loads there. Chat ignores agents and hooks, for example. See Claude Cowork plugins and skills for what Anthropic documents.