Introduction

If you've followed our posts on structured prompting and automation, you know we like AI tools that act like a reliable teammate, not just a chatbot. Claude Plugins push that idea one step further.

A skill teaches Claude how to do one thing well. A plugin packages up several skills, commands, and integrations into a single install — like handing a new hire a ready-made onboarding kit instead of a stack of loose notes. Here's what plugins actually are, how they compare to skills, and how to start using them.


What Is a Claude Plugin?

A Claude Plugin is a shareable package that extends Claude Code with custom functionality. A plugin folder can include any mix of:

  • Skills — capabilities Claude pulls in automatically when relevant
  • Slash commands — explicit actions you trigger by name, like /deploy
  • Agents — specialized subagents with their own instructions and tools
  • Hooks — automatic triggers, like running a linter on every save
  • MCP servers — connections to external tools and data sources

A manifest file (plugin.json) ties it all together with a name, description, and version. Install the plugin once, and every piece inside it becomes available immediately.

Are Plugins Turned On by Default?

No. Plugins are entirely opt-in. A fresh Claude Code install ships with zero plugins active — you start with plain Claude Code and add only what you choose. The one exception is that the official marketplace itself (claude-plugins-official) is pre-registered as a source the moment you start Claude Code, but registering the source isn't the same as installing anything from it. Nothing runs until you explicitly install a plugin from it.


Plugins vs. Skills vs. No Plugin at All

Here's the practical comparison most people actually want:

Approach What you get Best for
No plugin, default Claude Code Core Claude Code, nothing extra Quick one-off coding tasks, trying things out
A standalone skill (.claude/skills/) One capability, manually added A personal habit or one project's quirk
A plugin Multiple skills + commands + agents + hooks + MCP, installed and versioned together Team-wide standards you want everyone using the same way, every time

In short: skip plugins entirely for quick experiments, write a skill when you've got one repeatable task, and reach for a plugin once that task grows into "this should be the same for my whole team."


The Matrix Moment: Why Plugins Feel Like Cheating

Picture Neo in The Matrix, sitting in the chair while Tank loads a program. Seconds later, Neo opens his eyes: "I know kung fu." No training montage — just an instant download of a ready-to-use skill set.

That's installing a plugin. One command, and Claude immediately "knows" your deploy process or review checklist — kung fu, downloaded. The catch: Neo trusted Tank. You need to trust whoever built the plugin, too.


Best Practices

Install from official or curated sources first. claude-plugins-official is curated directly by Anthropic; the community marketplace runs submissions through automated safety screening. Still, read the README before installing anything that touches credentials.

Test locally before sharing. Use claude --plugin-dir ./my-plugin to try a plugin without a full marketplace install.


Final Thoughts

Plugins are what happens when "I wrote a useful skill" becomes "my whole team should have this." They're optional, off by default, and only worth installing once you've got a repeatable workflow worth standardizing. Check what's running with /plugin list, stick to trusted sources, and you get the "I know kung fu" moment without the side effects.

Further Reading