Claude Code Plugins vs Skills: What's the Difference?

What is a Claude Code plugin?

A plugin is a distribution format. It is how you install a bundle of capabilities into Claude Code in one move: skills, slash commands, custom agents, hooks, even MCP server configurations can travel together inside a single plugin. You add it from a marketplace or a Git repository, point Claude Code at it, and everything it carries becomes available in your terminal at once.

The plugin itself does not know how to do anything. It is packaging and delivery, the installer that puts working parts where Claude Code can find them and registers them so they load when needed. That distinction matters more than it sounds. A plugin’s job is plumbing: versioning, dependency order, where each file lands, what gets exposed as a command versus what runs silently as a hook. When teams talk about “the plugin,” they usually mean the whole shipment, but the value lives in the parts the shipment carries.

The strength of the format is consistency at scale. If five people on a team should all have the same audit checklist, the same deploy command, and the same pre-commit hook, a plugin makes that one install instead of five manual copies that drift apart by Friday. The weakness is that a plugin is an abstraction over its contents: open one and you might find skills, but you might also find things that are not skills at all.

What is a Claude skill?

A skill is the working part. It is a file of instructions that teaches the model how to do one specific job well: the checklist for a technical SEO audit, the framework for a pricing page, the discipline that separates a professional output from a generic one. A skill loads into context when the task calls for it and shapes how the model works, then steps back out when the job is done.

The honest limitation is that a skill is method, not machinery. It cannot reach a live database or a Search Console account on its own; that is an MCP’s job, and the two are often paired. What a skill does carry is judgment: the order of steps, the edge cases a beginner forgets, the “do this, never that” rules a domain expert internalizes. A weak skill is just a list of tips. A strong one encodes the decisions, so the model behaves like someone who has done the job a hundred times.

If you want the longer grounding, start with what Claude skills actually are; the short version is that a skill is the knowledge layer, and a plugin is one way that knowledge gets delivered.

Plugins vs skills: how do they differ?

One is a container, the other is content. A plugin answers “how does this get installed and distributed?” while a skill answers “what does the model now know how to do?” A plugin can hold many skills plus commands, agents, and hooks; a skill is a single unit of competence that works wherever its file is loaded.

Comparing them head to head is a category mix-up, like comparing a shipping crate to the tools packed inside it. The crate matters for moving and storing the tools; it does nothing useful once you have unpacked. The practical line is just as clean: you browse and install plugins, but you work with skills. Day to day, you reach for a skill because you need a job done; you reach for a plugin because you need a set of jobs installed cleanly. Where this trips people up is that a single skill can be both a standalone file and the headline contents of a plugin, so the same thing wears two labels depending on how it arrived.

Do skills work as plugins (and vice versa)?

Yes, and this is the part most explanations miss. A skill is plain instruction text, so the same file loads two ways: dropped directly into your skills folder, or packaged inside a plugin and installed through Claude Code’s plugin system. Nothing about the file changes between the two paths; only the delivery does. Every skill in the forgehouse catalog works in both modes, same file, two routes, which we confirmed across the full library before shipping it.

The reverse is only half true, and that asymmetry is the real lesson. A plugin can contain skills, but it can also carry hooks, slash commands, custom agents, and MCP configs, things that are not skills at all and do not behave like them. So you can always turn a skill into a plugin, but you cannot assume everything in a plugin is a skill. Treat the plugin as a mixed shipment and read its manifest, rather than assuming the box is full of one thing.

Which should you use, or both?

Use the skill when you want one capability now: download it, load it, work. Use the plugin route when you want a set installed cleanly across a team or kept in sync from a marketplace, so everyone stays on the same version without manual copying. In practice you end up with both, because plugins are simply how skill collections travel once more than one person is involved.

If you are choosing your first capability rather than a delivery format, pick by the job, not the packaging. The SEO and AEO skills are the most common starting point because the work is concrete and the payoff is fast, while forgehouse Brain wires a full library into one system for people who want the whole shelf at once. And because skills are plain text, they are not locked to Claude: the same files carry over to any MCP-compatible AI such as ChatGPT, Gemini, or Cursor, so the method you buy once is not stranded on a single tool.