What's the difference between Claude plugins and skills?

A skill is the instructions for one job. A plugin is an installable package that can contain several skills plus connectors and sub-agents. Plugins help a team get a ready setup in one step, and they work across chat, Cowork and Claude Code. The skills inside still decide how each job is done.

Are Claude commands the same as skills?

In Claude Code, mostly yes. Custom commands have been merged into skills: a command file and a skill with the same name both create the same `/name` command, and old command files still work. Claude Code also has built-in commands, so not every `/` item is a skill someone wrote.

What's the difference between Claude agents and skills?

A skill is a set of instructions Claude follows. A subagent is a separate helper with its own context and tool permissions that takes on a self-contained piece of work and reports back. You might give a subagent a skill to follow. Use skills for methods and subagents for splitting work off.

Should I use a Claude project or a skill?

Use both, for different things. Put one client's or initiative's chats, files and instructions in a project. Put a method you repeat across clients in a skill. A proposal skill can then work in every client project without its instructions being copied into each one.

When should I use CLAUDE.md instead of a skill?

Use CLAUDE.md for short facts and conventions Claude Code should know in every session, such as folder names or spelling rules. Use a skill for procedures and long reference material, because a skill's instructions load only when needed. If part of CLAUDE.md reads like a set of steps, move it into a skill.

When should I use a skill instead of a project or a plugin?

Use a skill whenever you have a method you want followed the same way each time, wherever the work happens. Choose a project when you need one place for a piece of work's background. Choose a plugin when a team needs several skills and connectors installed together. Most teams start with a skill.

Claude Skills vs Plugins, Projects and Commands

Quick Answer

A skill is a method for one job, and a plugin is a package that bundles skills with connectors and other parts. A project is a workspace for one piece of work's chats and files, while a command starts something when you type / (in Claude Code, custom commands are now skills). A subagent is a separate helper with its own context, and CLAUDE.md holds facts Claude Code reads every session.

What Each One Is

Term	What it is	Use it for	Example
Skill	Saved instructions for one job, loaded when the job comes up	A repeatable method	Write a proposal our way
Plugin	An installable bundle of skills, connectors and sub-agents	Rolling out a ready setup for a role or team	A sales plugin with the proposal skill and a CRM connector
Project	A workspace with its own chats, files and instructions	Keeping one client or initiative together	Everything for the Harbour Foods account
Command	Typing `/` to start a skill or built-in action	Starting a known procedure on purpose	`/proposal` in Claude Code
Subagent	A separate helper with its own context and tools	Splitting off a self-contained piece of work	A research helper that reports back
CLAUDE.md	A file Claude Code reads at the start of every session	Facts and conventions for a codebase or folder	"Use NZ spelling. Invoices live in /finance."

These are working distinctions. Not every feature behaves the same in every Claude surface.

How They Differ in Practice

Skills versus projects. Anthropic's own explainer puts it neatly: "Projects say 'here's what you need to know.' Skills say 'here's how to do things'" (Anthropic: Skills explained). A project holds the background for one piece of work, with its own chat history, uploaded files and project instructions. On Team and Enterprise plans you can share a project with view or edit access. Anthropic is also rolling out a new project experience in stages, starting with Claude Code (Anthropic: What are projects?). A skill can be used in any chat or project when its job comes up.

Skills versus plugins. A plugin is packaging. It bundles skills, connectors and sub-agents into one install, and it's saved to your account so it works in chat, Cowork and Claude Code. Some parts, such as hooks and sub-agents, run only in Cowork and Claude Code (Anthropic: Use plugins in Claude). Use a plugin to distribute a setup; use a skill to define the method inside it. Claude Cowork Skills covers plugins in Cowork.

Skills versus commands. In Claude Code, custom commands have been merged into skills. A file in .claude/commands/ and a skill in .claude/skills/ both create the same /name command, and older command files keep working (Claude Code: Skills). Built-in commands also exist, so not everything that starts with / is a skill you wrote.

Skills versus subagents. A subagent is an independent helper with its own context and tool permissions (Anthropic: Skills explained). A skill is instructions; a subagent is a separate worker. In Claude Code, a skill can even run inside a subagent when the job is better done in isolation.

Skills versus CLAUDE.md. Claude Code loads CLAUDE.md into every session, so it suits facts and conventions. A skill's full instructions load only when it's used, which suits procedures and long reference material. Anthropic suggests making a skill when part of CLAUDE.md has grown into a procedure rather than a fact (Claude Code: Skills).

Worked Example: One Proposal, Every Piece

Ironbark Consulting, a fictional firm, is preparing a proposal for a prospective client, Harbour Foods (also fictional). Here's where each piece fits:

The project holds the Harbour Foods discovery notes, the brief and past conversations about the account.
The proposal skill sets the method: sections, pricing rules, tone and the checks before finishing.
A sales plugin gives the whole sales team the proposal skill, a research skill and a CRM connector in one install.
A subagent could research Harbour Foods' recent news separately and report back, without cluttering the main conversation.
A command lets a consultant in Claude Code start the skill on purpose by typing /proposal.

None of these decides what evidence your firm needs before it quotes, or whether Claude can open the price list. That still comes from the skill's instructions and from the access Claude has.

Try It: Choose the Right Piece

Write down the job in one sentence.
If it's a method you'll repeat across clients, make it a skill.
If it's background for one client or initiative, put it in a project.
If a whole team needs the same set of skills and connectors, package them as a plugin.
If part of the job is self-contained research or checking, consider a subagent.
If you work in Claude Code and need a standing fact, put it in CLAUDE.md. If it's a procedure, make it a skill.
Start with one skill and add the other pieces only when a real need appears.

Test It: Three Cases

A normal request. In the Harbour Foods project, ask for the proposal. The skill should load, and the result should use the project's notes.
Incomplete input. Remove the budget from the notes. The skill should ask about it, not invent one from another client's project.
An unrelated request. In the same project, ask for a meeting summary. The proposal skill shouldn't take over.

Where Contexta Fits

These pieces all live inside Claude. Contexta is for the parts that need to be the same in every AI your team uses. A skill tells your AI what to do. Your knowledge base shows it how, in detail. Your company context tells it who it's for. The Contexta harness brings them together, in every AI.

Contexta works much like the CLAUDE.md and skills split, across tools. Your organisation's rules, covering terminology, naming and formats, load at the start of every session, with team rules on top. Skills load when the job comes up. Supporting material stays in your knowledge base, in your own SharePoint or Google Drive. Company, team and personal spaces do some of the organising a project does, giving each team its own home and rules, and they work in every connected tool.

A Claude project remains useful for client work inside Claude. Every change to a Contexta skill or rule is versioned, with compare and restore. Claude is tested with Contexta; see Integrations and How It Works.