The Free Anthropic Plugin That Audits Your Claude Code Setup

Most people never set Claude Code up

They install it, start typing, and that's it. It works. But it's working with none of the tools that make it fast, and there's no obvious moment where anything tells you what you're missing.

Anthropic ships a free plugin that fixes exactly that. It reads your real project and tells you which pieces you don't have yet.

It is called Claude Code Setup, and it is read-only. It looks at your code and makes recommendations. It does not touch your files.

What it looks for

It scans your codebase and suggests the top one or two things in each of five categories.

The point is that the answers are specific to your project. A Python data pipeline and a React marketing site get different lists.

Before you start

• Claude Code — already installed and signed in.
• A real project — open it in the folder you actually work in. Run this on an empty folder and it has nothing to read.

Install it

Inside Claude Code:

/plugin install claude-code-setup@claude-plugins-official

This does not install right away. It opens the plugin's details so you can review it, then asks where to install it:

• Install for you (user scope) — the plugin works in every project on this machine.
• Install for all collaborators on this repository (project scope) — it is enabled for everyone who works in this repository.
• Install for you, in this repo only (local scope) — it works in this repository only, and only for you.

Pick one and Claude Code installs it. From a normal terminal, claude plugin install claude-code-setup@claude-plugins-official installs it at user scope without starting a session.

If you'd rather look around first, run /plugin and go to Discover. It's in there under productivity, published by Anthropic.

Then just ask for it in plain English

This is the part people get wrong. There is no slash command to run it. You talk to it:

recommend automations for this project
help me set up Claude Code
what hooks should I use?

It reads your project and comes back with a list.

What to do with the list

Don't install everything it suggests. Start with the one that removes work you're doing by hand right now.

If you're fixing formatting after every edit, take the hook. If you keep pasting documentation into the chat, take the MCP server. If you keep re-explaining the same review checklist, take the subagent. One at a time, and you'll actually notice each one working.

The playbook the plugin can't install for you

The plugin tells you which tools you're missing. It can't tell you how to work. These three habits do more for your output than anything in that list, and they cost nothing.

1. Plan before you touch a line

The failure everyone has had: you ask for a feature, it starts editing six files, and twenty minutes later you're untangling changes you never agreed to.

The fix is to make it show you the approach before it writes anything. Claude Code has a plan mode that holds it back from editing until you approve, and you can also just say it: "plan this first, don't change any files yet."

Fixing a plan costs you one sentence. Fixing six files costs you an afternoon.

2. Keep your rules file stupid short

Your CLAUDE.md is the file Claude reads before every session. Left alone it grows — a note here, a correction there — until it's hundreds of lines of instructions competing with each other.

Long rules files make it worse, not better. Anthropic's best-practices page warns that a bloated CLAUDE.md makes Claude ignore your actual instructions, because the important rules get lost in the noise. Keep only what's actually true and actually load-bearing, and delete anything describing behaviour you already get for free.

The same plugin directory has a free Anthropic plugin for this, claude-md-management. It audits the file you already have.

3. Make it grade its own homework

The failure: it says "done," you ship, and you find out two days later that it wasn't.

The fix is to make review a step instead of an assumption. Before it calls anything finished, have it check its own work against what you originally asked for. Subagents are built for exactly this — a specialist reviewer that reads the change with fresh eyes and no attachment to having written it.

It is far better at finding a problem when it's looking for one than it is at never creating one.

Run the plugin once to see what you're missing. Run the playbook every day. The second one is where the difference actually comes from.

Three things that trip people up

• Waiting for a slash command to run it — installing the plugin is the slash command. Using it is a sentence.
• Expecting it to change your setup — it doesn't. It's read-only by design. It tells you what to add and you decide.
• Expecting it to clean up your rules file — that's a different free Anthropic plugin, claude-md-management, in the same directory. This one recommends automations; that one audits your CLAUDE.md.

Frequently asked questions

Is it free?

Yes. It's published by Anthropic in their official plugin directory. You need a Claude plan for Claude Code itself.

Will it change my files?

No. The plugin is read-only. It analyzes and recommends, nothing else.

How do I run it?

Install it once, then ask in plain English: "recommend automations for this project."

What if I don't like what it suggests?

Ignore it. Nothing is applied until you say so.

Is there anything else worth grabbing while I'm in there?

The same directory lists more than 300 plugins. claude-md-management is the other Anthropic one most people want.

Get it

The plugin: github.com/anthropics/claude-plugins-official/tree/main/plugins/claude-code-setup (opens in a new tab)

The full directory: github.com/anthropics/claude-plugins-official (opens in a new tab)