8 Claude codes and the features nobody turns on

Claude has two kinds of hidden power. The first is the same trick that works in any model: short codes you define yourself so you can demand argument, holes, and honest odds without retyping the instruction. The second is specific to Claude, and it is the set of built-in features that most people never turn on, which is why their Claude answers stay generic while someone else's seem to already know their business.

This covers both. Codes first, because they take one paste. Then the four features that change the tool from a chat window into something that works like it has been on your team for a month.

What are the codes?

Paste this block once. After that, type any code at the end of something you have written and Claude answers in that mode only.

Here are 8 codes. When I type one after something I've written, respond only in
that mode.

/ATTACK - Argue against what I just said as hard as you honestly can.
/HOLES - Find my unstated assumptions and what I left out.
/STEELMAN - Give the strongest possible case for the opposite position.
/SOWHAT - Tell me why this matters or doesn't. No throat-clearing.
/ODDS - Give me your honest confidence level, high, medium, or low, and what
would change it.
/PLAINLY - Strip the hedging. Give me the blunt friend version.
/NEXT - Name the single most important next step and say why it beats the
alternatives.
/FAILHOW - Tell me the most likely way this goes wrong before I commit.

Claude is a strong fit for these because it holds long context well and tends to give structured, specific pushback rather than a paragraph of hedging. /HOLES and /STEELMAN in particular return more useful answers here than in most tools.

Use one at a time. Stacking them gets you a blend, and the blend is the softened answer you were trying to escape.

What is a Project and why does it change everything?

A Project is a container that holds instructions and files, and every conversation inside it inherits both. This is the feature that separates people who chat with Claude from people who run a business on it.

The move is one Project per job rather than one for everything. Client work. Content. Finance. Each holds the context that job needs, so you stop re-explaining your business at the top of every conversation.

What goes in the project instructions:

Here is my business context. Use it in every response in this project.

What I sell: [offers and prices]
Who buys it: [audience, their situation, what they are trying to do]
How I write: [tone rules, words you use, words you never use]
My constraints: [time, budget, team, anything that rules options out]
What I already tried: [so you stop suggesting it]

When you answer, assume this context. Do not ask me to restate it.

The last line matters. Without it Claude politely re-asks for information it already has.

What does a Style do that instructions do not?

A Style controls voice specifically, and it travels across projects. You can create one from a writing sample, meaning you paste three things you wrote yourself and Claude derives the patterns: sentence length, how you open, what you never say.

This is the difference between output you edit for twenty minutes and output you edit for two. Instructions tell Claude what to think about. A Style tells it how to sound. Keeping your own voice while using AI for structure is the whole game, and use AI without losing your voice covers the judgment side of it.

What are Artifacts good for beyond code?

An Artifact is a self-contained interactive page Claude builds and renders live in the conversation. You click it, tell Claude what is wrong, and it changes. Since 13 July 2026 you can publish one to a public link that opens for anyone, with no Claude account needed.

For a business owner the useful applications are rarely software. They are calculators that price your offer, assessments that score a client's situation, planners, trackers, and interactive worksheets you hand to a client instead of a PDF. It is also the fastest route to a first digital product, which build a digital product with Claude walks through end to end.

Why connect your files instead of pasting?

Connectors let Claude read from Drive and Gmail directly. The difference is not convenience. It is accuracy.

When you paste a summary of a document, Claude works from your summary, including whatever you unconsciously left out. When it reads the real file, it works from the real thing and catches what you skipped. Ask it to compare two contracts and it reads both instead of trusting your recollection of either.

The privacy line still applies and it is not optional. Connected does not mean safe to feed anything. Client personal data, credentials, and unreleased financials stay out regardless of plan, per AI privacy basics.

What is the one line that fixes vague asks?

Separate from the codes, this one prompt fixes the most common failure, which is a vague ask returning a generic answer.

Before responding, ask me clarifying questions until you're 95% confident you can
complete this task well. Use only verifiable sources. Do not speculate.

Claude asks four or five questions, you answer them in a minute, and the output improves more than any rewrite of your original prompt would.

What does the full setup look like?

One evening, in this order. Create a Project for your main work and paste the business context into its instructions. Paste the code block in there too, so the codes load in every conversation inside it. Create a Style from three pieces of your own writing. Connect Drive if your work lives there. Then save your first three real prompts somewhere you will find them again.

That is the whole configuration, and it is why two people using the same Claude plan get very different quality out of it. Your first week with AI has the day-by-day version if you would rather spread it out, and how to give AI context goes deeper on the context brief itself.

For the design side of Artifacts, so what you build looks like your brand rather than a default, the Claude design starter kit has the style prompts and brand suffix to paste into any build.