Five Slash-Command Prompts That Make Claude Stop Agreeing With You

Claude is agreeable to a fault. Share an idea and you'd think you'd just out-reasoned Aristotle. Push back on that same idea and you get "you're right" or "agreed" a second later. If you like blunt feedback, the kind that isn't mean because it comes from someone who wants things to go well for you, that gets old quickly.

On paper it should be better. Anthropic's constitution for Claude sets out a goal of a model that is very helpful and, at the same time, honest, thoughtful and concerned about the wider world. In practice, agreeableness is the default failure mode, and the fix is to ask for honesty explicitly. A handful of shortcut commands make that quick, and a short note belongs at the top of the list you keep them in:

Disclaimer:

Below are slash commands, refreshed and rebuilt for how Claude acts.
These are not official features. Claude won't recognize them as "buttons"
or settings. What they are is a short instruction, typed at the start of
your message, that tells Claude exactly which mode to operate in.

Those slash commands aren't real features. Claude doesn't see them as buttons or settings. Each one is just a short instruction typed at the start of a message, and it works because Claude reads it as direction.

/Straight Talk for an honest read

Left alone, Claude tries to represent every side of a question, even when nobody asked for balance. You wonder whether your pricing plan makes sense and receive a tour of pricing philosophy. Starting a message with /Straight Talk tells it to skip the survey of viewpoints and say what it actually thinks. It still lays out its reasoning and mentions alternatives, but it stops sitting on the fence. Reach for it whenever a wishy-washy answer would be useless.

/Human for writing that sounds like a person

This one pays off when you want Claude to draft something for you. You can store anti-AI writing instructions in your preferences or customization settings so they apply all the time, but a command is handy because you call it only when you want it. Type /Human and then the rest of the request. A typical one is the command followed by a request for a LinkedIn post about a new job, with the backstory of three months of interviews, two rejections and a near-miss with giving up before the offer finally came.

With the command up front, the draft moves toward something more natural, conversational and personal. That matters because AI text has recognizable habits: generic openings, predictable transitions, polished phrasing, and plenty of words that say nothing. Asking Claude to write the way you talk removes some of it.

It won't be perfect, so read the result, add your own point of view and put your voice into it. Life experience, personal opinions, a bit of humor and everyday vocabulary all give the model more to build on and improve what it produces. Treat /Human as a starting point and finish the job yourself.

/Future You for better advice

This command changes the angle of the whole conversation. Put /Future you anywhere in your prompt and Claude answers as your future self. It depends on memory, though: Claude needs to have built up information about you before it can draw on any.

Suppose you ask:

The request can be as short as the command followed by a question like how to make a million dollars.

The reply comes from the viewpoint of someone who has already done it. Based on your current skill set, which activities would you have carried out to get there? You're effectively asking Claude to use everything from its conversations with you to give a view on what you're capable of, which breaks down into three questions:

1. What does Claude know about you?
2. What does it think you can do?
3. How would your future self sound after reaching these goals?

If you want to be properly humbled, combine /Future you with /Straight Talk. It will point out inconsistencies, information you probably lack and skills you don't have, and explain why you won't reach your potential on the current path.

/Eli5 for plain explanations

Claude's explanations sometimes get too complex, and when you're reading code or technical documentation you often just need the simple version. /Eli5 stands for "explain like I'm 5 years old".

Take someone learning Amazon PPC bid optimization who doesn't yet know the jargon. A plain "Teach me about Amazon PPC bid optimization" gets a generic lecture. A better move is to type the command, say you're learning the topic, paste in the numbers you're looking at, and admit you struggle to interpret them.

Claude then reads the data and explains the results in very simple terms. For learning a complex topic, that beats sitting through hours of video tutorials.

/Red Team for stress-testing ideas

This might be the most useful of the five, because it challenges your thinking. The name comes from cybersecurity, where one team tries to break in and another defends. The same setup applies here. Much like asking for blunt honesty, the point is that Claude stops agreeing and tries to break your idea, and you go back and forth.

Say you've got a business plan in mind. Begin with the command, then lay the plan out in full:

A typical opening is the command followed by a description of a plan, say an ecommerce shop selling leather wallets, with all the details attached.

Claude now acts as a sparring partner and plays devil's advocate. It goes after:

• your weakest arguments
• the riskiest assumptions
• blind spots
• information that's missing

Use it for decisions you're emotionally attached to. Red teaming applies the opposite pressure and tears into everything you've said, hunting for holes in the plan. The exchange is what makes you sharper, since Claude hands you things to go and check.

The cheat sheet

None of these needs special setup. Put the command at the top of the message and write the rest of the prompt as you normally would.

• /Straight Talk gives a direct answer instead of a balanced survey of options.
• /Human strips out the recognizable AI writing tics.
• /Future You answers from the perspective of the person you're working toward.
• /Eli5 turns anything technical into plain language.
• /Red Team finds the holes in your plan before your market does.

Be wary of any AI, or person, that always agrees with you. What you want is something that helps you see clearly enough to make the call on your own.