Five Codes That Turn Claude Into a Strategist

Most people use Claude like a search bar. Question, answer, next question. These five prompt patterns change what it does with the request. Try one on a real problem and you will feel the difference in about five minutes.

You already know the loop. You ask Claude something, it gives you a tidy list, you skim it, you move on. The answer is fine. It is also shallow, because you asked a search-bar question and got a search-bar answer.

The five codes below are not hidden buttons or secret settings. There is nothing to switch on. They are patterns you type as part of your message, and they change how Claude approaches the work. One builds instead of explains. One rewrites so the output reads human. Three of them force real analysis where you would normally get a summary.

Read one, run the example on something you are actually dealing with right now, then move to the next. That is the whole method.

What a Prompting Code Actually Is

A code is a short instruction you add to your prompt that tells Claude how to think, not just what to answer. Some go before your question, some go after. None of them require setup, an account change, or a plugin. You type the pattern, Claude shifts its approach, you get a different kind of output. That is all a code is.

Here is the full set at a glance.

Five patterns. Each one solves a different kind of problem. Work through them in order.

Code 01. /artifacts: Build It, Do Not Describe It

Add /artifacts after your prompt and Claude builds the thing instead of explaining how you would build it.

Most people ask Claude to describe a tool. You get steps, a rough plan, maybe some code you then have to assemble yourself. With /artifacts you skip that. Claude generates a working piece you can use in the same window: a calculator, a dashboard, a small internal tool, a landing page, a quick game to test an idea.

For a business owner this is the difference between a paragraph about a spreadsheet and a working spreadsheet. You wanted the result, not the recipe.

Copy This Prompt

Build me a simple ROI calculator for a service business.
Inputs: monthly retainer, hours saved per week, hourly rate.
Output: monthly savings and payback in weeks. /artifacts

Paste that, and Claude hands back a calculator you can type numbers into. No copying code into another app. No wiring anything up.

When to reach for it: any time the deliverable is a thing, not an explanation. If your prompt starts with "make me a," this is the code.

The mistake that makes it fail: asking Claude to tell you how to build the thing, then adding /artifacts out of habit. If you describe the steps, you get steps. Ask for the finished piece and let Claude do the building.

Code 02. /ghost: Make It Sound Human

Add /ghost before your prompt and Claude strips the machine tells out of the writing.

AI writing has a signature. Stiff phrasing. Constant hedging. The same three transition words in every paragraph. Your reader clocks it in a sentence, and the moment they do, they stop trusting the message. /ghost pulls those tics out. What comes back has natural rhythm and plain words, the way you would actually write it if you had the time.

Use it on anything a real person reads: client emails, social captions, replies, follow-ups, proposals.

Copy This Prompt

/ghost Write a 4-line reply to a client who asked why the
project slipped a week. Honest, calm, no corporate filler.

Pair it with a normal draft: write the first version without the code so Claude gets the structure right, then re-run that draft with /ghost to fix the voice. You get the thinking from the first pass and the human sound from the second.

The mistake that makes it fail: using /ghost and then leaving the output untouched. Read it once out loud. If a line still sounds like a press release, tell Claude which line and say fix it. The code gets you most of the way; your ear closes the gap.

Code 03. *: Get Every Angle at Once

Type a * before your question and Claude works it from several viewpoints in one answer instead of handing you a single take.

A normal answer gives you one opinion. Useful, until the obvious opinion is the wrong one. The star tells Claude to look at the same question as the optimist, the skeptic, the operator, and the customer, then show you where those views disagree. You are buying the trade-offs, not a verdict. For any decision where being wrong is expensive, that is what you want to see before you commit.

Copy This Prompt

* Should a 5-person agency raise prices 30% this quarter?
Give me the founder view, the client view, and the
cash-flow view, then where they conflict.

The last line does the heavy lifting. Naming the specific angles you care about, then asking where they conflict, is what turns a bland "it depends" into a real map of the decision.

When to reach for it: any "should I" question you are about to answer yes to before you have stress-tested it.

The mistake that makes it fail: dropping the star on a factual question. "What is the VAT rate in Portugal" does not have four perspectives. Save it for judgment calls, where more than one reasonable person would disagree.

Code 04. OODA: A Decision Framework, Not an Opinion

Drop OODA before your prompt and Claude runs the problem through a decision framework instead of dumping a generic list.

OODA stands for Observe, Orient, Decide, Act. It was built for making fast calls under pressure, when waiting for perfect information is not an option. Put it in front of your prompt and Claude works the situation in those four moves: what is actually happening, what it means for you, what you decide, what you do first. You get a read and a next move, not a menu of options you still have to choose between.

Copy This Prompt

OODA A competitor just undercut my price by 40%.
Walk the situation through Observe, Orient, Decide, Act
and give me my next move this week.

When to reach for it: the reactive moments. Something changed, you need a decision now, and you can feel the analysis paralysis coming on. OODA forces the answer to end in an action, not a shrug.

The mistake that makes it fail: using it for a slow, deep question where you have weeks to think. OODA is built for speed. For the big structural problems, the next code is the one you want.

Code 05. Prometheus: Go All the Way Down

My personal favorite. Prometheus forces Claude to break the problem into layers, work each one, then reassemble it into the most complete answer it can build.

The other four codes are fast. This one is deep. It tells Claude to decompose the question, solve each layer on its own, then put the pieces back together in sequence. It goes further than any prompting pattern I have used. Save it for the big, messy questions where a shallow answer would actually cost you money.

Copy This Prompt

Prometheus framework: How should I restructure my service
business so it stops depending on me for delivery?
Break it into layers, solve each, then give me the
sequenced plan.

That is not a question you want a quick list for. It is the kind of question that decides your next year, and Prometheus treats it that way.

When to reach for it: the decision that matters most this quarter. Give it your hardest question, not your fastest one.

The mistake that makes it fail: using it for everything. Run Prometheus on a simple email and you get three screens of over-analysis for a two-line task. Depth is the point, so spend it where depth pays.

Stack the Codes

The real move is combining them. Use Prometheus or OODA to think the problem through. Use the star to pressure-test the plan that comes out. Use /artifacts to build whatever the plan needs. Use /ghost to write the message that goes out to a client at the end.

Say you are deciding whether to hire your first project manager. You start with Prometheus framework: should I hire a PM or fix my delivery process first? and get a layered breakdown instead of a yes. Then you take the plan it produces and run it back with a * in front, asking for the founder view and the cash-flow view, to find the hole before you commit payroll to it. Once you land on hiring, /artifacts builds you a simple scorecard to rate candidates against, and /ghost writes the job post so it reads like a person wrote it, not a template.

That is four codes on one decision, in the order a strategist would use them. Think, stress-test, build, write. It is a workflow, not a single clever prompt.

Honest: What These Codes Can and Cannot Do

These are prompt patterns, not official commands. Nothing here is a documented Claude feature you can look up in a manual. Prometheus and OODA are names for a way of instructing the model; the star and the slash words are cues, not switches. That matters for one reason: if the output ever ignores the pattern, you do not file a bug, you just make the instruction more explicit. Spell out the four OODA moves in your prompt. Name the exact angles you want the star to cover. The pattern is a starting point, and your specifics steer it.

Two of these are fast tricks and three are thinking tools, and it is worth being clear about which is which. /artifacts and /ghost change the format of what you get. The star, OODA, and Prometheus change the depth of the thinking. Reaching for a depth code on a trivial task wastes your time and Claude's. Reaching for a fast trick on a decision that deserves real analysis gives you a confident answer to a question you never actually examined.

None of this replaces knowing your own business. Claude working a problem through OODA still needs the real numbers, the real constraints, the context only you have. Feed it a vague situation and you get a polished answer to the wrong problem. The code sharpens the thinking; you supply the facts.

If You Only Do One Thing This Week

Start with the star. It is the lowest-effort code and the fastest to convince you, because you already have a decision sitting on your desk that you have only looked at from one side. Type a * in front of it, name three angles, ask where they conflict, and read what comes back.

If the star earns your trust, learn Prometheus next, because that is where the real gap shows up between a list and a plan. The other three you will pick up as the need comes.

Try one. The worst case is you spent five minutes and go back to asking Claude for lists. The likely case is you stop treating it like a search bar for good.