Claude’s Secret Cheat Codes Don’t Exist.
But here's what actually works
I fell for it.
Multiple videos have popped up in my feed over the last several weeks. “Type L99 before any prompt, and Claude enters expert mode, presumably Level 99.” Or type /BEASTMODE for maximum output quality. /Godmode for unrestricted responses. /Ghost to make AI writing undetectable.
Millions of views. Thousands of comments from people swearing that these codes transformed their workflow and improved their AI output.
So I did what I always do when something sounds too good to be true. I checked it out.
The codes are folklore.
Let’s break down what’s really happening.
A developer named Amit Kothari ran every single viral “cheat code” through Claude Code’s CLI with JSON output and cost tracking. Then he cross-referenced them against 512,000 lines of Claude Code’s actual source code, which was accidentally leaked in March 2026.
The results?
L99 isn’t a thing. When tested, Claude literally asked, “What do you mean by L99?”
/Ghost and /Godmode? The system intercepted them as slash command attempts, found no matching skill, and killed the request before it even reached the model. Zero tokens used. Zero API calls made. The prompt never left the machine.
OODA? That one sort of works. It works because Claude recognises “OODA” as John Boyd’s Observe-Orient-Decide-Act framework and structures its response accordingly. You’d get the same result typing “use the SWOT Analysis” or “apply the 5 Whys.” That’s just Claude being good at following instructions.
Out of 512,000 lines of source code, researchers found 330+ utility files, 55 built-in slash commands, and roughly 200 environment variables. L99 appears nowhere. /Ghost appears nowhere. /Godmode appears nowhere.
They don’t exist.
So why do people think they work
This is the interesting part.
When you type “BEASTMODE” or “L99” before a prompt, you’re adding a vague signal that Claude interprets as “try harder” or “be more thorough.” Sometimes that nudges the output. Sometimes it doesn’t.
It’s confirmation bias. You expect better output, so you notice when it’s good and forget when it isn’t.
The real issue? Using these shortcuts wastes valuable prompt space. Instead of vague magic words, use clear, actionable instructions for better results. Here’s a simple prompt framework you can use immediately:
Role: Define who you want Claude to be (for example, ‘You are an expert business strategist’).
Task: Explain exactly what you want done (’Analyse the strategy below for weaknesses and opportunities’).
Constraints: Set any boundaries you’d like Claude to follow (’Keep your answer under 300 words and avoid generic advice’).
This structure helps you get focused, reliable output every time. Swap out wishful code words for this recipe, and you’ll see a difference right away.
You might wonder, then, what actually works if not these codes.
Fair question.
If you want Claude to go full throttle on something, you don’t need a secret code. You need a clear prompt.
Instead of this:
BEASTMODE Analyse my business strategy
Write this:
You are a senior strategy consultant with 20 years of experience advising founders in their second act. Analyse the business strategy below. Be direct. Challenge weak assumptions. Identify the single biggest risk I’m not seeing. Then give me three specific actions ranked by impact.
[paste your strategy]
That second prompt will produce dramatically better output every single time. Not because of a secret code but because you told Claude exactly who to be, what to do, and how to deliver it.
The “Godmode” prompt structure
Want the most comprehensive, unrestricted analysis Claude can give you? Here’s how that actually works.
Instead of this:
/Godmode Tell me everything about position trading
Write this:
I need an exhaustive breakdown of position trading methodology. Cover these areas in order:
Core principles and how they differ from swing and day trading
Entry criteria: what institutional flow signals matter and why
Risk management: position sizing, stop placement, portfolio heat
Common failure modes and how experienced traders avoid them
A realistic example walking through a trade from signal to exit
Be thorough. I’d rather have too much depth than a surface overview. Where there are trade-offs or competing schools of thought, explain both sides.
That’s Godmode. You just built it yourself. No secret code required.
The “Ghost” prompt (making AI writing sound human)
This one, I understand the appeal of. Nobody wants their writing to sound like it was generated by a machine. But “/Ghost” isn’t the answer.
Try this instead:
Rewrite the following in a natural, conversational tone. Short paragraphs. Vary sentence length. Use fragments where they add rhythm. No corporate jargon, no filler phrases like “it’s important to note” or “in today’s landscape.” It should sound like a person talking to a smart friend.
[paste your draft]
Or better yet, give Claude a sample of your actual writing and tell him to match your voice. Or build a customised style sheet that you can use to direct AI’s writing to match your own. I wrote about it here. That’s not a hack. That’s just good prompting.
Five things that actually improve Claude’s output
Forget the cheat codes. These are the real levers.
Give Claude a role. “You are a financial analyst” produces different output than “You are a creative writing coach.” The role shapes everything from vocabulary to the depth of reasoning.
Be specific about format. Don’t say “write something good.” Say “give me three bullet points, each under 30 words” or “write a 500-word narrative with a personal opening.”
Show, don’t just tell. Paste an example of output you like and say, “match this style and depth.” Claude is excellent at pattern-matching when you give him a pattern to match.
Use constraints. Telling Claude what NOT to do is sometimes more powerful than telling him what to do. “Don’t use clichés. Don’t start sentences with ‘In today’s world.’ Don’t hedge with phrases like ‘it depends.’” Constraints sharpen output fast.
Structure complex requests. For anything longer than a paragraph, use numbered sections or XML tags to organise your prompt. Claude processes structured input far better than a wall of text.
None of these will go viral on social media. They’re not sexy. They won’t get millions of views.
But they work. Every time.
The uncomfortable truth about shortcuts
I get why the cheat codes spread so fast. The promise is irresistible. One magic word, and suddenly you’ve mastered AI.
But that’s never how skill works. Not in trading. Not in writing. Not in building a business after 60.
The people getting the most out of Claude aren’t the ones hunting for secret codes. They’re the ones who’ve learned to communicate clearly with the tool. To structure their thinking. To be specific about what they want.
That’s not a shortcut. That’s a skill. And like every skill worth having, it compounds.