If you spend any time on Instagram or LinkedIn, you’ve probably seen the posts. Slick dark graphics promising “secret Claude codes” that unlock hidden modes. Type /godmode for an aggressive power version. Type /ghost to make the writing sound human. Add L99 at the end for “senior engineer quality.” Stack them together for god-tier results.
The posts pull thousands of likes and shares. People comment “all” and “repo” by the hundreds to get the full list. And almost none of it works the way it claims.
I run a small business and I use Claude every day for real work. So when these started flooding my feed, I did the thing nobody making the posts seems to do. I tested them. Then I went and read what people who actually pulled apart the tool found. The short version is that most of these codes are doing nothing, a few are doing something but not for the reason you think, and the whole framing of “secret codes” is wrong from the start.
Here’s what’s actually going on, because understanding this will save you time and probably make your results better.
There is no hidden command menu inside Claude that these codes unlock. When you type /godmode into a normal Claude chat, the words /godmode get read as plain text, same as if you’d typed “please.” The model sees them and moves on. There’s no switch being flipped behind the scenes. A developer named Amit Kothari ran these straight through Claude’s developer tool with cost tracking turned on, and /godmode came back as “Unknown skill: godmode” in eleven milliseconds, zero tokens used. The request never even reached the AI. It got rejected as a command that doesn’t exist. He cross-referenced the codes against a half-million lines of the actual source code that leaked earlier this year and found zero evidence any of them are real features.
So why do some of them seem to work? This is the part worth understanding.
Take /ghost, the one that supposedly makes writing sound human. When you paste it in front of some robotic text, Claude often does clean up the writing. Not because /ghost triggered a secret mode, but because the model has seen that pattern enough times in its training to associate it with “rewrite this to sound less AI.” It’s reacting to the word the way it reacts to any instruction. Plain English does the exact same job. Type “rewrite this so it doesn’t sound AI-generated” and you get the same result, often better, because you can be specific about what you actually want.
OODA is another one that appears to work. People type it and Claude structures its answer around observe, orient, decide, act. That’s not a hidden feature either. OODA is a real decision-making framework, and Claude recognizes it the same way it would recognize “SWOT analysis” or “pros and cons list.” You’re not unlocking anything. You’re naming a framework the model already knows, and it’s organizing around it. Useful, sure. Secret, no.
Then there’s /godmode itself. When testers gave it a fair shot, all it did was make the answer longer. Not better. Not more powerful. Just more words. One writer made the point perfectly: if you want Claude’s most thorough analysis, you don’t need a code. You write a clear, demanding request. Tell it to be exhaustive. Tell it to cover competing viewpoints. Tell it you’d rather have too much depth than a shallow overview. That’s the whole trick. You built “godmode” yourself by writing a better prompt.
This is the thing the viral posts get backwards. They sell the idea that the power is locked away behind insider codes, and you just need the secret list. The reality is the opposite. The power was never hidden. It’s in writing clear instructions about what you want, in what order, with what level of detail. Anthropic publishes a free guide on exactly how to do this. The people getting the most out of Claude aren’t typing magic words. They’re describing their task well.
I’m not saying these prompt prefixes are worthless. If typing /ghost out of habit gets you to think “make this sound human,” and that gets you a better result, fine, keep doing it. It works as a mental shortcut for you, not as a feature in the tool. The danger is believing the marketing. People genuinely think they’re missing out on a locked version of the tool that everyone else has access to, and they go hunting for longer and longer lists of codes instead of just learning to ask for what they need.
There’s also a distinction the posts blur completely. Claude Code, the developer tool you run from a command line, does have real slash commands. Things like /clear to reset the conversation, /compact to compress a long session, /cost to see what you’ve spent. Those are documented and real. But they’re housekeeping commands for managing a coding session, not magic personality switches. The viral posts mash the real developer commands together with the made-up codes so the fake ones borrow credibility from the real ones. If you’re not using the developer tool, none of the slash commands apply to you anyway.
So here’s what I’d actually tell a fellow business owner who saw one of these reels and got curious.
Skip the code lists. They’re engagement bait, and chasing them teaches you a habit that doesn’t carry over to actually getting work done. What carries over is being specific. Tell Claude what you want, what format you need it in, and enough about your situation for it to do the job. Treat it like a sharp assistant who’s good at the work but doesn’t know your business yet. That’s it. That’s the whole skill the codes are pretending to replace.
Do that and you’ll beat any secret code on the internet. The honest answer is just less fun to post about than a screenshot full of slashes and emojis, so nobody’s making a reel about it. But there was never a locked door. You already have the whole tool.
