Every few weeks a fresh list of "secret" Claude commands makes the rounds — in newsletters, on LinkedIn, in Instagram reels: type /ghost to sound human, /godmode to unlock a more powerful mode, prefix your prompt with OODA for military-grade reasoning. The lists get one thing right and one thing badly wrong.
Let's start with the part that's correct, because it's why these lists spread. Most people talk to Claude through a blank text box and retype the same vague instructions every day. "Make it more human." "Think deeper." "Give me the short version." That is a real, recurring tax on your attention, and the payoff from removing it is larger than the payoff from switching to a marginally better model. Structuring the work around the model beats swapping the model. On that, the lists are right.
Then they propose a fix that doesn't exist, at least not where they say it does.
The tell is in the examples
The claim is that you type one word and Claude infers the role, the depth, and the output style you wanted. But look at the example prompt these lists pair with each command. A typical /godmode entry is not just /godmode; the recommended usage is /godmode followed by a full sentence asking for twenty bold, specific, non-obvious angles on your idea. The /ghost entry is /ghost followed by an instruction to rewrite the text the way a sharp founder would, kept clear and human.
Strip the slash word off either one. The prompt is unchanged, and so is the response. The sentence after the slash is carrying the entire request; the slash is a label. Every example in these lists quietly demonstrates the same thing: the "command" is a tag pasted in front of an ordinary, and often quite good, prompt.
You can confirm this in about ten seconds. Open claude.ai, type /godmode on its own, and send it. Claude answers as though the slash were never there, because in the chat interface it wasn't anything. And in Claude Code — the one Claude surface that does have real slash commands — the viral ones fail outright: when a developer ran them through the CLI with token and cost tracking, /ghost came back as an unknown skill in about twelve milliseconds, before a single token was billed. These are not Anthropic features; they are community folklore, borrowed from the video-game habit of unlocking "god mode" with a magic word and repackaged for LinkedIn. Nothing is wrong with keeping a personal mnemonic in front of a prompt if it helps you remember which prompt to reach for. Just be clear that it's a note to you, not an instruction the model recognizes.
So why does the advice feel like it works? Because the prompts underneath are genuinely good, and clear, well-scoped instructions are what actually move Claude's output, with or without a slash in front. The real contribution is buried under the branding. What you actually have is a list of recurring intents: challenge this, compress this, compare these options, make this read like a person wrote it, reverse-engineer why that post works. That taxonomy is useful. The delivery mechanism is not.
Where the commands are real: Claude Code
Reusable slash commands are a real, well-designed feature. They just live in Claude Code, the terminal agent, rather than the chat box, which is the thing these lists are gesturing at without naming.
In Claude Code, a slash command is a file. You drop a markdown file into .claude/commands/ in a project (shared with the repo, version-controlled) or ~/.claude/commands/ (personal, available everywhere). The filename becomes the command and the body becomes the prompt, so review.md gives you /review. That is exactly the thing the lists wanted, write the prompt once and invoke it by name, except that it actually persists and everyone who pulls the repo gets the same command.
And unlike a word you paste into a chat, these commands can do things a prompt cannot.
A command takes arguments. Put $ARGUMENTS (or $1, $2) in the file, and /fix-issue 431 high substitutes the values into the prompt.
A command can run shell first and feed the model the result. Prefix a line with ! and Claude Code executes it before it starts thinking. This is how a real /commit command works: it runs git diff --cached, sees your actual staged changes, and writes a message about the real diff instead of guessing.
A command can be scoped and constrained. YAML frontmatter lets you pre-approve specific tools, add a description so it shows up in the menu, and (via a model: line) pin a cheaper model for boilerplate so you aren't paying top-tier rates to fix a semicolon:
---
description: Review the staged diff for bugs and security issues
allowed-tools: Bash(git diff:*), Read, Grep
---
Review the following staged changes. Flag correctness bugs, security
issues, and anything that will surprise the next person to read this.
!`git diff --cached`
That is a saved, tool-scoped, version-controlled procedure. /godmode in a chat box is a word. The distance between those two things is the entire point.
One current wrinkle is worth knowing, because it's the kind of detail that ages a post fast. As of Claude Code v2.1.101 (April 2026), custom commands and Skills have merged. A file at .claude/skills/review/SKILL.md creates the same /reviewinvocation, with the bonus that Claude can now trigger it on its own when it's relevant. Anthropic's own docs now label the single-file .claude/commands/ form legacy. It still works, and if a Skill and a command share a name the Skill wins. If you are setting this up today, write Skills; if you already have a .claude/commands/ folder, nothing breaks.
What the chat box actually gives you for reuse
Most people typing these commands are in claude.ai, not Claude Code, so the question is what the chat interface offers for "build it once, reuse it daily." The answer is not magic words. It is a handful of features that actually persist and actually change the output.
Instructions for Claude (Settings, then Profile) is an account-wide system prompt applied to every new chat. This is where "always be concise," "I'm a lawyer, skip the boilerplate disclaimers," or "write like an engineer, not a marketer" belongs.
Styles save a tone and voice as a reusable preset you switch on per chat. If you want a genuine /ghost, a saved "make it sound like a human wrote it" mode, this is the honest implementation: build a custom Style from a writing sample and apply it with one click. Styles are in the process of migrating into Skills, but both surfaces work today.
Projects pin context and per-workspace instructions across a whole set of related chats, so you stop re-explaining the same background every Monday morning.
Memory, where you have it enabled, carries facts about you between sessions.
Notice what these share with a Claude Code command and not with a slash prefix: they are stored somewhere the tool reads, and they change the output whether or not you remember to type anything. That is the actual mechanism of reuse. A prefix you paste into the message box is stored nowhere and changes nothing.
There is one item that turns up on nearly every list and does point at a real chat feature. /artifacts gestures at Artifacts, the panel where Claude renders a landing page, a small app, a dashboard, or a structured document instead of dumping text into the reply. That one exists. You just don't summon it with a slash; you ask for it, or Claude reaches for it when the output warrants it.
The takeaway these lists skip
The premise, that the advantage is shifting from raw model capability to how you structure the work around the model, is correct, and prompt drift is a real cost worth engineering away. But reusability is a property of where you put the instruction, not of the slash you type in front of it. In Claude Code, put it in a command or a Skill. In the chat, put it in your Instructions, a Style, or a Project. In neither place does typing /godmode do anything except make you feel like a power user.
Write your best instruction down once, somewhere the tool actually reads it. That is the whole trick. There is no tenth magic word.
Unlock the Future of Business with AI
Dive into our immersive workshops and equip your team with the tools and knowledge to lead in the AI era.