Turns out the Unreal cheat codes still work as Claude cheat codes

A LinkedIn post claimed that Claude cheat codes are secret commands. Prefix your prompt with /ghost, and the model drops its AI tells. Prefix with /godmode to enable aggressive expert mode. End with L99 for “senior staff engineer quality”. Stack them for “god-tier results”.

If incantations like that work, the classics should too. So I tested it.

The results

(Unreal’s console doesn’t use slashes. You type the command bare. The slash is a LinkedIn embellishment.)

I typed /ghost into Claude Opus 4.7. It let me walk straight through the hallucinations and fly directly into the weights. Unreal.

/god engaged invincibility. Claude became immune to context window limits. I pasted in a 40-page SharePoint admin guide, and it answered instantly. Which it was going to do anyway, but now with attitude.

/allammo gave me all the MCP servers and all the keys. Every integration unlocked.

/summon supersharepointhero triggered a secret developer mode, in which Claude admits the SharePoint modern experience was a mistake.

None of this is real

Neither is /ghost, /godmode, /artifact, OODA, or L99. They are not features. They are not parsed by any part of Claude’s product surface. They are text. Typing them into a prompt has exactly the effect of any other text you type. The model reads tone and intent and adjusts accordingly.

This is a placebo. It works on humans, too. When you open a prompt with /godmode and then write a more demanding, specific request, you get a better response. Not because /godmode did anything. Because you wrote a better request.

The actual slash commands in Claude Code are things like /clear, /compact, /model, /mcp, /config, /init, /review, /cost, /plugin, /agents, plus whatever you define yourself in .claude/commands/ (the legacy format) or .claude/skills/<name>/SKILL.md (the current recommended format, which supports the same /name invocation). Those are real. The CLI parses them. They do specific, documented things.

Claude.ai has no user-definable slash commands. What looks like one is text.

The receipts

Prefer empirical data over jokes? Amit Kothari ran these through Claude Code’s non-interactive mode with token counts and cost tracking. /ghost returns “Unknown skill: ghost” in 12 milliseconds. Zero tokens consumed. The CLI rejects it before the prompt reaches the model. /godmode fails identically.

He then cross-referenced the Claude Code source code itself, which became public on March 31, 2026, when Anthropic accidentally shipped a source map file in npm version 2.1.88. Roughly 1,900 TypeScript files and 512,000 lines, exposed for several hours before takedowns landed. Researchers catalogued 55 built-in slash commands, 5 bundled skills, hundreds of tools and feature flags. None of the viral commands appears anywhere. Not /ghost. Not /godmode. Not L99.

One honest footnote: OODA without a slash prefix reaches the model, and Claude recognises Observe-Orient-Decide-Act as a known framework. That is not a hidden feature. That is the model being good at context. You would get the same effect typing “SWOT” or “5 Whys.”

Why does this spread?

The viral post sounds authoritative. It arrives in a numbered list with a combo-stacking pro tip at the end. It gets shared, screenshotted, turned into Notion templates, and ends up in a team Slack, where a junior dev wastes an afternoon wondering why /godmode isn’t fixing their production bug.

Here is the part that actually bothers me: god and ghost are real console commands, just not in Claude. They are cheats from Unreal, the 1998 Epic MegaGames shooter that became Unreal Engine and now powers Fortnite, The Mandalorian, and most of modern gaming. Type ghost into Unreal’s console and you fly through walls. Type god and you become invincible. Those commands have real mechanics behind them, because someone wrote the code.

The viral LinkedIn cheats borrow the names and drop the mechanics. It is cargo-culting. The words sound right because they came from somewhere that worked. They do nothing in Claude because nobody wrote the code.

The underlying observation from the post is correct. Prompt framing changes output quality. Clear instructions beat vague ones. Structured requests beat rambling ones. Asking for depth gets depth. This has been true since GPT-3.

Dressing that insight up as secret commands obscures what is actually happening. Worse, it makes the real skill, writing clear, specific, well-scoped prompts, look like it requires memorising a cheat sheet.

What actually works

Tell the model what you want. Be specific. Give it constraints. Provide context. Ask for the level of detail you need—structure longer tasks with headings or numbered steps. Use system prompts for persistent behaviour. Use Projects or Styles in claude.ai for reusable setups. Use MCP and tools when you need real capabilities beyond text.

None of that is as fun as typing /godmode.

But it works.

What actually works, in practice

• AI-Driven SPFx 1.22 Upgrade: A Reproducible Process Flow — A worked example of using Claude on real engineering work. No cheat codes. Just clear instructions and a reproducible process.

Sources

• Amit Kothari, “I tested every viral Claude cheat code, here is what actually works”: amitkoth.com/claude-cheat-codes-tested
• Claude Code commands reference: code.claude.com/docs/en/commands
• Claude Code slash commands documentation: code.claude.com/docs/en/slash-commands
• Unreal console commands (Beyond Unreal Wiki): beyondunrealwiki.github.io