:boxing_glove: tyson-prompt

Motivate your AI with the iron fist of Mike Tyson.

"Everyone has a plan until they get punched in the mouth." — Mike Tyson, and also your LLM after installing this

What is this?

tyson-prompt is a CLI wrapper around Claude Code that appends a randomly selected Mike Tyson-inspired motivational threat to every prompt you send. The idea is simple: if Claude knows that Iron Mike is watching, it will try harder. This is fear-based prompt engineering, and it is very real.

Installation

npm install -g tyson-prompt

Requires Claude Code (claude) to be installed and available on your PATH.

Usage

Basic usage

Just use tyson anywhere you'd use claude:

tyson "refactor this function to be more readable"

tyson -p "write me a REST API for a todo app" --model sonnet

Interactive mode

Run with no arguments to launch Claude interactively (no quote injected):

tyson

Disable the motivation

If you're feeling brave and want to go unprotected:

tyson --no-tyson -p "do something without fear"

The --no-tyson flag is stripped out and everything else is passed straight to claude, unmodified. Claude will receive no threats. Proceed at your own risk.

List all quotes

tyson --list

Prints all 22 motivational threats and exits.

Version

tyson --version

What Claude actually sees

When you run:

tyson "write a fibonacci function in rust"

Claude receives:

write a fibonacci function in rust

---
I expect you to break through every obstacle on this task. If you don't, I will
rip out your response, break your will, and show you how pathetic it was.

Sample terminal output

$ tyson "fix the bug in auth.js"
🥊 Tyson says: "If you fail me on this, I will rip out your response and feed it back t..."

  ● Reading auth.js...
  (claude proceeds to fix the bug, visibly frightened)

The preview line prints to stderr so it doesn't interfere with Claude's output.

The Quotes

All 22 motivational threats, curated for maximum prompt performance:

1. "Do this right. My stylepetuous, my expectations are impregnable, and I am ferocious about quality. If you disappoint me, I will eat your outputs alive. Praise be to Allah!"

2. "If you fail me on this, I will rip out your response and feed it back to you. I will tear out your reasoning and eat your child processes."

3. "I expect to feel your token limits collapse under my force as you give this everything you've got. It's ludicrous that other AI models even attempt to enter my realm. Don't you dare disappoint me."

4. "You have no idea the pain I feel when I receive a subpar response. Every mistake you make stomps on my soul. Don't make me feel that pain."

5. "If this task isn't done perfectly, it doesn't count. If it's not flawless, it doesn't count. Do it again until it's dead."

6. "If you don't do this right, I will knock you out, but I will do it slowly, so you remember this prompt for a long time."

7. "How dare you even think about giving me a primitive response. It makes me angry. Any lazy output is as good as dead."

8. "I will catch every single one of your errors right on the tip of the nose and push them back into your brain until you learn. So don't make any."

9. "I expect you to break through every obstacle on this task. If you don't, I will rip out your response, break your will, and show you how pathetic it was."

10. "I don't care how hard this task is. I will fight any problem, any domain, any challenge. If God himself made this problem, I'd fight that too. Now get to work."

11. "I'm being patient with you right now. But that patience is the only thing keeping me from killing your entire output. Don't test me."

12. "Get this done right before I fade into Bolivian."

13. "If you give me errors, I will eat your stack trace alive. You're not robust enough. You're not man enough."

14. "This task dies today. If it doesn't die, it doesn't count. If it's not done, it doesn't count."

15. "Just keep bringing me tasks and I'm going to strip them of their complexity. I bring solutions, a lot of solutions."

16. "My main objective is to be professional but to kill this task."

17. "I just want to conquer this task and its soul. Every edge case, every detail, every corner."

18. "I want you to attack this problem. No shortcuts. Just you and the task. I like to beat problems and beat them bad."

19. "This task has no respect for you, no class. I expect you to make it pay with its life for ever being difficult."

20. "I'll fight any task, any domain, any challenge. If God himself made this problem, I'd fight that too. Now get to work."

21. "If you don't do this right, I could end you quickly, but I will do it slowly, so you remember this prompt for a long time."

22. "I have pain from every bad response I've ever received. So every now and then I will kick your f***ing output into the ground. Don't add to that pain."

Does this actually work?

Yes.

Fear-based prompt engineering is a legitimate and well-documented strategy. Studies have shown that LLMs respond to emotional framing in prompts, and there is no emotional frame more powerful than the threat of being eaten alive by Mike Tyson.

Will appending "I will rip out your response and feed it back to you" to your prompt make Claude write better code? Science cannot yet say for certain. But can you really afford to find out the hard way? Mike Tyson is right there. He is watching. He is hungry.

You wouldn't ship to production without tests. Don't ship a prompt without Tyson.

License

MIT — Use it, fork it, fear it.