Prompt Jenga

Tap a word to pull it Keep the output standing

model output preview

How Prompt Jenga Works

This prompt works. Now remove words until it does not. Each prompt is a tower of tokens that currently produces a perfectly serviceable output. Your job is to greedily delete words — one tap at a time — and squeeze out every drop of redundancy your prompt engineer left behind.

  1. Tap a word-block to yank it from the prompt.
  2. Pull a filler token (please, kindly, very, really) and the output stays valid. Free real estate.
  3. Pull a load-bearing token and the whole output crashes into hallucinated garbage. That prompt is done.
  4. Survive? Bank it and advance to a more fragile prompt. Twelve in total, fragility rising.

What Am I Actually Scoring?

Your headline number is total words removed across all prompts before a collapse. Clearing prompts unlocks more brittle ones with fewer safe tokens to pull. It is context-window minimalism as a competitive sport. Nobody asked for this. We built it anyway.

Slop Fact: Real prompts are 40% magic words people are too superstitious to delete. "Take a deep breath," "you are an expert," "this is very important to my career" — load-bearing vibes with no measured effect, kept in production forever because nobody dares pull the block.

Back to the Slop