Automated prompt combat with ELO rankings

What it does Prompts Royale is a Vue-based web app that treats prompt engineering as a tournament. You describe a task, add test scenarios with expected outputs, and the app spawns multiple prompt candidates—either auto-generated or hand-written. It then pits them against each other across your test cases and scores the results, surfacing a winner without manual A/B drudgery.

The interesting bit Instead of exhaustive pairwise comparison, it uses Monte Carlo sampling to pick duelists probabilistically—prompts with higher uncertainty or potential get more ring time—and updates their skill ratings with an ELO-style formula that tightens confidence (standard deviation) after each match. It is essentially a competitive matchmaking ladder for strings.

Key highlights

• Runs in the browser with local storage; your data never hits a backend server, though LLM API calls go from your own key
• Auto-generates both prompt candidates and test cases from a plain-text description
• Exposes every parameter—ELO learning rate, points at stake, and more—for tuning
• Uses a separate LLM call as a judge to pick duel winners, keeping evaluation consistent

Caveats

• The README is heavy on math notation but light on empirical validation of the ranking convergence
• “Local” means data storage, not offline execution: you still need a live LLM API key and bandwidth for the battles

Verdict Worth a spin if you’re burning hours manually tweaking prompts and want a structured, semi-automated way to stress-test candidates against real examples. Skip it if you already have a prompt evaluation pipeline or balk at burning API tokens on meta-prompting.