The prompt that makes AI agents judge each other

Matt Shumer’s Gauntlet Loop uses separate builders and harsh critics. I turned it into a free prompt generator. ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

Matt Shumer’s Gauntlet Loop uses separate builders and harsh critics. I turned it into a free prompt generator. ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌

The Prompt Index Prompts • Toolbox • Blog New  •  Free Gauntlet prompt generator What if your AI had to impress a harsh critic? Most prompts produce one answer, one design or one block of code — then stop. The Gauntlet Loop tells capable AI agents to do something more useful: build, inspect the real result, criticise it against a demanding standard, and keep improving the biggest weakness. Forge my Gauntlet prompt → Free to try  •  no prompt-engineering knowledge needed  •  requires an agentic environment The technique, in 30 seconds Matt Shumer popularised the method while building Claude of Duty. Instead of micromanaging every step, the prompt gives a lead agent an ambitious target and a real quality bar. The agent then runs a builder-versus-critic cycle: The clever bit was not the “utterly perfect” swagger. It was stopping the builder from grading its own homework. BUILD → CRITIQUE → IMPROVE → /LOOP What makes it different? Three mechanics turn that idea into a repeatable workflow: 1 A concrete quality bar “Make it great” is vague. A reference, test suite or measurable benchmark gives the critic something real to judge. 2 Separate builders and critics A fresh, harsh critic inspects the actual output — not the builder’s polished progress report. 3 A loop with boundaries The biggest remaining gap goes back for another round. Sensible prompts also cap time, cost and attempts, because “until perfect” is not a safe finish line. I built the setup for you Describe the build. Get the Gauntlet. Tell the generator what you want to create, what “world class” looks like, and which tools or stack to use. It writes the ready-to-paste prompt for you — setting a Triple-A quality bar, assigning builder sub-agents, giving each output a fresh harsh critic, looping on the biggest gap, enforcing sensible boundaries, and finishing with a whole-project review. Forge my Gauntlet prompt → Want the theory first? Read the complete Gauntlet Loop guide. ONE IMPORTANT CAVEAT A Gauntlet prompt needs an agentic environment that can inspect files, run tools and delegate work — such as Claude Code or Codex. A normal chat window can imitate the wording, but not the complete workflow. Try it on the project you have been putting off because it feels too large for one prompt. Give the critic a real bar, set sensible boundaries, and see what survives the Gauntlet. Stephen The Prompt Index P.S. If you forge one, hit reply and tell me what you’re building — I’d genuinely like to see what people put through the Gauntlet. Telegram  •  X / Twitter  •  YouTube © 2026 The Prompt Index. All rights reserved. Unsubscribe  |  Visit website Ferryhill Durham County Durham DL17 9LH United Kingdom Powered by EmailOctopus