One shotting AAA Games - The Gauntlet Loop
One shotting AAA Games - The Gauntlet Loop
If you haven't seen it you should give this a try. The gauntlet loop can one-shot basic prototypes of AAA games using a very recursive prompt. Works amazingly well with Claude Opus 5. Follows as below:
Code:
I want you to build a first-person shooter at the level of the most recent Call of Duty games. It should be utterly perfect, visually beautiful, with every single thing done at AAA quality—from textures to physics to anything you could think of. Fan out sub-agents and have sub-agents tackle each one individually so that the game is utterly perfect. You should /loop on each item and have a separate sub-agent check it visually to ensure it looks triple A. That separate sub-agent should be a really harsh critic, and if it doesn't look triple A, it should keep going. Don't stop until each sub-agent is utterly wowed with the quality when compared with the actual Call of Duty game. It should literally compare them side by side blind and say which one looks better. Do this in ThreeJS. /loop until it's utterly perfect. Fan out sub-agents and ultracode.
I modified this to do it as like a Need for Speed racing game... let's see how this goes. i'm also using GPT 5.6 SOL Pro
share results plox! i've seen other demos of codex/deepseek/kimi but it seems opus is significantly better at 3d (for now)
Single pass on extra high, only worked on it for 11 minutes LOL realized I didn't put it on pro, I put it on Extra High by mistake, but even still. What I changed with your prompt was - With GPT, it responded after giving this to me with
"I cannot truthfully label this as a finished AAA game on the scale of Underground 1/2, nor claim independent sub-agents performed a blind NFS side-by-side comparison: This environment doesn't provide that sub-agent orchestration, and its npm mirror/browser-network restriction prevented a live WebGL visual capture here"
What I've learned with GPT is you have to give it a starting point. You have to give it assets, you have to give it a video of what you want, it needs more direction but it seems with GPT, visual direction does more than described direction.
Code:
I want you to build a street-racing drag racing game at the level of the Need for speed Underground 1 and 2 games. It should be utterly perfect, visually beautiful, with every single thing done at AAA quality - from textures to physics to anything you could think of. Fan out sub-agents and have sub-agents tackle each one individually so that the game is utterly perfect. You should /loop on each item and have a separate sub-agent check it visually to ensure it looks triple A. That separate sub-agent should be a really harsh critic, and if it doesn't look triple A, it should keep going. Don't stop until each sub-agent is utterly wowed with the quality when compared with the actual NFS games. It should literally compare them side by side blind and say which one looks better. Do this in ThreeJS. /loop until it's utterly perfect. Fan out sub-agents and ultracode.
Not bad for a first run. i guess ultracode made it use xhigh? I found that asking it to create visual harnesses speeds up development/iteration time.
Truthfully not sure... I'm still pretty new to knowing more of these specific terms of like "ultracode" or even just using Skills lol.
Usually my starting prompts are like 25x in length to yours, but also include videos, screen shots, 3d models, audio files... I make it too damn easy for the ai to spit something out
Usually my starting prompts are like 25x in length to yours, but also include videos, screen shots, 3d models, audio files... I make it too damn easy for the ai to spit something out
so "ultracode" is a claude specific term i.e. high effort - i think codex translated it. i guess loops are universal now which makes it work towards a goal
