Codex made a racing game with multiple racers, eight maps and items from one prompt, acting as designer, developer and QA tester by playing it.
Codex made a racing game with multiple racers, eight maps and items from one prompt, acting as designer, developer and QA tester by playing it.
Codex got one detailed prompt for a Three.js voxel kart racer, plus an image generation skill and a web game development skill. It then worked alone, acting as designer, developer and QA tester by playing the game. It was repeatedly re-prompted from a random list of ten generic prompts, such as finding what's missing, adding features, testing them and fixing bugs.
The build produced a racing game with different racers, eight maps and space-bar items, using over 7 million tokens. Earlier iterations are published so you can see how it improved. The source gives no quality assessment.
The whole run used more than 7 million tokens from a single initial prompt, so expect heavy usage. The prompt is summarized, not verbatim.
Build a browser-based [racing] game in [folder] with [number] racers, [number] maps, and [power-ups or items]. Then play-test it yourself, fix any bugs or balance problems you find, and write a short QA report. You are done when the game runs with one command and you have listed how to start it and what you fixed.
Claude Code got an old DirectX game running on modern hardware, decompiled much of it, built a headless parallel runner, and now writes full level guides with screenshots.
RemainderBot uses a founder's goals and task lists to write PRs with spare Claude and Codex quota, producing 111 PRs including call prep and prototypes.
A reviewer connected Dots to Slack and asked it to use its own computer to choose flight dates; it was slow and had not picked dates after a minute.