0xreversed/with.ai
WebsitesWorksSource available
Target: Chrome T-Rex Runner (open-source browser game)

Rebuilt Chrome's T-Rex Runner as a tested HTML5 game using REA and Codex

Chrome T-Rex Runner (open-source browser game): Codex rebuilt an open-source browser game from REA evidence.
Image: GitHub (zhxunzh84/rea-dino-reverse-engineering)

What was done

REA inspected the JavaScript of an open-source browser game through Chrome DevTools Protocol, and Codex rebuilt it with regression tests.

zhxunzh84 used REA and Codex to study the scripts of an open-source browser dinosaur runner and rebuild it as a self-contained HTML5 Canvas game with its own tests. A live demo runs in the browser with no build step.

How it was done

REA, an MCP tool for reverse engineering, found the target browser page through Chrome DevTools Protocol and inspected the loaded JavaScript without changing it. The captured source and evidence went to Codex, which analysed the behaviour, wrote the implementation, reviewed it and orchestrated the tests.

The README names one subtle detail as the main lesson: movement uses elapsed time, but speed increases once per eligible update. Keeping that distinction mattered more than matching the original look, so the artwork is drawn from scratch.

Results

The result is a single HTML file with a speed slider, auto acceleration, swept collision checks and a session best score. Embedded unit and integration tests run the actual game logic, and a Node.js harness runs the HTML script with mocked browser interfaces. The author describes it as practice in evidence-based investigation and checking AI-assisted code.