If you're a software engineer who has ever said "I want to make a game" and then bounced off Unity tutorials, this page is for you. The games in this arcade are built on universal-game-engine — a free, MIT-licensed template for making web games with an AI assistant, built around the one thing AI-generated code desperately needs: proof it works.

The pitch, for engineers

Game code is usually untestable by construction — logic tangled into render loops, Math.random() everywhere, correctness checked by playing. This template inverts that: the entire simulation is a pure, deterministic function over plain data (seeded RNG, fixed timestep, zero DOM), so it runs headless in Node and fast-forwards an hour of gameplay in milliseconds. Bots play the real game through the same command API as the UI. Tests prove every ending is reachable; Monte-Carlo bands guard the balance; a purity gate fails the build if anyone — human or AI — sneaks nondeterminism into the sim. Rendering is a thin, swappable adapter. The whole thing deploys free as a static bundle.

• GitHub: liamitus/universal-game-engine — use it as a template
• Live demo of the included starter game

Dream one up right now

Copy this prompt into your AI of choice. It seeds the assistant with the template and a workflow that starts with finding a game you'd actually enjoy building — then builds it test-first.

I want to make a game — a small web game I can actually finish: built with an AI coding assistant, properly tested, and deployed free as a static site.

Use this template: https://github.com/liamitus/universal-game-engine — universal-game-engine, a testing-first web game template designed for AI-assisted development. It has a deterministic, headless TypeScript simulation core (seeded RNG, fixed timestep) fully separated from rendering, a command API shared by the UI and by test bots, and a Vitest harness where bot playthroughs prove the game is winnable and Monte-Carlo balance tests keep the numbers honest. Clone it, read its README and CLAUDE.md, and follow the workflow they describe.

Step 1 — Find the game: ask me up to three questions about what I enjoy, then pitch three concepts sized for a solo dev (one core loop, one screen, a week of evenings). We will pick one together.
Step 2 — Build it the template way: design doc with tuning targets first, then the sim (state, systems, commands), then tests proving a bot can reach every ending, then rendering last.
Step 3 — Ship it: when npm run check is green, deploy the static build for free (a GitHub Pages workflow is included).

The buttons prefill the prompt. An agentic coding tool (Claude Code, Codex, Cursor) will take it furthest — it can clone the repo and run the tests — but chat works: paste the prompt, dream up the game, and bring the plan to your editor.