Cell Wars

Cell Wars is a turn-based territory game that proves strategy does not need a hundred-hour tutorial. The board is a 7x7 grid of cells, and you control one color while an AI opponent controls the other. On your turn you pick one of your cells and either clone it to an adjacent empty square — keeping the original — or jump it two squares away, sacrificing the source. The heart of the game is conversion: every enemy cell adjacent to your landing spot flips to your color. The board fills up move by move, and when it is full, whoever owns more cells wins. It is a complete, genuinely deep abstract strategy game with rules you can fully understand in ninety seconds.

The depth comes from a constant tension that every Ataxx-style game carries: cloning grows your numbers but puts weak cells everywhere; jumping keeps you compact but cedes territory. Add moving blocker cells that neither side can capture, and every turn becomes a genuine puzzle of geometry and timing. Games run five to ten minutes — long enough to matter, short enough for a break — and the built-in difficulty selector means the game grows with you. Easy mode makes almost random moves and is beatable by anyone; Normal plays a real positional game; Hard looks ahead at your best reply before committing, and it punishes greed without mercy.

The interface is clean on purpose: tap or click a highlighted cell to see your options, confirm to move, and captured cells flip with a short animation so you can always read what just happened. A score bar runs along the top showing the territory balance in real time, which makes comebacks visible and defeats educational. On desktop it plays by mouse; on phones everything is thumb-sized taps.

As an AI-generated game, Cell Wars is our most interesting experiment, because the hard part was not the rules — it was the opponent. Gemini implemented the ruleset cleanly, but the first AI player was a pushover. Our strategy-ai-opponent prompt made the difference: it forced explicit scoring weights (captures over mobility over safety) and a one-ply lookahead for Hard mode. The result is an AI that punishes mistakes without feeling unfair, and a prompt in our library that any strategy game idea can reuse.