How did you use AI?
Two jobs. Claude (Anthropic) wrote the code under my direction. FLUX.2 klein 4B, run on Cloudflare Workers AI, drew every dog. The art is 432 sprites. Each of the 12 pieces is one dog, kept consistent by feeding the model a reference image of that dog’s master sprite. The gallop frames were drawn over pose shadows taken from Muybridge’s 1887 “Dog Maggie galloping,” which is public domain. Everything was generated on a flat grey ground so the white fur survives when the background is keyed out. Scripts check every frame before it ships: team collar color, silhouette, which way the dog faces, and whether the hat floats off its head. The facing check alone caught 20 dogs drawn backwards. The AI in the gameplay is the dogs. A Great Pyrenees understands the command and decides for itself. Every legal move gets odds: 75% the dog does what it wants, 15% its second choice, and 10% a squirrel, which is any other legal move. What it wants comes from instinct: guard, chase what’s worth catching, cover or run when threatened, and the King hides. The card shows you the real numbers before you send it. The CPU uses expectimax search, because it has to plan around dogs that won’t listen to it either. Full chess rules sit underneath, refereed by chess.js and checked against it on 49,560 positions.

List of Team Members (name or pseudonyms)
M Dawg. Code written with Claude, art generated with FLUX.2 klein.

How does your game fit the restriction?
It plays in the phone browser straight from the itch page. Touch only, no install. Built portrait-first at 390x844 and checked in landscape. The dogs are drawn at 1.4 times the square height so they read on a small screen, and the shed fur was made bigger after it disappeared at phone size.