Cube Tag Lab

Cinevva's analyzer read this game's design doc and its source, then wrote the brief below. "Make your own" hands it to the game creator as your opening message, so you start from a real spec rather than a one-line idea.

Create a browser-based 3D game called “Cube Tag Lab” using Three.js-style rendering. The game is an arena-building and autonomous chase simulation, not a direct player-controlled action game.

Scene and camera: Show a dark navy/black background with light fog, a flat circular arena about 36 units across, a dark blue ground disk, a fine square grid, and a glowing translucent blue ring around the edge. Use soft shadows, a bright hemisphere light, a directional key light, and a cool fill light. Start with an elevated oblique orbit camera looking toward the center; allow the camera to orbit by dragging and zoom with the mouse wheel/pinch when not playing. During Play, disable orbit controls and keep the simulation view centered on the arena. The visual style is polished, toy-like, colorful low-poly 3D with a dark sci-fi lab presentation and translucent glass-like UI panels.

Arena objects: Seed the arena at startup with eight randomly positioned objects within the circular boundary. Cycle through three types: cyan rectangular blocks (box geometry, roughly 1.6–3.2 units wide, 1.6 high), yellow cylinders (about 1.7 high), and lime-green spherical bumpers. Objects cast and receive shadows. Treat all obstacles as circular collision volumes on the XZ plane. Blocks may be rotated. Agents are kept inside the arena by clamping/wrapping them back inside the circular boundary. Include an Edit mode where the user can add more objects or delete them, including while setting up another round.

Agents and chase rules: Place two 1.2-unit cubes at opposite sides of the arena, initially around x=-6 and x=+6. The Red cube is the runner and the Blue cube is the chaser/IT. Give each cube a colored body and a friendly AI face texture visible on all sides: pale face area, dark round eyes, curved smiling mouth, and subtle colored cheek accents. On Play, both cubes move autonomously across the XZ plane. Blue steers toward Red using a seek weight plus wandering and obstacle avoidance; Red steers away from Blue using flee plus wandering and obstacle avoidance. Use acceleration toward a maximum speed of about 6, velocity damping, obstacle collision sliding, and extra outward impulse when a spherical bumper is contacted. If an agent is stuck for a short time, inject a random movement burst. Rotate each cube to face its movement direction.

A round starts when Play is pressed and resets both agents to their starting positions. Blue wins when it comes close enough to touch Red; Red loses and the Red cube disappears in a burst of approximately 44 small red emissive cube particles with gravity, rotation, and fade-out. The round also has a 60-second timer in the actual simulation: if time reaches 60 seconds without a catch, Red wins and Blue explodes in blue particles. At the end, show temporary floating labels attached to the agents: the winner gets “Good job!”, and the loser gets either “Wrong — you got caught.” or “Wrong — time ran out.” Then return to the editing/idle state after the short effect. Include a Stop button during play to end the simulation and restore the cubes without particles.

Learning behavior: Store visible steering weights independently for Red and Blue. Each has seek, flee, wander, and avoid values, initially around 0.6, 0.6, 1.0, and 0.8. After a Blue catch, increase Blue seek and avoid, reduce Blue wander, and increase Red flee and avoid while reducing Red wander. After Red survives the timeout, increase Red flee and avoid while reducing Red wander, and increase Blue avoid, seek, and slightly wander. Clamp these values to sensible ranges. Use the current weights in the steering behavior so replaying rounds on different layouts can produce different behavior over time.

Editing controls and UI: The fixed top bar should show the title “Cube Tag Lab” and buttons. In the normal idle state show Edit and a blue Play button. While playing, hide Edit and Play and show a red Stop button. Clicking Edit toggles the editor overlay controls: Reset, Clear, and HUD. Clear removes all placed obstacles. Reset returns agents to their start positions and resets the round state. HUD toggles a dark translucent panel, preferably near the lower left, showing elapsed time out of 60 seconds, distance between agents, and two compact cards: Red (runner) and Blue (chaser), with their seek/flee/wander/avoid values color coded red and blue.

Desktop editing: When Edit is open, use keyboard shortcuts 1 for block, 2 for cylinder, 3 for bumper, R to rotate the block preview in 22.5-degree increments, and mouse wheel to change the selected object preview size between about 1.0 and 5.5. Left click places the selected object on the ground at the raycast cursor location; right click deletes the clicked placed object. Show a translucent preview mesh following the cursor while in placement mode. When Edit is closed, mouse dragging is for orbiting the camera rather than placing.

Touch editing: On coarse-pointer/mobile devices, show a bottom translucent build bar only while Edit is open. Include Cam, Place, and Del modes; Block, Cyl, and Bump type chips; minus and plus size controls with a numeric size label; and a Rotate button. Camera mode enables orbiting, Place mode enables tapping the ground to add the selected object, and Delete mode enables tapping an object to remove it. Use large rounded buttons suitable for touch.

UI styling: Use a responsive full-screen layout with no page scroll, white/light-blue bold typography, rounded dark translucent controls with blur, a bright blue primary button, and a red danger button. Keep the Three.js canvas full screen behind the UI. Avoid inventing weapons, scoring, lives, sound, powerups, or direct avatar controls; the central interaction is designing obstacle layouts and watching the two face-cubes learn and chase.