oneshotlm
Compare all models on this prompt →

Snake · qwen3.5-plus-20260420

Build a playable Snake game on a canvas with arrow-key controls, a growing tail, food, score, and a game-over and restart state.

⛨ sandboxed Open in new tab ↗
transcript · 4 messages
user

Build a playable Snake game on a canvas with arrow-key controls, a growing tail, food, score, and a game-over and restart state.

assistant
thinking
The user wants a Snake game built as a single HTML file with embedded CSS and JS. Let me build a complete, polished Snake game with:

1. Canvas-based rendering
2. Arrow key controls
3. Growing tail mechanic
4. Food spawning
5. Score tracking
6. Game over state (wall collision + self collision)
7. Restart functionality

Let me write this all in one file.
write /tmp/ws/index.html (9574 bytes)
toolResult

Successfully wrote 9574 bytes to /tmp/ws/index.html

assistant

The Snake game is ready at `/tmp/ws/index.html`. Here's what's included: **Gameplay:** - Classic snake mechanics on a 30×30 grid - Arrow keys + WASD controls - Food spawns randomly (avoiding the snake body) - Snake grows when eating food, score increases - Game over on wall collision or self-collision **Visual features:** - Dark themed UI with subtle grid lines - Gradient snake body (bright green head → darker tail) - Eyes on the snake head that follow the direction of movement - Glowing red food with radial glow effect - Red flash on game over **UI states:** - **Start screen** — title + instructions + "Start Game" button - **Playing** — score display in the header - **Game Over** — final score shown + "Play Again" button - High score persisted in `localStorage` **Restart:** Click the button, press Space, or press Enter.

TAKE IT TO CODE

A single self-contained HTML file — copy it and open in any browser, or reuse the prompt with your own pi-agent run.

RUN

modelqwen3.5-plus-20260420
vendorQwen
harnesspi-agent
generated2026-08-01

EVALUATION

2/5 partial

Snake game shows start screen and game-over state, but game appears to end immediately without visible gameplay/movement.

Vision grade · anthropic/claude-sonnet-5

MEASUREMENTS

Raw readings from the capture, not a verdict — the grade above is the only judgement. Percentages are how much of the screen changed; the measure under-reports sparse motion on a dark background.

idle motion0.0%
on click0.0%
on drag0.0%
on wheel0.0%
on enter + space3.5%
on arrow keys0.0%
on w/a/s/d0.0%
frame spread13.7 / 255
console errors0
js errors none

METRICS

runtime71.0s
tokens in6.7k
tokens out3.8k
cached0
cost / run$0.0088