OpenCode (DeepSeek V4 Pro via OpenRouter) in a Vercel Sandbox. Starter only (AGENTS.md, serve.py, player/, helpers/dom/view_kit). No hardcoded demo code. Cursor did not edit workspace/.
2026-10-15 · task frvQ6ISheY0I · opencode · openrouter/deepseek/deepseek-v4-pro-0813
Build a fake office desktop I can click through. On the dock I can open a browser, an email app, a spreadsheet, and a chat app. Each round gives me one ordinary job, like sending a message, filling a cell, or looking something up and writing it down. What matters is the state of those apps when I say I am done. If the email was sent, or the cell has the right value, I score, even if I clicked around first. Extra clicking should cost a little, so wandering is worse than a short path. Clicking something that is not actually on screen should do nothing. Show the desktop, the windows, and the score. A click is how I play. Different rounds are different jobs. The same job should start from a clean desk every time.
Running the prompt-created workspace on this machine — not a mock. Save an episode, then open Replay below to scrub the timeline.
opening · 0 ms
Exact files the agent wrote — not rewritten by hand.
# AI : shared Action contract — the page posts {"move": "..."}; keep this shape.
from pydantic import BaseModel
class Action(BaseModel):
move: str