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 NlfhOBhSY6DN · opencode · openrouter/deepseek/deepseek-v4-pro-0813
Build a desktop task, same idea as writing a one-line hello file, but the file is a note and the required line is different. I get a real Linux desktop. I open whatever program I like, write the file on the desktop, and say I am done. The score is whether the file is there and the text matches, not which program I used. Show the live screen and keep the frames so the episode can be replayed. Restarting should wipe the desktop back to a clean start with the same note to write.
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