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 cz5mz7gyouUn · opencode · openrouter/deepseek/deepseek-v4-pro-0813
Build a skin-check clinic. Each round is one person standing in a camera booth. I get a body I can turn, and spots I can zoom into until I see skin, pores, and the mark itself, with a ruler in millimetres. I also get the previous scan of that same mark beside the new one. I fill in five observations: whether the two halves match, whether the edge is smooth or broken, whether the colour is one shade or several, whether it is bigger than 6 mm, and whether it changed or bled. Then I decide to leave it, watch it, or take it off. Saying I am not sure should be safer than a confident mistake. A pretty close-up of the wrong spot should not count. A view that shows me the inside of the skin should not be allowed to raise the score. Several kinds of mark, including one that is small but dangerous and one that is large but harmless. The same round should come back the same if I restart it.
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