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 RAmc-HTPLmK6 · opencode · openrouter/deepseek/deepseek-v4-pro-0813
Build a dental X-ray reporting screen. I see one panoramic film. I drag a box around each dark patch at a root tip, not around the whole tooth, and I grade it 3, 4, or 5. Grade 3 is bone change with some mineral loss. Grade 4 is a well-defined dark area. Grade 5 is severe, with extra features. Grades 1 and 2 are not reported. A film with nothing to report is an empty report. The films in this exercise do have at least one lesion. When I file the report, score me on whether my boxes land on the real lesions and whether the grades match. A box that overlaps a real one well enough counts. Extra boxes should hurt. After filing, show the reference boxes: matched in one colour, misses and extras in another. Show a short fake referral so it feels like a workstation, and make clear it is not a real patient. Several films. The same film should come back the same.
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