Wave-2 OpenCode (DeepSeek V4 Pro) snapshot after starter/player fixes: human_label helper, AGENTS.md UI rules (spaced labels, view CSS, visible plots, default playable reset), and the examples/sandbox players hide legal_moves buttons when view.js renders. Cursor did not edit the environment workspace.
2026-10-15 · task Nu7SGlP8E4ho · opencode · openrouter/deepseek/deepseek-v4-pro-0813
Build a short diagnostic game, like a medical word puzzle. A patient has one or two ordinary diseases, and their symptoms overlap, so the first impression is not enough. I can order tests. Each test tells me something useful and also costs something. I mark up to two diagnoses from a fixed list, and I can change my mind. When I submit, I am scored on whether I named the right diseases, and the cost of the tests I ordered comes off that score. Ordering every test should be a bad habit. Show the complaint, what each ordered test came back with, the diagnosis list with my marks, and the score only after I submit. Do not show the true diseases before I submit. Offer several patients. Opening the same patient twice should show the same complaint and the same test results.
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