WOW! Might also be worth trying the exact task with Kimi K3 to see how much better/worse it is. It was going better than GPT 5.6 Sol Ultra on many large-scale design/SWE tasks.
Yes, that would be very interesting. Depending on the prompt, the result will probably be something completely different, but it would be interesting to see the level of gameplay and immersion.