[pokpokpok]

Happy to answer any questions! I think one of the most interesting elements here is the way that the grounding a game environment allows agents to ratchet their engineering progress and run more autonomously than you might be able to for normal engineering tasks.

[pagwin]

The demo gif uses Claude Code but looking at the readme it seems like the idea is for it to be a good environment for various machine/reinforcement learning type tasks.

If that's the case what led to the inspiration to use Runescape and are there any notable non-LLM machine/reinforcement models you think might have an interesting time with this?

[pokpokpok]

I am super curious about using and fine-tuning smaller vision-language-action style models! There are also some interesting RL projects out there focused only on PvP: https://github.com/Naton1/osrs-pvp-reinforcement-learning