Even with his context-tracking mechanism, the gameplay failures sound like running out of context in the late game, especially the frequent failures of the "check for opponent win conditions every 20 moves." Wondering how much info about the game win state gets captured in the game digests, and how much he could improve the gameplay even with the MCP limitations by focusing there.

I also noticed they where not using XML for game state output, from what I understand most LLMs still benefit from having outputs like this put into XML tags