I'm skeptical. When I tried "word games" with arbitrary rules (such as not mentioning specific words) the LLM often goes off script and breaks the rules. An LLM that is "too helpful" effectively ruins the game, and prompting cannot be relied on to enforce rules.