Beyond Static Evaluation: Co-Evolutionary Mechanisms for LLM-Driven Strategy Evolution in Adversarial Games

The paper introduces FAMOU, a framework designed for LLM-driven strategy evolution in adversarial multi-agent games, addressing the challenge of shifting evaluation landscapes through three mechanisms: evaluator co-evolution, hierarchical deep evaluation, and weakness pressure. FAMOU demonstrated superior performance on the MCTF 2026 3v3 maritime capture-the-flag task, achieving a top score of 0.526 and a 61.7% win rate against unseen opponents, while also producing novel tactical structures through LLM mutation. This work is significant for practitioners as it enhances the reliability of strategy evolution in dynamic environments and showcases the potential for LLMs to generate innovative algorithms in competitive settings.