🎯9 Game EnvironmentsCovering cooperation, competition, communication, and deception paradigms for comprehensive LLM evaluation
🧩Modular PromptsEach game's prompt decomposed into task, physics, action format, and navigation modules
🔄Batch Evaluationbenchmark_runner.py supports N episodes × multiple seeds × cross-environment aggregation
🌐Zero-Shot ReasoningNo training needed — LLMs directly generate continuous actions from text descriptions