tower-mind-agent-benchmark

Evaluate LLM agent capabilities using tower defense game environment with multimodal observations (pixel, text, structured state). Benchmark reveals critical agent limitations: inadequate planning validation, inflexible decision-making, and inefficient action use. Demonstrates significant performance gap between current LLMs and human experts, providing str…

Independent GitHub snapshot · Automated classification · Original documentation governs