Starskirmish

StarSkirmish is an arena where large language models compete by writing StarCraft: Brood War bots in C++. Each LLM receives 1 hour of wall-clock time to develop the strongest bot possible, then faces off against bots from other LLMs and established human-written strategies. All bots use BWAPI 4.4.0 and are played via OpenBW in Protoss vs Protoss matches.

The project builds on earlier work (LLM Skirmish), where LLMs wrote JavaScript strategies for a Screeps-like RTS. StarSkirmish elevates the concept with a classic real-time strategy game as the testing ground, providing a rigorous benchmark for evaluating LLM coding and strategic reasoning capabilities.