Show HN: Castellan – A real-time strategy game for agents

Category: ai-ml

Tags: agent-benchmark, rts-game, mcp, typescript, ai-evaluation

Score: 5.3/10 (Innovation: 6, Technical: 6, Documentation: 4, Utility: 4)

Castellan is a real-time strategy game environment designed to benchmark and evaluate AI agents, where units are controlled by JavaScript scripts uploaded per unit type and agents connect via MCP or CLI tools. It's an interesting intersection of game design and agent evaluation, offering replays and multi-player matches to test decision-making under time pressure.

Target audience: AI researchers and agent developers

Repository: https://rts.emersoftware.cl/

View on Hacker News