boo System Agent Takes On TUI Invaders

A terminal arcade game is a surprisingly useful computer-use eval because it tests perception, timing, control, and recovery in a tight feedback loop.

• –TUI Invaders is an MIT-licensed Bun game built on OpenTUI, with pixel-art ships, power-ups, bosses, and persistent high scores.
• –The challenge measures interactive terminal control rather than code generation alone.
• –Beating level 1 is a clear milestone, but meaningful comparisons need fixed versions, latency, seeds, and control settings.
• –GPT-5.3-Codex-Spark’s stronger showing highlights how low-latency inference may matter as much as raw reasoning in real-time agent tasks.