AutoEnvScaling: self-play system where one model builds terminal agent training tasks

Dongxi NLP (东锡) announces AutoEnvScaling, a system in which a proposer agent builds complete Harbor tasks and sandbox environments for terminal agents.

According to the post, one model acts as both builder and solver in self-play reinforcement learning, so the model improves without humans in the loop.

The post does not report benchmark results or other measured outcomes.