XNews

NVIDIA Skill2Env Compiles Community Agent Skills into RL Training Environments

NVIDIA Skill2Env Compiles Community Agent Skills into RL Training Environments

NVIDIA's Skill2Env converts 3,400 public agent skills into 8,000 terminal tasks with programmatic tests and behavioral rubrics. Just 300 steps of RL training on Qwen3.8-27B improved Terminal-Bench 2.1 by 4.7 points. The method uses a four-stage pipeline (Plan, Diversify, Create, Verify) and reveals that rubric-calibrated RL improves behavioral alignment but not benchmark scores.

Source posts