NVIDIA 发布 Skill2Env：将公开 Agent Skills 编译为 RL 环境生成近 8000 条训练任务
NVIDIA 把公开 Agent Skills 自动转成 RL 训练环境，产出近 8000 条任务，还开源了流水线思路，做 Agent 训练的可以看看。
NVIDIA 的 Skill2Env 流水线把公开的 Agent Skills 转成可执行终端任务：Codex planner 读取 SKILL.md 并拆分工作流，Codex creator 为每个任务构建程序化测试和行为 rubric。从约 3400 个爬取的 Skills 中产出 7971 条任务，覆盖 13 个领域，软件工程占比不到四分之一，用 GPT-5.6 Sol 生成的 API 花费超过 9 万美元。Qwen3.8-27B 经过 300 步结果导向 RL 训练后，Terminal-Bench 2.1 成绩从 49.4% 提升到 54.1%，S2EBench pass@1 从 33.4% 提升到 37.7%。加入 rubric 奖励在基准上增益较小（Terminal-Bench 2.1 为 50.1%），但人类偏好的模型轨迹占比从 54.5% 升至 73.0%。
Exciting work from NVIDIA.
(bookmark it)
Interesting to see this approach to turn public Agent Skills into RL environments.
Lots of excitement around RL environments so this is a great read.
Skill2Env compiles each Skill into executable terminal tasks.
A Codex planner reads the SKILL.md bundle, researches related public assets and splits the Skill into workflows.
A Codex creator then builds each task with programmatic tests and a behavioral rubric taken from the Skill's own quality criteria.
From about 3.4k crawled Skills, the pipeline produced 7,971 tasks across 13 domains, with software engineering under a quarter of the corpus. Generating them with GPT-5.6 Sol cost over $90k in API usage.
After 300 steps of outcome-only RL, Qwen3.8-27B improved from 49.4% to 54.1% on Terminal-Bench 2.1 and from 33.4% to 37.7% pass@1 on S2EBench, their hand-verified held-out benchmark.
Adding the rubric to the reward gave smaller benchmark gains, 50.1% on Terminal-Bench 2.1. Given the source SKILL.md, a judge preferred the rubric-trained model's trajectories over the base model's on 73.0% of tasks, against 54.5% for the outcome-only model.
Paper: https://t.co/DYm5hevsHd
Chat with Paper: https://t.co/WSSNWE3erY
