NVIDIA's Skill2Env Turns Agent Skills into RL Environments
Rubric-augmented rewards trade raw benchmark score for trajectories a judge rates far higher, a hint that outcome-only RL over-optimises for narrow metrics while behavioural rubrics shape more genuinely skilled agents. Teams training terminal agents can reuse the open pipeline to convert existing Agent Skills corpora into RL environments instead of hand-crafting tasks. The US$90,000 generation cost also sets a concrete budget baseline for anyone planning similar corpus builds.
NVIDIA's Skill2Env pipeline converts public Agent Skills into executable reinforcement learning environments by compiling each SKILL.
md bundle into terminal tasks.A Codex planner splits each Skill into workflows, and a Codex creator builds tasks with programmatic tests and a behavioral rubric drawn from the Skill's quality criteria.
From about 3,400 crawled Skills, the pipeline produced 7,971 tasks across 13 domains, with software engineering under a quarter of the corpus;
generation with GPT-5.6 Sol cost over US$90,000 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, a hand-verified held-out benchmark.Adding the rubric to the reward yielded 50.
1% on Terminal-Bench 2.1, but a judge preferred the rubric-trained model's trajectories over the base model's on 73.
0% of tasks.
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
This report is categorized under AITOP's "Papers" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include NVIDIA, Skill2Env, RL environments, Qwen3.8-27B.
Source attribution: curated from DAIR.AI (@dair_ai). Cross-source corroboration: 9 independent sources have covered this development (including shao__meng, NVIDIA AI, arXiv cs.LG), confirming industry-wide signal strength.
