Skill2Env: Capability-Oriented Environment Synthesis from Skills for General Agents
Weiyi Xu, Xiaowen Yang, Wen Da, Hang Xu, Canwei Li, Hongjie You, Pusen Dong, Yucheng Zeng and 2 others
AllSpark Team
Published September 27, 2026 · on the feed September 29, 2026 · ▲ 35 · arXiv 2609.33772
The abstract
Executable environments are critical for post-training agents on tasks that require tool use and multi-step interaction, but constructing executable tasks together with their environments remains difficult to scale. Skills provide reusable domain knowledge, operational procedures, and tool-use instructions, but a substantial gap remains between the information contained in a skill and a concrete, challenging task with a complete executable environment. To address this gap, we introduce Skill2Env, a capability-oriented framework that starts from a skill and uses agent capability demands to guide task and environment synthesis. Skill2Env represents these demands through reusable difficulty patterns and instantiates them into task blueprints that specify objectives, challenges, environment facts, information boundaries, and acceptance criteria. These blueprints guide the joint construction of task instructions, execution substrates, workspaces, and rubric-based evaluators around source skills. We further propose Iterative Task Hardening, which uses solver execution evidence to identify insufficiently challenging task designs, strengthen or extend their difficulty-pattern instantiations, and revise the corresponding blueprints and environments. Using 1.5K high-scoring trajectories generated from Skill2Env environments for supervised fine-tuning, we observe consistent improvements across a broad range of agent benchmarks, demonstrating the effectiveness of capability-oriented environment synthesis for agent post-training.
The Brief
2:19A short podcast summary of the paper. Written and voiced by AI.
Transcript
Skill2Env. Capability-Oriented Environment Synthesis from Skills for General Agents. By Weiyi Xu and nine others, from the AllSpark Team. Published on the twenty-seventh of September. No code repository is linked to it.
Post-training agents for tool use and multi-step interaction needs executable environments, and building executable tasks together with their environments is hard to do at scale. Skills look like a promising raw material. A skill packages reusable domain knowledge, operating procedures, and instructions for using tools. But there is a wide gap between what a skill contains and a concrete, challenging task that comes with a complete environment able to run it.
Skill2Env bridges that gap by asking a different first question. Rather than what a skill describes, it asks what capabilities an agent would need to exercise it, and lets those demands steer the synthesis of both the task and its environment. The framework is capability-oriented from the start.
Capability demands are represented as reusable difficulty patterns. These patterns are instantiated into task blueprints, and each blueprint spells out objectives, challenges, environment facts, information boundaries, and acceptance criteria. The blueprints then guide the joint construction of everything around a source skill: the task instructions, the execution substrate, the workspace, and an evaluator based on rubrics.
A second mechanism, Iterative Task Hardening, keeps the tasks from being too easy. It looks at evidence from solvers actually executing the tasks, picks out designs that are not challenging enough, strengthens or extends how their difficulty patterns are instantiated, and then revises the matching blueprints and environments.
The evidence comes from training. The authors take one point five thousand high-scoring trajectories generated in Skill2Env environments and use them for supervised fine-tuning. The result is consistent improvement across a broad range of agent benchmarks. The abstract does not give the individual scores, only that the gains hold across that range.
Think of a skill as a manual and an environment as a workplace. A manual on its own teaches nothing until someone is handed a job that forces them to use it, and a job that is too easy teaches little more. Skill2Env tries to manufacture those jobs, starting from the capability a skill calls for and then hardening the result against a real solver until it bites.
Paper Quest · 7 chapters · about 37 min
Play this paper
Learn it by doing: make the calls the authors made, guess the result before you see it, and defend it in a viva graded on your own words. Chapter 1 is right here.
Chapter 1 🧩 · The problem
Training agents needs worlds to act in. Here, the raw material is plentiful but shapeless.
You are post-training an agent for tool use and multi-step work. It needs executable tasks: instructions, usable tools, a workspace full of files and state, and a way to grade the result.
The road ahead
- 1 · The problem ~6 min Agents learn tool use from executable tasks, and turning a skill's know-how into a hard, runnable task is the step that does not scale.
- 2 · What came before ~3 min
- 3 · The key idea ~9 min Design each task around the abilities an agent must use, turn those into challenge types, write a plan, build from it, and harden easy tasks.
- 4 · Choose your own research ~3 min
- 5 · Build the method ~8 min A curated skill becomes demands, then patterns, then a blueprint, then a built environment, which is hardened and finally used for fine-tuning.
- 6 · What they found ~3 min
- 7 · The viva ~5 min Three questions from the examiners, answered in your own words. Win it for the paper’s trophy.
Free, with Google or an email address. Your stars and progress are kept.