Skill2Env: Capability-Oriented Environment Synthesis from Skills for General Agents

The paper presents Skill2Env, a capability-oriented framework that bridges the gap between reusable skills and executable environments for agent training. Skill2Env uses agent capability demands to guide the synthesis of task blueprints, specifying objectives, challenges, and acceptance criteria, paired with Iterative Task Hardening to strengthen under-challenged designs using solver execution evidence. Training on 1.5K high-scoring trajectories generated from these environments yields consistent improvements across multiple agent benchmarks.

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.