EnvHarness: Awakening Static Worlds for Agent Learning
LLM agents learn by interacting with environments, yet these environments are hand-built and static: blind to an agent’s weaknesses, and quickly left behind as it improves. While recent environment generation methods attempt to address this, they require domain-specific pipelines, rely on expensive or unreliable verifiers, and still produce static environments. To alleviate the engineering burden of rebuilding environments from scratch, we propose Environment Harness (EnvHarness), a programmable layer of plug-in components that wraps a static environment to reshape its behavior without modifying the underlying logic. Operating through standard interfaces, EnvHarness applies across diverse domains while ensuring every reshaped environment retains its original verifier. To automate this process, we introduce EnvRigger, which treats the target policy as a black box, observing its execution trajectories to synthesize EnvHarness components targeting diagnosed flaws, and validating them via fresh rollouts. Across five benchmarks in four domains, EnvHarness outperforms both original environments and domain-specific environment generation pipelines, achieving up to a 9.0-point improvement on held-out instances with 9.8% fewer execution steps.
Introduction. As LLMs are deployed as autonomous agents, the source of learning shifts from curated text data to interactive environments. Whether navigating web pages (Gur et al., 2024), resolving issues in a codebase (Yang et al., 2024), or controlling an embodied platform (Wang et al., 2023), agents rely on their respective environments to acquire learning signals (Yao et al., 2022b). These environments act as interactive counterparts that present specific tasks, manage changing states, respond to actions, and evaluate success (Li et al., 2026). Unfortunately, building them requires substantial human effort to hardcode the interaction logic and verifiers (Merrill et al., 2026; Zhang et al., 2025). Consequently, the resulting environments remain rigidly static, behaving identically regardless of which agent interacts with them or how much that agent has improved (Hu et al., 2026).
Discussion / Conclusion. We introduce EnvHarness, a programmable layer that turns a static, existing environment into a controllable one. EnvHarness wraps a frozen benchmark with three plug-in components, Stage, Contract, and Chain, and reshapes it entirely through the standard reset/step interface, making it possible to isolate a skill, extend a task’s horizon, or calibrate difficulty in environments that were never built for any of these purposes. Since EnvHarness never touches internal code, a single implementation works seamlessly across different domains. Furthermore, by leaving the original tasks unchanged, every reshaped environment safely retains its trusted, human-built verifier. To fully automate this customization, we introduce EnvRigger, an autonomous loop that diagnoses policy weaknesses from execution trajectories and synthesizes targeted EnvHarness components to provide precise learning signals. This reframes environment construction as a wrapping problem rather than an authoring one, and suggests a practical pathway toward scalable environment supply for agent learning.
Lines of inquiry this paper opens 24
Research framings built by reading the notes related to this paper — the questions it feeds into.
How do evaluation biases undermine LLM quality assessment systems? How can LLM user simulators model realistic goal-driven conversation? How should we design LLM systems to maintain alignment and control? Does self-reflection enable models to reliably correct their errors? What articulatory information do speech signals carry that text cannot? How do knowledge injection methods compare across cost and effectiveness? How faithfully do LLMs reflect their actual reasoning in outputs and explanations? Why do language models reinforce false assumptions instead of correcting them? Do accurate-looking LLM outputs hide structural failures in learning and reasoning? How can models identify insufficient information and respond appropriately without guessing? How do language models inherit human biases from training data? What capability tradeoffs emerge when scaling model reasoning abilities? How should inference compute be adaptively allocated based on prompt difficulty? What pretraining choices and baseline capability constrain reinforcement learning gains? Why do continual learning scenarios trigger catastrophic forgetting and interference? Why does finetuning cause catastrophic forgetting of model capabilities? How do training priors constrain what context information can override? How do language models establish social grounding in human dialogue? Can AI-generated outputs constitute genuine knowledge or valid claims? How do multi-agent systems achieve genuine cooperation and reasoning? Can prompting strategies overcome LLM biases without model fine-tuning? What causes silent corruption to amplify through delegated workflows? Is embodied interaction necessary for language meaning and genuine agency?