Breaking the Environment Wall: A Unified Framework for Preparing and Evolving Agent-Native Environments

Many real-world tasks (e.g., office workflows, scientific experimentation) require LLM agents to interact repeatedly with their environments for context-dependent operations. However, such environments are often not agent-ready. First, information is often scattered and fragmented across the environ