RECENT Framework Grounds Embodied Agent Skills Using Small Language Models

RECENT decouples skill semantics from environment-specific execution bindings by representing skills as code and using localized refactoring to adapt them via small language models, avoiding LLM dependency at inference time. Targets embodied agents in dynamic, partially observable environments where LLM calls are impractical.