1 paper
Fangxin Shang, Yehui Yang
External skills can improve action-oriented LLM agents without changing model weights, but persistent skill updates are risky when they are distilled from sparse or noisy trajector…