Aligning LLM Agents by Learning Latent Preference from User Edits | Read Paper on Bytez