From Slop to Skill: An Agent Proficiency Engine Using RLM | Dark Hacker News