brycewang-stanford/auto-empirical-research-skills/ai-security-papers-guideskill versions