brycewang-stanford/auto-empirical-research-skills/ai-model-benchmarkingskill versions