Auto Empirical Research Skills

Auto Empirical Research Skills (AERS) is a large curated catalog — maintained by brycewang-stanford under the Stanford REAP (Center on China’s Economy and Institutions) banner and paired with the CoPaper.AI product — that describes itself, as of July 2026, as drawing on a pool of 23,000+ agent skills, vendored and indexed down to 69 collections holding 1,150 individually catalogued skills, for empirical research across eight social-science disciplines: economics, political science, psychology, sociology, international relations, public administration, education, and communication. Rather than a single hand-written skill, AERS ships a router SKILL.md that classifies an incoming task by research stage — causal inference and econometrics, Stata/R/Python analysis, literature review, citation checking, manuscript writing, de-AIGC editing, or replication — and loads only the one matching child skill, explicitly instructing the agent never to read the whole catalog at once.

The routing table is organized by identification strategy: difference-in-differences and staggered event studies, instrumental variables, regression discontinuity, synthetic control, panel fixed effects, double machine learning/CATE/causal forests, and Bayesian modeling each point to a specific vendored collection (for example `skills/50-brycewang-aer-skills/` for AER-level DiD/IV/RDD work, or `skills/40-py-econometrics-pyfixest/` for panel fixed effects in Python). Separate collections cover full-language empirical-analysis pipelines in Python, Stata, and R (the `00.x-Full-empirical-analysis-skill_*` flagship skills), plus dedicated skills for literature review, citation checking (`skills/62-PHY041-claude-skill-citation-checker/`), manuscript proofreading, and Chinese-language de-AIGC rewriting (`skills/48-copaper-ai-chinese-de-aigc/`).

AERS also foregrounds a "trust surface": the README reports 17 numeric benchmark tasks with gold values recomputed from real data on every run, 37 of 183 behavioral eval-harness rubric items scored, and a security audit badge reporting 52/52 checks clean. Its flagship pipeline advertises 20 built-in methodology skills and a self-built StatsPAI library (900+ MIT-licensed statistical functions) aimed at producing a reproducible empirical paper in around 20 minutes, positioning AERS as an amplifier for the time-consuming mechanics of empirical work — data cleaning, regression tables, citation formatting — while leaving research judgment to the researcher.