Auto-Empirical Research Skills (AERS)

🌐 Language / 语言: English | 中文

Stanford REAP × CoPaper.AI · An academic-industrial AI toolkit for empirical research Crafted by Stanford's empirical methodology team — covering the full pipeline from data cleaning to top-journal submission

The Definitive Collection of AI Agent Skills for Empirical Research — 119 GitHub Repos / 23,000+ Skills

A curated, opinionated list of 119 GitHub repositories and 23,000+ AI Agent Skills for empirical research in economics, political science, sociology, psychology, public health, education, management, finance, and public policy — organized by research workflow, from topic selection to journal submission.

In 2026, the way we do empirical research is being redefined.

CoPaper.AI — an empirical research AI assistant incubated by researchers at Stanford REAP / SCCEI (Stanford Center on China's Economy and Institutions) — can complete a publication-quality empirical paper in 20 minutes: from data import, descriptive statistics, causal inference models, and robustness checks to formatted result tables, all in one go. The secret isn't a more powerful model — it's Skills: encoding senior researchers' methodological expertise into structured workflows, so the AI knows "what a complete DID analysis should include" instead of waiting for you to remind it step by step.

This repository is the Agent Skills landscape we compiled while building CoPaper.AI. We organized hundreds of Skills repos and tens of thousands of Skills scattered across GitHub, communities, and academia by research workflow stages, so you can pick what you need.

🎓 Three Layers of Trust · Why It's Us Building This

| Layer | Anchor | Lever | |---|---|---| | 🏛️ Academic lineage | Stanford REAP / SCCEI — Stanford Center on China's Economy and Institutions | A research center with a sustained publication record in empirical economics methodology and a deep tradition in applied causal inference | | 🔧 Engineering delivery | CoPaper.AI empirical research AI assistant | Ships with 20 econometric methodology Skills (DID/IV/RDD/PSM/DML, etc.), Supervisor + 4 sub-agent multi-agent architecture, one-sentence triggers, automatic result output | | ⚙️ Open-source engine | StatsPAI — the causal-inference engine that powers CoPaper.AI | 900+ functions · one import statspai as sp · JOSS in submission · MIT-licensed. Every DID/IV/RD/SCM estimate CoPaper.AI produces is driven by StatsPAI; this Skills collection is itself part of the StatsPAI ecosystem |

🔒 Use with confidence: every one of the 52 Skills / 2,940+ files in this repo passed our systematic security audit — 52/52 CLEAN, zero FLAGGED, zero exfiltration, zero reverse shells, zero prompt injection.

💡 Want it out of the box? Skip the Skills assembly — try → copaper.ai and let the Stanford methodology team run the empirical pipeline end-to-end for you.

🆕 Changelog

• 📕 skills/50-brycewang-aer-skills: This repo's sister project brycewang-stanford/AER-skills vendored in whole, with the same StatsPAI-style sync loop (scripts/sync-aer-skills.sh + .github/workflows/sync-aer-skills.yml — Monday 06:00 UTC weekly diff, PR on drift). Positioning: a top-5 economics submission skill stack (AER / AER:Insights / AEJ family), extending the StatsPAI / 00.x "analysis" line to the "manuscript + submission" line.
  • 🧱 Nine skills covering the full submission pipeline: aer-topic-selection (AER vs Insights vs AEJ routing) → aer-identification (identification audit: modern DiD / weak IV / boundary RDD pitfalls) → aer-robustness (referee-anticipating robustness matrix) → aer-introduction (Keith Head five-paragraph intro) → aer-tables-figures (AER booktabs typesetting) → aer-replication (AEA Data and Code Availability Policy package, openICPSR-ready) → aer-submission (preflight: 100-word abstract, disclosure, cover letter) → aer-rebuttal (R&R letters written against the revised manuscript, not the old draft) → aer-workflow (orchestrator that tells you which skill to use next).
  • 🆚 Differentiation from existing skills: StatsPAI / 00.x solve "how to run the analysis correctly"; AER-skills solves "how to write the paper to top-5 acceptance threshold" — the AER 100-word abstract / AER:Insights 7000-word limit / 45% desk-rejection rate / AEA mandatory replication are top-5-specific constraints that generic scientific-writing skills (Nature-Paper-Skills etc.) do not cover. Identification-first: if your design is fragile, no prose will save it.
  • 🔁 Vendor-sync loop: git clone --depth=1 upstream → rsync -a --delete --exclude='.git' mirror the whole tree → diff content hashes before/after, exit 0 on no drift, exit 1 on drift to trigger peter-evans/create-pull-request@v6 on chore/sync-aer-skills branch. Supports manual workflow_dispatch for on-demand sync.
  • License: MIT — consistent with StatsPAI / 00.x; commercial and academic use both allowed.
  • First upstream commit: 7e9c44d (2026-05-25, includes modern-aer-exemplars.md with 30+ subfield-organized papers).

• 🛡️ SECURITY-SCAN-REPORT.md: We ran a systematic security audit across all 52 Skills / 2,940+ files in this repository. No malicious prompts, viruses, trojans, reverse shells, or other malicious content were found. Bottom line: every Skill in this repo is safe to use.
  • 🔍 Six-phase defense-in-depth methodology: (1) automated grep across 13 risk categories (pipe-to-shell, reverse shell, credential exfil, decode-and-run, mining/RAT signatures, prompt injection, etc.) → (2) 100% manual review of all 6 hook-bearing Skills and their 40+ hook scripts → (3) three parallel agents auditing SKILL.md prose, agent definitions, and reference docs separately → (4) supplemental integrity checks (hidden Unicode, encoding anomalies, ultra-long lines, HTML injection, network-related imports).
  • 📊 Result distribution: every "sensitive" hit verified as one of three legitimate categories — defensive security rules (deny rules, bash-safety hooks, credential detectors), legitimate academic API calls (arXiv / CrossRef / PubMed / FRED / World Bank / OECD / BLS), or standard Claude Code workflow hooks (scaffolding / state save / context monitor — all local file operations, zero network IO).
  • 🔑 Key insight: 17-DAAF is actually the strongest "security-aware" reference template in this batch (14 defensive hooks + 32 deny rules + active credential scanning). Largest size ≠ highest risk.
  • 📈 Visual infographics: 5 zhihu-style information graphics embedded in the report ([overview