AERS: Auto-Empirical Research Skills
23,000+ curated skills across 8 social science disciplines for end-to-end reproducible empirical research.
- Stars
- ★ 4.7k
- Last updated
- 4d ago
- License
- NOASSERTION
- Primary language
- Stata
- FA score
- 51/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesClaude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- Social science researchers needing end-to-end automation from idea to paper.
- Not a fit if
- Users who do not work in Claude Code or Codex-style agents
- Teams that require a clearly stated commercial license
- Source review
- 51/100 · Major gaps
What does this agent do, and when should you use it?
Auto-Empirical Research Skills (AERS) is a curated library of 76 collections containing 1,096 skills for empirical research, maintained by Stanford REAP and CoPaper.AI. It covers 8 social science disciplines and provides a 9-stage end-to-end pipeline from research question refinement, literature review, data acquisition, identification strategy, estimation modeling, robustness auditing, to publication-quality tables/figures, writing, and AI-content reduction. The library features the in-house StatsPAI causal engine (900+ functions), full pipelines in Python/Stata/R, and a bilingual de-AIGC skill, all validated by 17 numeric benchmarks and 37 behavioral evals.
AERS provides 1,096 skills across 76 collections (numbered 00-72). Users invoke skills via natural language; the Agent auto-selects based on description. StatsPAI offers 'sp.causal(...)' for one-line causal inference (DID/RD/IV/SCM/DML). Python/Stata/R stacks wrap libraries like pandas/statsmodels/reghdfe/fixest. The 48 de-AIGC-skills reduces AI-detection rates in academic writing. The 69 Paper-WorkFlow meta-orchestrator chains the 9 stages into an automated pipeline, allowing manual intervention at any step. Installation is via Claude Code plugin marketplace or copying skill folders; catalog/skills.json tracks all skills.
- Social science researchers needing end-to-end automation from idea to paper.
- Economists running panel data causal inference with publication-grade tables.
- Authors submitting to journals requiring low AI-detection scores, using de-AIGC skills.
- Policy analysts quickly gathering evidence via literature review and data skills.
- Interdisciplinary teams leveraging 8 disciplines for varied empirical studies.
How do you install or deploy this agent?
Simplest: provide the repo URL to Claude Code and specify scope (directory/project/global). Manual:
- Plugin marketplace (Claude Code v2.1+):
claude plugin marketplace add brycewang-stanford/Auto-Empirical-Research-Skills
claude plugin install aer-skills@auto-empirical-research-skills- Copy a single skill:
git clone --recurse-submodules https://github.com/brycewang-stanford/Auto-Empirical-Research-Skills.git
cp -R skills/00.1-Full-empirical-analysis-skill_Python .claude/skills/ # project
cp -R skills/00.1-Full-empirical-analysis-skill_Python ~/.claude/skills/ # globalHow do you use this agent?
In a new Claude Code session, describe the task in natural language; the Agent picks the right skill. Example: 'Run a Callaway–Sant'Anna event study with panel data and produce HonestDiD robustness and journal-style tables.' Or trigger the full pipeline: 'Start a new paper project: air pollution and Chinese labor market, CS design + provincial panel.' You can intervene at any step; outputs are saved for manual adjustment.
What are this agent's strengths and limitations?
- Comprehensive 9-stage pipeline covering the entire research workflow.
- In-house StatsPAI causal engine with 900+ functions, supporting multiple methods.
- Tri-stack support (Python/Stata/R) for user preference.
- Validated by 17 numeric benchmarks and 37 behavioral evals.
- Bilingual de-AIGC skills for international and Chinese journals.
- Tied to Claude Code for installation and usage.
- Large repository (76 collections, 1,096 skills) may require learning curve.
- Some community skills are upstream snapshots, possibly unmaintained.
- License is NOASSERTION, requiring careful compliance review.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| AERS: Auto-Empirical Research Skills This agent | 51 · Major gaps | Agent plugin / skillFree + model costs | ★ 4.7k | 4d ago | Stata | Claude Code |
| Wisp Science | 71 · Some gaps | Desktop appFree + model costs | ★ 1k | today | Rust | Codex · Claude Code · OpenAI API · Claude API |
| Qinyan Academic Skills | 48 · Major gaps | CLIFree + model costs | ★ 966 | 3mo ago | Python | Codex · Claude Code |
| OpenScience: Open-Source AI Workbench for Scientific Research | 45 · Major gaps | CLIFreemium | ★ 4.1k | today | TypeScript | OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows the repository includes a security policy (SECURITY.md), pinned dependencies (requirements.txt), least-privilege workflow permissions (check-external-links.yml etc. use contents: read and issues: write), and safety guards in the install script (test_install_skills.py tests refusal to overwrite source directories). However, no user confirmation mechanism, detailed data flow transparency, specific sensitive data handling, explicit external effects, or rollback mechanism were found. Source attribution is present via upstream links and license audit, but publisher identity is unverified. Thus, most criteria score low, but dependency security (pinned versions, Dependabot, OpenSSF Scorecard) and source attribution (upstream links) score higher.
Internal consistency is good: README claims 76 collections and 1096 skills, consistent with the catalog; test files (conftest.py, test_install_skills.py, test_numeric_check.py) show test infrastructure. Dependency availability: requirements.txt pins versions, but no install instructions or dependency list provided. Failure messages: tests show error messages, but no user-facing failure handling documentation. Thus, self-consistency and dependency availability score moderate, failure messages low.
README clearly targets empirical researchers and AI agent developers, with multiple usage scenarios (whole-repo import, single-skill import). Capability boundaries are described (router skill, not loading all skills), but trigger precision (SKILL.md trigger conditions) not detailed in provided files. Environment fit: supports Codex, CodeBuddy, Claude Code, but no specific environment requirements. Thus, audience and scenarios, capability boundaries, environment fit score moderate, trigger precision low.
Information architecture is clear: README provides table of contents, skill list, and grouping. Install notes mentioned in README and INSTALL.md, but no detailed steps. Naming stability: project renamed, but redirect provided. Examples and FAQ: README provides quickstart and verification, but no FAQ. Known limitations: not explicitly listed. License: CC BY-SA 4.0 explicit. Versioning and changelog: not provided. Maintenance responsibility: README and SECURITY.md show maintainer response, but not explicit. Thus, information architecture, install notes, naming stability, examples, license, maintenance responsibility score moderate, known limitations and versioning low.
Output usability: skills aim to generate reproducible empirical papers, but no output format examples. Marginal value: provides 1096 skills, but no comparison with existing tools. Cost-benefit: no performance or resource usage data. Thus, output usability and marginal value score moderate, cost-benefit low.
Claim traceability: README claims skills are verified, but no specific verification data. Cross-source corroboration: upstream links provided, but no independent verification. Fact-inference separation: not explicitly distinguished. Thus, claim traceability scores moderate, cross-source corroboration and fact-inference separation low.
- Publisher identity is unverified; treat skills from unknown sources with caution.
- No user confirmation mechanism; installing or executing skills may lack explicit consent.
- Data flow and sensitive data handling details are insufficient; review skill code.
- No rollback mechanism; installation may be hard to undo.
- Dependency security is pinned, but no install instructions provided, potentially affecting reproducibility.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v2026.07 AERS v2026.07 — first tagged release
Release notes excerpt
Highlights - First tagged release. Every gate is green: make check (validate + 183 unit tests + eval-harness + numeric benchmarks), validate-catalog, quality-evals, and OpenSSF Scorecard on main. - Rigor double loop at 13 method families. Each family closes taxonomy tag → eval scenario (does the agent avoid the method's known traps?) → deterministic numeric benchmark (does it recover the true effect?). New this cycle: heterogeneous treatment effects (CATE) and quantile treatment effects (QTE). …View on GitHub ↗