Description
Ask Claude Code to run an empirical paper and it writes the DID as a plain OLS, does one sample swap as the only robustness check and hands you tables a journal would send back. General models lack the muscle memory of econometrics. Auto-Empirical Research Skills (AERS), maintained by Stanford REAP and CoPaper.AI, is an agent skills library: 76 collections and over a thousand skills vendored into one repository, so once installed into Claude Code or Codex a request like "run a Callaway–Sant'Anna event study on this panel with HonestDiD robustness" makes the agent pick the right skill on its own.
The library is organised by the nine stages of a paper: topic refinement, literature review, data acquisition, identification strategy, estimation, robustness audit, tables and figures, writing and peer review, de-AIGC and submission, each mapped to concrete skill collections. Paper-WorkFlow chains the nine into one end-to-end pipeline whose every step you can take over by hand before letting it continue. The seven in-house skills at the core (the StatsPAI causal engine, explicit Python, Stata and R stacks, de-AIGC, the AER submission kit and the orchestrator) are pinned down by numeric benchmarks and behavioural evals rather than claims.
Installation is handing the repository URL to the agent and saying where to put it, a single Claude Code marketplace command, or copying one skill folder. A built-in search page filters the thousand-plus skills by method, stage, language and licence; every community collection links back to its upstream with licence and commercial-use status, and a security scan report ships with the repo.
Nine stage pipeline: Every stage from idea to submission maps to specific skills; the Paper-WorkFlow orchestrator runs them end to end with artefacts on disk and manual takeover at any step.
StatsPAI causal engine: sp.causal(...) runs DID, RD, IV, SCM, DML and matching in one call, backed by 900+ estimation functions.
Three explicit stacks: Full empirical skills for Python (pandas, statsmodels, pyfixest), Stata (reghdfe, csdid, rdrobust) and R (fixest, did, HonestDiD with Quarto).
Robustness and replication audit: Ten point replication package checks, Honest-DiD, referee simulation and R&R rehearsal.
Writing and de-AIGC: Economics writing guides, LaTeX and Quarto typesetting, bilingual AI trace and watermark removal aimed at Turnitin, CNKI and similar detectors.
Literature and data: PRISMA systematic reviews, OpenAlex and CrossRef citation checks, SEC filings, public databases and scraping.
Verifiable: 19 numeric benchmark tasks and 42 behavioural eval scenarios; aers-score grades your own agent on the same exam.
Install options: Claude Code plugin marketplace, whole-repo import for Codex and CodeBuddy, or copy a single skill folder; a skill search page is included.
Transparent provenance: Each community collection links to its upstream repo with licence and commercial-use status, plus a security scan report.
The library is organised by the nine stages of a paper: topic refinement, literature review, data acquisition, identification strategy, estimation, robustness audit, tables and figures, writing and peer review, de-AIGC and submission, each mapped to concrete skill collections. Paper-WorkFlow chains the nine into one end-to-end pipeline whose every step you can take over by hand before letting it continue. The seven in-house skills at the core (the StatsPAI causal engine, explicit Python, Stata and R stacks, de-AIGC, the AER submission kit and the orchestrator) are pinned down by numeric benchmarks and behavioural evals rather than claims.
Installation is handing the repository URL to the agent and saying where to put it, a single Claude Code marketplace command, or copying one skill folder. A built-in search page filters the thousand-plus skills by method, stage, language and licence; every community collection links back to its upstream with licence and commercial-use status, and a security scan report ships with the repo.
Features
Nine stage pipeline: Every stage from idea to submission maps to specific skills; the Paper-WorkFlow orchestrator runs them end to end with artefacts on disk and manual takeover at any step.
StatsPAI causal engine: sp.causal(...) runs DID, RD, IV, SCM, DML and matching in one call, backed by 900+ estimation functions.
Three explicit stacks: Full empirical skills for Python (pandas, statsmodels, pyfixest), Stata (reghdfe, csdid, rdrobust) and R (fixest, did, HonestDiD with Quarto).
Robustness and replication audit: Ten point replication package checks, Honest-DiD, referee simulation and R&R rehearsal.
Writing and de-AIGC: Economics writing guides, LaTeX and Quarto typesetting, bilingual AI trace and watermark removal aimed at Turnitin, CNKI and similar detectors.
Literature and data: PRISMA systematic reviews, OpenAlex and CrossRef citation checks, SEC filings, public databases and scraping.
Verifiable: 19 numeric benchmark tasks and 42 behavioural eval scenarios; aers-score grades your own agent on the same exam.
Install options: Claude Code plugin marketplace, whole-repo import for Codex and CodeBuddy, or copy a single skill folder; a skill search page is included.
Transparent provenance: Each community collection links to its upstream repo with licence and commercial-use status, plus a security scan report.
