Auto-Empirical Research Skills (AERS) — first-party skills¶
license: CC BY-SA 4.0 (repo default); MIT for the mirrored first-party collections (StatsPAI, AER-skills, Paper-WorkFlow) · 22 skills · last update: 2026-07-22
Repo bundles 1,094 skills across 74 collections, ~90% vendored third-party (clo-author, academic-research-skills, ARIS, awesome-econ-ai-stuff, etc. — cataloged separately in RISE); this manifest catalogs the 22 first-party skills: the root router, the four flagship full-pipeline analysis skills (StatsPAI / Python / Stata / R backends), the chinese-de-aigc editing skill, the 15-skill AER-skills collection (mirror of brycewang-stanford/AER-skills), and the Paper-WorkFlow meta-orchestrator (git submodule of brycewang-stanford/Paper-WorkFlow). First-party status taken from the repo's own catalog/provenance.json origin labels.
Builds the robustness, heterogeneity, mechanism, and placebo battery AER referees demand, once main results exist and before the introduction's value-added paragraph is written.
Runs the AER-track analysis with StatsPAI — the agent-native Python engine and MCP server for causal inference, robustness, sensitivity, and publication-ready table export — after aer-identification fixes the design.
Classical 8-step end-to-end empirical workflow in the traditional Python stack (pandas/statsmodels/linearmodels/pyfixest/econml) — cleaning through publication tables/figures in AER house style, with epidemiology and ML-causal parallel modes; the non-StatsPAI, estimator-explicit counterpart.
Same 8-step empirical pipeline in the tidyverse + fixest R ecosystem — did/HonestDiD/rdrobust/gsynth/MatchIt/grf/DoubleML estimation with modelsummary/gt/ggplot2 publication outputs and Quarto reproducibility; epidemiology and ML-causal modes included.
Same 8-step empirical pipeline as a reproducible Stata .do workflow — reghdfe/ivreg2/csdid/did_imputation/sdid/rdrobust/synth/psmatch2 estimation, bacondecomp/honestdid/rwolf/oster robustness, esttab/coefplot outputs; epidemiology and ML-causal modes included.
Full empirical/causal analysis in Python via the StatsPAI vertical engine — AER/QJE-style DID/RD/IV/SCM/DML pipeline with estimating equation + identifying assumption, Table 1/2, event-study figure, robustness gauntlet; plus epidemiology, ML-causal, and Oaxaca-style decomposition modes and Word/Excel/LaTeX table export.
Internal-consistency audit of a near-final manuscript: headline numbers across abstract/introduction/results/tables, sample sizes, log-point vs percentage-point conversions, cross-references, and citation-bibliography matching.
Selects, implements, or stress-tests the causal identification strategy — DID (incl. staggered), IV (incl. weak-IV-robust inference), RDD, synthetic control, shift-share/Bartik — before introduction or results are written.
For primary-data and experimental projects, before the intervention: writes the pre-analysis plan, sizes the sample from a power calculation, and registers with the AEA RCT Registry.
Drafts or rewrites the introduction to the Keith Head / Bellemare five-paragraph formula with AER-specific conventions, and compresses abstracts to the mandatory 100-word limit.
Drafts and revises the body sections of an AER/AEJ manuscript — background, data, empirical strategy, results, mechanisms, conclusion — including equation conventions, results-paragraph narration, magnitude interpretation, and back-of-envelope policy calculations.
Chinese academic de-AIGC rewriting targeting CNKI/Wanfang/VIP/Turnitin-zh detectors — a five-step locate/diagnose/rewrite/self-score/recheck loop over 17 diagnostic rules for the five structural signatures of Chinese LLM prose, with per-section strategies.
Constructs and revises regression tables, descriptive-statistics tables, and figures in AER booktabs house style, with regression-table layout and figure-note conventions.
Evaluates whether a research idea clears the AER top-5 bar, routes between AER, AER:Insights, and the AEJ family, and sharpens a fuzzy contribution sentence into one publishable claim.
Router for the 15-skill AER-skills collection — sequences manuscript work from topic selection through rebuttal for AER, AER:Insights, and AEJ journals; routes, does not replace, the specialized skills.
Root router for whole-repo installs — classifies an empirical-research request by stage/method and dispatches to one of the 1,094 vendored skills via catalog JSON lookups instead of reading the repo wholesale.
Meta-orchestrator for a complete empirical paper (econ/social science): Stage 0-9 resumable pipeline from topic selection to submission with two hard human gates (Method Gate after estimation, Draft Quality Gate after polish); routes Python/StatsPAI, Stata, or R analysis backends and invokes existing skills rather than reimplementing them. Git submodule mirroring brycewang-stanford/Paper-WorkFlow.
Positions a manuscript against the economics literature — antecedents map for the introduction, cite/no-cite decisions, and verification that every bibliography entry is real, correctly attributed, and cited to the published version.
Assembles the AEA Data and Code Availability deposit — README writing and replication-package audit against the current AEA policy (including the February 2026 Data and Code Availability Policy) before the AEA Data Editor review.
Adversarial internal review before submission — simulates the AER desk screen plus three referee reports with calibrated severity, scores against the editorial rubric, and produces a prioritized revise list; rerun until the simulated verdict is at least major R&R.
Handles a Revise & Resubmit from AER/AEJ journals — triage, the concede/clarify/push-back decision per comment, and the point-by-point response-letter format editors actually read, aligned with manuscript revisions.
Final pre-submission audit for AER/AEJ journals — length, format, cover letter, per-author disclosure statements, file packaging, and routing among the AEA journal family.