Docs

Which skill should I use?

1,052 skills is a lot. This guide routes you to the right one by what you are trying to do, then by method and language. For interactive filtering use search.html; for the full tag index see TAXONOMY.md; for copy-paste end-to-end runs see GOLDEN_WORKFLOWS.md.

Trust note: a green hygiene score in SKILL_HYGIENE.md means a skill is well-formed. Whether it makes an agent produce correct econometrics is what eval-harness/ and benchmark/ measure. Prefer skills whose behavior is covered by an eval scenario.

Start from your goal

I want to…	Use	Notes
Run a full empirical pipeline in Python	StatsPAI or 00.1 Python	data → identification → estimation → robustness → tables/figures
Run the same pipeline in Stata / R	00.2 Stata · 00.3 R	language-matched ports of the pipeline
Audit my identification before writing	aer-identification	gates TWFE behind the 2×2 conditions; weak-IV / RDD pitfalls
Build the robustness gauntlet referees expect	aer-robustness	placebo triplet, spec curve, heterogeneity by theory
Write the introduction / abstract for a top-5	aer-introduction	Keith Head 5-paragraph formula; 100-word abstract
Format tables & figures to AER house style	aer-tables-figures	booktabs, figure notes, one claim per exhibit
Pre-submission compliance check	aer-submission	abstract length, disclosure, AI usage, ScholarOne package
Write a R&R / rebuttal letter	aer-rebuttal	written against the revised manuscript
Build a replication package	aer-replication	AEA Data & Code Availability Policy, openICPSR-ready
Get causal-method code templates	causal-inference-mixtape · MixtapeTools	10 designs, Python/R/Stata
Lower the AI-writing signal of a Chinese draft	chinese-de-aigc	structural rewrite, facts preserved
De-slop English academic prose	avoid-ai-writing · stop-slop
Run a literature review	literature-review	see also stage:literature in the taxonomy

Then narrow by method

Pick the design, then jump to the method tag in TAXONOMY.md:

• Staggered diff-in-diff? You almost never want plain TWFE. Look for staggered-did skills (Callaway–Sant'Anna, Sun–Abraham, Borusyak et al., de Chaisemartin–D'Haultfœuille). Covered by the eval statspai-staggered-did.
• Instrumental variables? Check the iv tag; you want first-stage F reporting and weak-IV-robust inference. Covered by statspai-weak-iv.
• Regression discontinuity? Check rdd; you want a density/manipulation test, bandwidth sensitivity, and robust bias-corrected CIs. Covered by statspai-rdd-diagnostics.
• Synthetic control / matching / DML? Tags synthetic-control, matching, dml.

Then narrow by language

Filter the language facet (python, r, stata, latex) in search.html. The 00.x pipeline skills exist in all three analysis languages so you can match your stack.

A 30-second decision tree

Do you have a dataset and need results?
├─ yes → identification first:  aer-identification
│        then run the pipeline:  00 / 00.1 / 00.2 / 00.3 (by language)
│        then robustness:        aer-robustness
└─ no, I have results and need a paper
   ├─ intro/abstract           → aer-introduction
   ├─ tables/figures           → aer-tables-figures
   ├─ submission compliance    → aer-submission
   ├─ referee response (R&R)   → aer-rebuttal
   └─ reproducibility package  → aer-replication

Polishing prose?
├─ Chinese → chinese-de-aigc
└─ English → avoid-ai-writing / stop-slop

See also

• GOLDEN_WORKFLOWS.md — ready-to-run multi-skill workflows
• The numbered workflow-stage notes: 01-选题与研究设计 … 10-审稿回复与学术答辩
• SKILL_CATALOG.md — the full generated catalog