Econ Agent Skills (Jonas Weinert)

license: CC0-1.0 per the repo LICENSE file and README ("CC0 1.0 — public domain"); GitHub's API reports the licence as NOASSERTION / "Other" — record both · 11 skills · last update: 2026-05-05

Source: https://github.com/JonasWeinert/EconAgentSkills

Maintainers: Jonas Weinert (JonasWeinert on GitHub)

Related project entry: econ-agent-skills

Compatibility: claude-code cursor codex gemini-cli

Ten opinionated SKILL.md skills across four workflow stages — data (working-with-data, api-data-fetcher, stata-data-cleaning), analysis (r-econometrics, stata-regression, python-panel-data), writing (academic-paper-writer, latex-tables), and communication (econ-visualization, beamer-presentation) — plus the repo-wide AGENT_POLICY.md decision ladder, cataloged here as an eleventh entry because the README calls it "the most important file in the project" and every skill defers to it. The opinions are the whole point: each analysis skill carries a decision tree that pins the agent to the modern DiD literature (Callaway-Sant'Anna, Sun-Abraham, Borusyak-Jaravel-Spiess, de Chaisemartin-D'Haultfoeuille, with at least two estimators reported) instead of the 2010-era TWFE it would otherwise reach for, to weak-IV-robust Anderson-Rubin inference instead of "first-stage F > 10", to rdrobust at the MSE-optimal bandwidth for RDD, and to wild cluster bootstrap below roughly 30 clusters — while DIME Analytics conventions (iefolder, iecodebook, ieboilstart, master do-files, immutable raw data, the four-tier replicability standard, the Reviewing Graphs and Submit Table checklists) supply the reproducibility layer, translated to Python, R, and LaTeX where DIME's own guidance is Stata-only. Three caveats a user should know. First, licensing: the repo LICENSE is the full CC0 1.0 text and the README says "CC0 1.0 — public domain", but GitHub's licence API reports NOASSERTION, so the public-domain dedication is the maintainer's stated intent rather than a platform-detected licence. Second, academic-paper-writer and beamer-presentation cross-link to a latex-econ-model skill that does not exist anywhere in the repo — a dangling reference, not a missing install. Third, the framing as an econometrics pack undersells the spread: four of the ten skills are writing and presentation rather than estimation. Upstream each skill also ships a deeper reference.md and runnable examples/ (R, Python, Stata, Julia, LaTeX); only the SKILL.md routing layer is bundled here. The repo has been unchanged since 2026-05-05.

Source YAML: skills/econ-agent-skills.yml

Skills

analysis (3)

Skill	Field	Stages	Description	Full text	Source	Updated
/python-panel-data	econometrics	data-analysis code-generation	Panel and causal-inference code in Python via pyfixest, linearmodels.PanelOLS, and statsmodels, with correct fixed effects, two-way clustering, IV2SLS, event studies through sunab, weak-IV diagnostics, and the wildboottest package below 30 clusters. Unusually candid about the ecosystem gap: for Callaway-Sant'Anna, Borusyak-Jaravel-Spiess, or de Chaisemartin-D'Haultfoeuille it routes the user to the R or Stata skill (or rpy2) rather than improvising, and documents the cross-language jump in the decisions log.	view	origin	2026-05-05
/r-econometrics	econometrics	data-analysis code-generation	Modern reproducible R for causal inference and panel econometrics built on fixest, with a decision tree that routes staggered adoption away from plain TWFE to Callaway-Sant'Anna (the default), Sun-Abraham, Borusyak-Jaravel-Spiess, or de Chaisemartin-D'Haultfoeuille and asks for at least two estimators when possible. IV gets weak-IV-robust Anderson-Rubin confidence sets rather than a first-stage F threshold, sharp RDD gets rdrobust at the MSE-optimal bandwidth with h/2, h, 2h sensitivity, and fewer than 30 clusters gets wild cluster bootstrap via fwildclusterboot.	view	origin	2026-05-05
/stata-regression	econometrics	data-analysis code-generation	Stata regression workflows on DIME Analytics conventions (ietoolkit, iefolder, a master do-file, ieboilstart, dynamic absolute path globals, iebaltab for balance, esttab for output) with modern estimators as the default: reghdfe for fixed effects, csdid / eventstudyinteract / did_imputation / jwdid for staggered DiD, ivreg2 with weakid plus weakivtest for weak instruments, boottest when clusters are few, and rdrobust for RDD. Every table is written to file by code — never copy-pasted from the console.	view	origin	2026-05-05

data-handling (3)

Skill	Field	Stages	Description	Full text	Source	Updated
/api-data-fetcher	economics	data-acquisition	Builds reproducible Python acquisition pipelines for economic APIs (FRED, World Bank, BLS, OECD SDMX, IMF, Eurostat, Yahoo Finance) with DIME conventions ported to Python: an immutable data/raw/ layer written with vintage suffixes, a run_all.py master orchestrator, dynamic absolute paths via pathlib, API keys treated as PII and read only from env vars, content-addressed deterministic caching, and a source codebook recording series ID, units, frequency, transformation, vintage, and access date.	view	origin	2026-05-05
/stata-data-cleaning	economics	data-analysis	Generates DIME-conventional Stata cleaning workflows — ieboilstart, the iefolder DataWork structure, iecodebook template/apply for repetitive renames and recodes, ieduplicates and iecompdup for duplicate resolution, extended missing values for survey codes, numerically stored labelled categoricals, isid and assert validation, encryption for PII. Follows DIME's rule of identifying and documenting irregularities before correcting them, because many fixes depend on design choices the PI owns.	view	origin	2026-05-05
/working-with-data	economics	data-acquisition data-analysis	The foundational data discipline every other skill in the pack assumes: state and assert the unit of analysis before any merge, regression, or figure; treat an ID as a contract (unique, non-missing, immutable, type-stable, documented); declare merge cardinality up front and inspect every non-match, with no such thing as a safe m:m; keep data/raw/ write-once; log every filter with N before and after; and maintain one master dataset per unit of observation. Gives parallel patterns for Stata, R, and pandas.	view	origin	2026-05-05

drafting (2)

Skill	Field	Stages	Description	Full text	Source	Updated
/academic-paper-writer	economics	paper-drafting revision-editing	Drafts and revises economics papers in journal structure on the premise that the identification argument carries the paper — the methods section leads with what variation drives the coefficient, not with the regression equation — and polices causal language, reserving "causes" and "drives" for a defended design and "is associated with" for descriptive results. Refuses to invent the headline number or the identification strategy: both are blocking questions for the author.	view	origin	2026-05-05
/latex-tables	economics	paper-drafting	Produces publication-ready LaTeX tables from code — esttab/outreg2 in Stata, modelsummary or fixest::etable in R, stargazer or pyfixest.etable in Python — written directly into the paper's tabs/ folder at DIME's "full replicability" tier, with booktabs three-line styling, threeparttable notes carrying SE type and cluster level, siunitx decimal alignment, and Cref-friendly labels. The .tex file is never hand-edited; the script is re-run instead.	view	origin	2026-05-05

figures (1)

Skill	Field	Stages	Description	Full text	Source	Updated
/econ-visualization	economics	data-analysis paper-drafting	Code-produced publication figures in ggplot2, matplotlib/seaborn, or Stata twoway/coefplot, exported as vector PDF straight into the paper's figs/ folder, with a chart-type decision tree (event-study coefficient plots, balance plots, recession-shaded time series, binscatters, density plots, choropleths) and colourblind-safe Okabe-Ito or viridis palettes — never red-green for treatment versus control. Follows DIME's Reviewing Graphs checklist and separates caption-dependent paper figures from standalone visuals that need an in-figure title and a source line.	view	origin	2026-05-05

infra (1)

Skill	Field	Stages	Description	Full text	Source	Updated
AGENT_POLICY.md	economics		The repo-wide human-in-the-loop decision ladder every skill defers to, sorting each choice into four levels: ASK and block (unit of analysis, treatment and outcome definitions, identification strategy, cluster level, sample restrictions, anything touching PII or raw data), DEFAULT and flag with an override, DOCUMENT in the decisions log and proceed, or simply PROCEED. Also fixes the cross-cutting rules — raw data immutable, numbers never typed into prose, no silent estimator swaps, every dropped observation logged with N before and after — and supplies the decisions-log header template whose [HUMAN: please confirm] tags let an unattended agent surface what it inferred.	view	origin	2026-05-05

slides (1)

Skill	Field	Stages	Description	Full text	Source	Updated
/beamer-presentation	economics	dissemination	Builds LaTeX Beamer talks that share the paper's notation.tex macros and the same tabs/ and figs/ artefacts, so one rerun of the analysis updates paper and slides together and no number is ever re-typeset on a slide. Structures to an explicit time budget (15-20 min conference, 45-60 min seminar, 75-90 min job-market talk, 10-15 min policy briefing), one claim per slide, the claim in the frame title and the evidence in the body, detail pushed to backup slides for Q&A.	view	origin	2026-05-05