lishn6/awesome-ai-econ-research-writing

Make AI Research Writing Better for Economists

Overview

This project collects prompt templates and agent skills for economics research writing. It is adapted from Leey21/awesome-ai-research-writing, an excellent AI research writing prompt and skills collection originally oriented more toward general AI/CS research writing.

This version keeps the original spirit of reusable research-writing prompts and skills, but redesigns the workflow for economics papers. The focus shifts to clear research questions, credible identification, transparent data and sample construction, accurate result interpretation, structural modeling logic, counterfactuals, welfare analysis, and restrained academic prose.

Credit and thanks go to the original author and contributors of Leey21/awesome-ai-research-writing. This repository is an economics-focused adaptation built on top of that idea.

The repository now separates reusable assets into two layers:

• Prompts: copy-and-paste templates for specific writing tasks.
• Skills: agent-readable SKILL.md workflows for Codex, Cursor, Claude Code, and similar tools.

The main focus is English economics paper writing, with Chinese-English translation and Chinese academic editing support where useful.

What This Economics Version Adds

This project reorganizes the original research-writing idea around two major economics paper traditions:

1. Reduced-form / empirical economics writing

   • Chinese-English translation for economics papers.
   • Introduction, literature review, data, empirical strategy, and results prompts.
   • Identification-focused prompts for DID, IV, RDD, event studies, panel fixed effects, synthetic control, and related designs.
   • Regression table writing for main results, robustness checks, heterogeneity, and mechanisms.
   • Referee-style review and R&R response-letter prompts.
   • An economics-paper humanizer for removing AI/slop patterns from completed drafts while preserving academic style, including a scoring rubric for AI flavor.

2. Structural-form economics writing

   • Separate workflow for model environment, primitives, assumptions, identification of structural parameters, estimation, model fit, counterfactuals, and welfare.
   • Branch-specific guidance for IO/demand estimation, dynamic discrete choice, search and matching, quantitative macro, trade and quantitative spatial economics, auctions, and market design.
   • A dedicated structural-economics-writing skill with branch references.

Repository Structure

prompts/
  economics_writing_style.md
  reduced_form_economics_writing.md
  structural_economics_writing.md
skills/
  economics-writing-style/
  good-question/
  reduced-form-economics-writing/
  identification-auditor/
  regression-table-writer/
  literature-positioning/
  referee-simulator/
  response-letter-writer/
  structural-economics-writing/
  econ-paper-humanizer/
images/
  example.png
  model-rank.png
  nano-banana.png

Prompt Files

File	Scope	Use When
prompts/economics_writing_style.md	General economics writing style	You need a reusable style layer for concise, restrained economics prose, overclaim control, results-versus-mechanism separation, word choice, figures, tables, numbers, units, dates, and citations.
prompts/reduced_form_economics_writing.md	Empirical and reduced-form economics writing	You need copyable prompts for translation, polishing, Introduction, literature review, data, empirical strategy, regression results, robustness, heterogeneity, mechanisms, referee review, or R&R letters.
prompts/structural_economics_writing.md	Structural-form economics writing	You are writing or auditing model-based papers involving primitives, parameter identification, estimation, model fit, counterfactuals, welfare, or structural branches such as IO, dynamic choice, search, macro, trade/spatial, auctions, or market design.

Skills

Each folder below contains a SKILL.md file that can be copied into a compatible skills directory such as .claude/skills/, .cursor/skills/, or another agent-supported skill path.

Skill	Path	Purpose
economics-writing-style	skills/economics-writing-style/SKILL.md	Shared economics prose style layer for reduced-form, structural, theory, policy, literature, results, figures, tables, referee reports, and response letters.
good-question	skills/good-question/SKILL.md	Turn a vague economics topic, literature gap, dataset, policy shock, or model idea into a sharp research question with stakes, evidence, rivals, falsifiers, and a feasible pilot.
reduced-form-economics-writing	skills/reduced-form-economics-writing/SKILL.md	Reduced-form and empirical economics paper writing workflow, including bilingual Chinese-English support.
identification-auditor	skills/identification-auditor/SKILL.md	Identification diagnostics for DID, event studies, IV, RDD, panel FE, synthetic control, and related designs.
regression-table-writer	skills/regression-table-writer/SKILL.md	Convert regression, robustness, heterogeneity, and mechanism tables into accurate paper text.
literature-positioning	skills/literature-positioning/SKILL.md	Organize literature reviews and contribution-to-literature sections.
referee-simulator	skills/referee-simulator/SKILL.md	Simulate economics journal referee reports and revision strategies.
response-letter-writer	skills/response-letter-writer/SKILL.md	Draft R&R response letters and point-by-point replies.
structural-economics-writing	skills/structural-economics-writing/SKILL.md	Structural economics writing for models, estimation, counterfactuals, welfare, and branch-specific workflows.
econ-paper-humanizer	skills/econ-paper-humanizer/SKILL.md	Remove AI-written texture and generic slop from completed English economics paper prose while preserving academic register, causal caution, citations, LaTeX, and identification boundaries. Includes a 60-point AI/slop scoring rubric.

How to Use

For one-off writing tasks, open the relevant prompt file and copy the template you need.

For agent workflows, copy a skill folder into the skill directory used by your tool. For example, with a local project that supports Claude/Cursor-style skills:

mkdir -p .claude/skills
cp -R skills/economics-writing-style .claude/skills/
cp -R skills/reduced-form-economics-writing .claude/skills/
cp -R skills/identification-auditor .claude/skills/
cp -R skills/econ-paper-humanizer .claude/skills/

Then ask naturally, for example:

• “Use good-question to sharpen this vague topic into an economics research question.”
• “Use the reduced-form-economics-writing skill to rewrite my Introduction.”
• “Check whether this DID design has a credible identification argument.”
• “Turn this regression table into a Results paragraph, but do not overstate causality.”
• “Review this structural model section as a referee.”
• “Use econ-paper-humanizer to remove AI flavor from this completed Introduction and score the remaining AI/slop signal.”

Using the Skills with Codex

Codex reads skills from your Codex skills directory. If you have not customized CODEX_HOME, this is usually ~/.codex/skills/.

To install all economics writing skills from this repository into Codex:

mkdir -p ~/.codex/skills
cp -R skills/* ~/.codex/skills/

To install only selected skills:

mkdir -p ~/.codex/skills
cp -R skills/economics-writing-style ~/.codex/skills/
cp -R skills/good-question ~/.codex/skills/
cp -R skills/reduced-form-economics-writing ~/.codex/skills/
cp -R skills/identification-auditor ~/.codex/skills/
cp -R skills/regression-table-writer ~/.codex/skills/
cp -R skills/econ-paper-humanizer ~/.codex/skills/

After copying the folders, restart or reload Codex so it can discover the new SKILL.md files. Then ask Codex to use a skill by name, or describe the task naturally. For example:

• “Use good-question to turn this literature gap into a testable economics question.”
• “Use the reduced-form-economics-writing skill to polish this Introduction.”
• “Use identification-auditor to check the identifying assumptions in this DID section.”
• “Use regression-table-writer to turn this table into a cautious Results paragraph.”
• “Use structural-economics-writing to review the model environment and identification logic.”
• “Use econ-paper-humanizer to score this paragraph for AI/slop flavor and then rewrite it conservatively.”

When using these skills with Codex, provide the relevant paper section, table notes, regression output, model equations, or referee comment in the same message or as project files. Codex should not infer missing coefficients, citations, sample restrictions, institutional facts, or empirical results.

Skill-by-Skill Usage

Each skill works best when you name the skill and provide the concrete evidence it should use. The examples below are written for Codex, but the same prompts work in other skill-aware agents.

good-question

Use this before committing to a topic, proposal, dataset, policy shock, or model idea.

Provide:

• field and subfield;
• current rough idea;
• what the closest literature already studies;
• the gap you think remains;
• available data, identification strategy, model, or constraints;
• target output: paper, thesis, proposal, pilot, or grant.

Example:

Use good-question to sharpen this economics research idea.

Field:
Current rough idea:
Existing literature already studies:
Gap I want to address:
Available data / method:
Target:
My biggest concern:

reduced-form-economics-writing

Use this for drafting, rewriting, translating, or reorganizing empirical economics paper sections.

Provide:

• target section: abstract, introduction, data, empirical strategy, results, robustness, mechanisms, or conclusion;
• research question, setting, data, identification, main results, and target journal or audience;
• the current draft if rewriting or translating.

Example:

Use reduced-form-economics-writing to rewrite this Introduction for an economics journal.

Research question:
Setting and data:
Identification strategy:
Main findings:
Contribution:
Draft:

identification-auditor

Use this when you want a candid check of whether an empirical design can support the causal claim.

Provide:

• treatment, outcome, unit of observation, sample, and timing;
• estimating equation or design description;
• identifying variation, controls, fixed effects, and clustering;
• proposed robustness checks or threats you are worried about.

Example:

Use identification-auditor to review this DID design.

Treatment:
Outcome:
Unit and sample:
Timing:
Equation:
Identifying variation:
Fixed effects / controls / clustering:
Main threats:

regression-table-writer

Use this to turn regression, robustness, heterogeneity, mechanism, placebo, or event-study tables into accurate paper text.

Provide:

• the table or regression output;
• variable definitions, units, transformations, fixed effects, controls, standard errors, and sample;
• baseline means when economic magnitude matters;
• whether the table is main results, robustness, heterogeneity, or mechanisms.

Example:

Use regression-table-writer to write a Results paragraph from this table.

Table purpose:
Outcome definition:
Treatment / key variable:
Units and transformations:
Controls / fixed effects / clustering:
Baseline mean:
Table:

literature-positioning

Use this for literature reviews, contribution paragraphs, citation maps, and related-work sections.

Provide:

• research question and contribution;
• key papers or citation list;
• what each literature strand already studies;
• what your paper adds relative to each strand.

Example:

Use literature-positioning to organize this contribution section.

Research question:
Main contribution:
Closest papers:
Literature strands:
What my paper adds:
Draft or notes:

referee-simulator

Use this to simulate an economics journal referee report or seminar-style critique.

Provide:

• abstract, introduction, proposal, or full paper sections;
• empirical design or structural model summary;
• main tables or result summaries;
• target journal or field if relevant.

Example:

Use referee-simulator to review this paper as an economics journal referee.

Target journal or audience:
Abstract:
Research question and contribution:
Design / model:
Main results:
Sections or draft:

response-letter-writer

Use this for revise-and-resubmit letters, referee replies, editor responses, rebuttals, or revision plans.

Provide:

• editor letter and referee comments;
• what changes have already been made;
• new analyses, tables, sections, or wording changes;
• places where you need to disagree or explain infeasibility.

Example:

Use response-letter-writer to draft point-by-point replies.

Editor letter:
Referee comments:
Changes already made:
New analyses or tables:
Where we disagree:
Manuscript locations if known:

structural-economics-writing

Use this for structural papers involving model environment, primitives, parameter identification, estimation, model fit, counterfactuals, welfare, and policy analysis.

Provide:

• branch: IO, dynamic discrete choice, search, macro, trade/spatial, auctions, market design, or other;
• agents, timing, states, choices, payoffs, constraints, information, and equilibrium;
• primitives, parameters, moments, estimation method, and model fit;
• counterfactuals and welfare objects.

Example:

Use structural-economics-writing to review this model section.

Branch:
Agents and environment:
Timing / states / choices:
Primitives and parameters:
Data moments and identification:
Estimation:
Counterfactuals:
Welfare object:
Draft:

econ-paper-humanizer

Use this after you already have a completed English economics paper draft, but the prose sounds too generic, AI-polished, over-structured, or full of vague importance claims.

The skill removes AI/slop patterns while preserving economics paper style. It is designed to keep citations, LaTeX commands, causal caution, identification boundaries, technical repetition, and journal-appropriate formality intact.

It also includes a 60-point scoring rubric for diagnosing AI flavor:

Dimension	What it checks
Directness	Whether the prose states claims, estimates, identifying variation, mechanisms, or model objects directly.
Rhythm	Whether sentence and paragraph lengths vary naturally rather than feeling templated or punchline-driven.
Reader Trust	Whether the prose avoids hand-holding, meta-commentary, and repeated obvious transitions.
Authenticity	Whether the passage sounds like an economist making real judgments about evidence, limits, and interpretation.
Density	Whether each sentence does analytical work rather than carrying filler, vague importance claims, or repeated framing.
Specificity / Evidence Discipline	Whether claims are tied to concrete data, variation, mechanisms, margins, parameters, or assumptions, with causal language matched to evidence.

Scores of 54-60 indicate low AI/slop signal, 46-53 mild signal, 36-45 noticeable signal, and below 36 strong signal. Any dimension below 7 should be diagnosed and revised.

Provide:

• target section: abstract, introduction, literature review, data, empirical strategy, results, mechanism, model, counterfactual, conclusion, or response letter;
• the current completed draft;
• whether you want only a rewrite, only an AI/slop score, or both;
• any facts, estimates, citations, or interpretation boundaries that must not change.

Example:

Use econ-paper-humanizer to score and rewrite this completed Introduction.

Section:
Goal:
Must preserve:
Do not infer:
Draft:

Design Principles

• Do not invent facts, citations, coefficients, samples, institutions, or results.
• Use Journal of Political Economy citation and reference style by default, following Chicago author-date conventions.
• Keep causal language aligned with the research design or model assumptions.
• Separate reduced-form empirical writing from structural-form writing.
• Treat robustness, heterogeneity, and mechanisms as evidence tied to specific concerns, not as decorative sections.
• Prefer clear, restrained English academic prose; provide Chinese explanation when it helps the author revise.

Roadmap

• Add journal-specific style profiles for general-interest and field journals.
• Expand structural branches with more specialized references.
• Add examples using real anonymized table formats and section drafts.
• Add installation notes for OpenSkills and common agent tools.