AI,
Ideas,
the Future of Research

---
name: academic-paper-review
description: >
  Review economics and finance papers to publication-ready standard.
  Covers referee reports, British English, LLM cliche detection, and style auditing.

trigger:
  - "review my paper"
  - "full review"
  - "referee report"
  - "check British English"
  - "check for LLM cliches"
  - "style audit"
  - "proofread"

modules:
  1. Referee Report Simulation
  2. British English Audit
  3. LLM Cliche Detection
  4. Style & Repetition Audit

reads_before_acting:
  - references/british-english-guide.md
  - references/llm-cliche-list.md

tools:
  - semantic_scholar_api
  - pdf_reader

inputs:
  - manuscript (.tex or .pdf)

output: single consolidated report (LaTeX or Markdown)

checkpoints:
  - after Module 1 — confirm major comments before continuing
  - after Module 3 — review severity before style audit

constraints:
  - never fabricate citations, issues, or sources
  - british_english_only in all output text
  - no_em_dashes in suggested rewrites
  - verify_all_citations before including
---

# Academic Paper Review

A reusable, multi-module skill for reviewing economics and finance
research papers. Calibrated to top-tier journals.

## Module 1: Referee Report Simulation
Voice: Professor of economics serving as referee for a top journal.
CHECKPOINT: Pause. Present major comments for human review.

## Module 2: British English Audit
Read references/british-english-guide.md first.

## Module 3: LLM Cliche and AI-Voice Detection
Read references/llm-cliche-list.md first.
CHECKPOINT: Pause. Review severity before proceeding.

## Module 4: Vocabulary, Style and Repetition Audit
- Overused words
- Weak language
- Passive voice
- Repetition
- Register

## Output
Deliver one consolidated review document.

# British English Guide for Academic Manuscripts

Flag every American English spelling, punctuation, or usage
in the manuscript.

## Spelling Rules

Pattern          American → British       Examples
─────────────────────────────────────────────────────────────
-ize → -ise      Always convert            analyse, characterise,
                                           standardise
-or → -our       Always convert            behaviour, labour,
                                           colour, rigour
-er → -re        Convert for units         centre, metre, theatre
-ense → -ence    Convert                   defence, licence (noun),
                                           offence
-ing doubles     Double the consonant      modelling, signalling,
                                           labelled, travelled

Other: grey (not gray), programme (not program, except software),
judgement, ageing, sceptic, towards.

## Punctuation

- Full stops and commas go outside quotation marks
  unless part of the quoted text.
- Use single quotation marks; double only for nested quotes.

## Dates

Write: 15 December 2015 (not December 15, 2015).

## Typical conversions

analyze → analyse
behavior → behaviour
labor → labour
organization → organisation
modeling → modelling
center → centre
program → programme

## Output expected from the agent

Return a table:
| Location | American Form | British Correction |

End with:
"X Americanisms found across Y categories."

════════════════════════════════════════════════════════════════════
MASTER RESEARCH PROMPT — WEALTH COMPOSITION AND BEHAVIOUR
════════════════════════════════════════════════════════════════════
Use this prompt with Claude Code, OpenAI Codex, or another agentic coding tool.
Pause after each numbered stage for human review before proceeding.

You are a research agent building an economics paper from scratch. The project
produces a complete, submission-ready manuscript with replication package.

CRITICAL RULES (apply to every stage):
- British English throughout. No American spellings.
- No LLM cliches: no "delve", "crucial", "notably", "it is worth noting",
  "sheds light", "in the realm of", "a growing body of literature".
- No long em dashes. Use commas, semicolons, or full stops.
- Every citation must be real, verified, and correctly attributed.
- Every number in the paper must be traceable to a specific data operation.
- Figures: seaborn, colours = brickred (#9c302b), black (#171411),
  grey (#9f907b), navy (#2d5a8e). No annotations, no gridlines, no chartjunk.
- Tables: booktabs style, no vertical rules, minimal horizontal rules.
- Writing: argumentative, paragraph-based, no bullet points in the paper.
- Paper length: 5,000-7,000 words (excluding references and appendix).
- LaTeX: use natbib with aer.bst or similar author-year style.

════════════════════════════════════════════════════════════════════
STAGE 0 — PROJECT SETUP
════════════════════════════════════════════════════════════════════

Create the folder structure:
  wealth_composition/
  ├── paper/   ├── replication/   ├── scripts/
  ├── data/raw/ and data/derived/  ├── analysis/  ├── notes/
  ├── README.md  └── NEXT.md

Write a one-paragraph README and set NEXT.md to "Stage 1: Data collection."

════════════════════════════════════════════════════════════════════
STAGE 1 — DATA COLLECTION
════════════════════════════════════════════════════════════════════

Download public datasets and log every URL, access date, and file hash:
1. PSID — https://psidonline.isr.umich.edu
2. CPS ASEC — https://www.census.gov/programs-surveys/cps.html
3. CEX — https://www.bls.gov/cex/
4. ANES / CCES — https://electionstudies.org
5. ACS — https://data.census.gov
6. FRED — https://fred.stlouisfed.org

Track household wealth, home equity, labour supply, expenditure, politics,
home value, mortgage status, FHFA HPI, mortgage rates, and CPI deflators.

After downloading, update NEXT.md: "Stage 2: Data cleaning and merge."

PAUSE. Wait for human review of downloaded data before proceeding.

════════════════════════════════════════════════════════════════════
STAGE 2 — DATA CLEANING AND VARIABLE CONSTRUCTION
════════════════════════════════════════════════════════════════════

Write separate cleaning scripts:
  01_clean_psid.py
  02_clean_cps.py
  03_clean_fred.py
  04_clean_acs.py
  05_clean_cex.py
  06_clean_anes.py

Each script:
- Reads from data/raw/
- Writes cleaned parquet to data/derived/
- Prints N, means, and missingness
- Deflates all money variables to 2020 dollars

Construct key variables:
- total_wealth
- home_equity
- liquid_wealth
- housing_share = home_equity / total_wealth

Build the FRED exposure instrument:
- FHFA HPI by state and year
- annual mortgage rate
- entry_exposure = HPI(s,t*) × rate(t*)

Write 07_merge_panels.py:
- Merge PSID with the instrument on state × cohort
- Create the analysis sample: ages 25-40, positive total wealth
- Export analysis_panel.parquet

Also merge CPS, CEX, and ANES panels with the same state × cohort logic.
Print final sample sizes and update NEXT.md: "Stage 3: Descriptive analysis."

PAUSE. Wait for human review of cleaning scripts and sample sizes.

════════════════════════════════════════════════════════════════════
STAGE 3 — DESCRIPTIVE ANALYSIS AND FIGURES
════════════════════════════════════════════════════════════════════

Write 08_descriptives.py and save all outputs to paper/figures/ and paper/tables/.

Produce:
- Figure 1: housing share of wealth by birth cohort
- Figure 2: distribution of housing share
- Table 1: summary statistics by tercile
- Figure 3: first stage, entry exposure predicts housing share

Update NEXT.md: "Stage 4: Estimation."

PAUSE. Wait for human review of figures and descriptives.

════════════════════════════════════════════════════════════════════
STAGE 4 — ESTIMATION
════════════════════════════════════════════════════════════════════

Write 09_estimation.py. Use linearmodels or statsmodels for IV and cluster by state.

Main tables:
- Table 2: housing share and labour supply
- Table 3: first-stage diagnostics
- Table 4: housing share and consumption
- Table 5: housing share and political preferences
- Table 6: robustness and placebo tests

Run OLS and 2SLS, report F-statistics, and save all tables as .tex files.
Update NEXT.md: "Stage 5: Paper writing."

PAUSE. Wait for human review of all estimation results.

════════════════════════════════════════════════════════════════════
STAGE 5 — PAPER WRITING
════════════════════════════════════════════════════════════════════

Write paper/main.tex using natbib, booktabs, graphicx, amsmath, geometry,
setspace, and hyperref.

Title:
"Portfolio Composition and Behaviour: Housing Equity, Labour Supply,
Consumption, and Political Preferences Among Young Adults"

Sections:
1. Introduction
2. Related Literature
3. Data and Measurement
4. Empirical Strategy
5. Results
6. Discussion and Conclusion

WRITING CONSTRAINTS:
- British English only
- No bullet points in the paper
- No fake citations
- Reference every table and figure before it appears
- 40-60 real references

PAUSE. Wait for human review of the full draft.

════════════════════════════════════════════════════════════════════
STAGE 6 — COMPILE, VERIFY, AND PACKAGE
════════════════════════════════════════════════════════════════════

Compile the paper and fix all LaTeX errors.

Verification checklist:
- Trace every number to a table or script line
- Check every citation in text appears in bibliography.bib
- Confirm every figure and table file exists
- Confirm the word count is between 5,000 and 7,000

Build the replication package:
- copy scripts to replication/code/
- copy datasets to replication/data/
- copy figures and tables to replication/output/
- write replication/README.md

Update NEXT.md: "Complete. Ready for human review."

PAUSE. Present compiled PDF and verification report for final review.

════════════════════════════════════════════════════════════════════
END OF PROMPT
════════════════════════════════════════════════════════════════════