‘Technology pulls itself up by the bootstraps by giving scientific researchers vastly more powerful tools to work with’

For millennia, humans have sought tools to make ourselves more productive. That is no different today, as Joel Mokyr, the 2025 Nobel Prize winner, suggested in a 2018 paper. Artificial intelligence – Claude Code, Codex, and there’ll probably be many other alternatives soon – has accelerated this to a new level. In the process of using Claude Code, I’ve developed several tools that might be useful for others. Most live on my GitHub page; the LEAP house-style skill sits in a separate repository.

Claude skills

Diebolt

Diebolt is a simulated peer review system for economics working papers. It creates a panel of independent referee agents – each with a distinct subspecialisation, personality, and fictitious university affiliation – who review your paper in isolation, with strict information barriers that mirror real peer review. An editor then synthesises their feedback into a prioritised briefing, weighting senior referees more heavily. You choose which revisions to implement, and Diebolt updates your LaTeX files accordingly. Grounded in the refereeing principles of Berk, Harvey, and Hirshleifer (2017). Named after Claude Diebolt, the CNRS research professor and founding editor of Cliometrica.

EHRstyle

EHRstyle applies the Economic History Review house style to a LaTeX manuscript prepared for submission to the Review. It encodes the full Notes for Contributors – anonymity, double-spacing, word budget, UK -ize spelling, the Oxford comma, the journal’s specific capitalisation conventions (e.g. lower-case global financial crisis), and the peculiar date and number rules – and ships a bespoke biblatex style that produces EHR footnote references with short-title form, ibid. / idem handling, surname-first bibliography, “2nd ser.” and roman-volume treatment for pre-1992 EHR citations, and automatic exclusion of working papers from the consolidated bibliography. Drop-in templates for the anonymous title page and cover letter are included.

Janluiten

Janluiten is a lifelong sounding board for research ideas, named in honour of Jan Luiten van Zanden, Professor of Global Economic History at Utrecht University and my own PhD advisor. It helps you decide whether the idea on your desk is the right one to spend the next year on. It listens first, then asks – drawing on the long view of how research programs rise and fall, on cultural evolution, and on behavioural science. The deliverable is mentorship on three questions: what to work on, with whom, and why. The axiom underneath it all is that interest is the best predictor of success.

Kris

Kris catches hallucinated and chimeric references in .bib files. It runs Claude and Codex (GPT-5.4) agents in parallel using two genuinely independent verification methods – Claude does an API cascade across CrossRef, OpenAlex, Semantic Scholar, and arXiv with title-Jaccard and author-overlap scoring; Codex does DOI resolution, reverse search, retraction watch, ORCID checks, venue plausibility, citation-graph plausibility, and Internet Archive corroboration. References are randomly assigned to Codex agents so there are no fixed agent pairs, and where the two sides disagree an adversarial challenge round lets each side review the other’s evidence and either concede or refute. The output is an evidence-backed scorecard plus a list of confirmed fakes – designed so no hallucinated reference slips through silently. Built on top of two earlier ideas, with credit to their authors: PHY041’s claude-skill-citation-checker for the API-cascade method, and Imbad0202’s academic-research-skills for the contextual cross-model verification approach. Named in honour of Kris Inwood, the Guelph economic historian whose long collaboration with South African scholarship – and whose donation in 2020 of his personal library to LEAP – made many of the citations this skill is designed to verify reachable in the first place.

LEAPstyle

LEAPstyle applies the LEAP Economics house style to any academic output – LaTeX working papers, beamer slide decks, R graphs, or writing and editing. It handles preambles, the LEAP colour palette, the ggplot2 graph theme, and a comprehensive set of writing guidelines that match how LEAP publishes. Built for LEAP (Laboratory for the Economics of Africa’s Past) faculty and students who want their essays, papers, and presentations to carry a consistent visual and written identity. Lives in a separate repository from the other skills.

TannieDi

A LaTeX-to-Word round-trip tool for language editing. Many economics papers are written in LaTeX, but professional copy-editors typically work in Word. TannieDi automates the conversion: pack converts your LaTeX project into a clean Word manuscript (with figures and tables in a separate file), and unpack applies the editor’s tracked changes back into your original .tex files – preserving all LaTeX formatting, equations, and cross-references. Named after Di Kilpert, a language editor who has copy-edited many LEAP papers.

Tyler

Tyler converts a folder of academic PDFs into a token-efficient markdown wiki for literature review. Point it at a directory of papers and it produces one lightweight .md file per paper, plus an index, so Claude Code can load an entire literature into context without burning tokens on raw PDF parsing. Useful when you’re starting a new project and want to absorb fifty papers before writing a word yourself. Named in honour of Tyler Cowen, the economist behind Marginal Revolution and a famously voracious reader whose blog pioneered rapid, generous synthesis across fields.

Download

The general-purpose skills (Diebolt, EHRstyle, Janluiten, Kris, TannieDi, Tyler) are available at github.com/johanfourieza/econtools. The LEAP house-style skill lives separately at github.com/johanfourieza/leap. To install a skill in Claude Code, download the relevant skill.md (or SKILL.md) file and place it in your .claude/skills/{skillname}/ directory (e.g., .claude/skills/diebolt/skill.md). Each skill’s folder contains any additional files it needs to run.