Adrien Matray

Economist studying trade policy, finance, and macroeconomic outcomes. Research on tariffs, industrial policy, and global trade.

[Adrien Matray @AdrienMatray]

@cblatts's /review-plan skill is genuinely one of the best skills ever. The only catch: you run it again and again by hand.

So I made /review-plan-auto, it loops the review across multiple rounds on its own. 🔁

https://t.co/jjimOVWLbn

[Adrien Matray @AdrienMatray]

@PorzioTommaso is one of the most exciting macro-dev persons I have the pleasure to read. Cannot wait to read this!

[VoxDev @vox_dev]

📢Technology is central to economic development.

In this VoxDevLit on June 15th, @juli_caunedo (@CornellEcon) & @PorzioTommaso (@Columbia_Biz) will review the economic evidence on technology and development.

Register➡️https://t.co/ZQOzuEybDy

[Barry Eichengreen @B_Eichengreen]

https://t.co/0bbOWycY71

[Adrien Matray @AdrienMatray]

@startupecon sooooo it is not that clear. If your claude.md in your working directory has instructions that contradicts your global claude for instance, this can be a bit of a mess (definitely happened to me at least)

[Adrien Matray @AdrienMatray]

(1/N) Does your Claude sometimes ignore your preferences? Suddenly forget your rules? Run Stata the wrong way out of nowhere?

You are not alone. And it is almost always the same mistake.

[Adrien Matray @AdrienMatray]

@startupecon oh yes 100% this has been another source of conflict I had with Claude, it took a moment before I realized that my claude.md in projects would over-rules the "general claude.md". Now I have in place that everytime I create a project-specific claude.md it automatically checks that

[Adrien Matray @AdrienMatray]

I wrote a full guide for economists who have never used Claude. Mental model, setup, hard rules, FAQ on the Dropbox mistake, common failures.

Take a look and tell me what is still missing.

https://t.co/98FD3Hhbo7

[Adrien Matray @AdrienMatray]

Good news: you can force Claude to behave how you want.

Write your non-negotiables as Hard Rules in your global CLAUDE.md. Add preferences. Iterate when Claude breaks one. It takes back and forth, but Claude misbehaving is NOT a fatality.

[Adrien Matray @AdrienMatray]

@Haonan_Zhou Doesnt that impose for you to have the code in a completely separate folder outside dropbox? I did not want to take this route because it is common for people to work on multiple computers and hence having everything inside dropbox allows to do that easily

[Adrien Matray @AdrienMatray]

1/n ⚠️⚠️ how to use github and dropbox for economists. Economists are now told, for good reason, that if we use AI for research code, we should also use GitHub.
That advice is right. But the way many economists actually work creates a trap.

Most projects still live in Dropbox.

[Adrien Matray @AdrienMatray]

@abhishekn 1) github does not handle well very large datasets (typical in econ) 2) economists use dropbox and are used to it, this system allows to have coauthors who continue to use github :)

[Adrien Matray @AdrienMatray]

I wrote up the workflow here for research assistants and coauthors:
https://t.co/zcK0SWtoH7

The main point: Dropbox is the storage layer. GitHub is the collaboration layer. Mixing those roles is where mistakes happen.

[Adrien Matray @AdrienMatray]

And before editing anything, always ask Git what branch the folder is actually on:
git status --short --branch
Do not infer branch state from the Dropbox folder name alone.

The folder name is a label. Git is the source of truth.