Claude Code in VS Code
for Academic Economists

Setup

Installation in 5 Steps

Installing Claude Code in VS Code and getting started is straightforward.

1. Open VS Code and open the Extensions panel by pressing Ctrl+Shift+X.
2. Search "Claude Code" and install the official Anthropic extension.
3. Click the Claude icon () in the left sidebar or on the top right of a window pane.
4. Sign in with your Claude account or API key.
5. Open your full project folder in VS Code (File → Open Folder), not just a single file. Claude needs the whole directory tree to be useful. This can be any folder on your computer — a git repository, a Dropbox folder, or a plain folder on your desktop all work equally well.

Next

After Installing — What to Do Next

Claude Code works out of the box, but a few one-time setup steps make it dramatically more useful for empirical research. Everything below is covered in detail in the sections that follow — this is the short roadmap.

1. Install the extensions for your research stack. Stata, Python, LaTeX, CSV viewers — Claude reads what's installed and can run code, render tables, and preview PDFs directly in the editor.
2. Create a CLAUDE.md file for each project. Tell Claude once about your data, sample definitions, and coding conventions. It reads this at the start of every session so you never have to repeat yourself.
3. Add skills for the workflows you repeat. Turn a standard robustness check, a referee-style review, or a slide-deck build into a one-line slash command you can invoke in any project.

Recommended Extensions

A Starter Pack for Economists

Install these alongside Claude Code to cover the full research stack. Search for each by the extension ID in the Extensions panel (Ctrl+Shift+X).

Stata

Python

LaTeX

Data Files

"latex-workshop.latex.autoClean.run": "onBuilt",
"latex-workshop.latex.clean.method": "glob",
"latex-workshop.latex.clean.fileTypes": [
  "*.aux", "*.bbl", "*.blg", "*.log",
  "*.out", "*.toc", "*.fls", "*.fdb_latexmk",
  "*.nav", "*.snm", "*.vrb"
]

File Formats

What Claude Code Can Read — and What to Convert First

Claude Code works best with plain text. The more a file looks like characters-on-a-page (as opposed to a rendered binary), the better Claude can read, edit, and reason about it. For economists this has practical consequences: a referee report in .docx or a codebook as a scanned PDF are both usable, but giving Claude a text version of the same file produces noticeably better results.

A markdown file (.md) is the simplest useful form of this: a plain-text document with lightweight formatting conventions — # for headings, *italic*, **bold**, bullet lists, and links written as [text](url) — that renders into nicely formatted output but stays fully readable and editable as raw text.

A rough guide to how Claude handles the formats you're most likely to encounter:

Handling PDFs

PDFs are where most researchers get stuck. They look like documents, but under the hood they are a layout format, not a text format — the "text" is often a sequence of glyphs positioned on a page, with no inherent notion of paragraphs, columns, or tables. Claude Code can read PDFs, but results vary enormously depending on how the PDF was produced.

A practical rule of thumb:

For converting PDFs to markdown, a few options work well:

Customisation

CLAUDE.md and Skills

CLAUDE.md — Persistent Project Context

A CLAUDE.md file at your project root is read automatically at the start of every Claude Code session. It's the place to tell Claude things that are always true about your project — conventions, variable naming, dataset structure, what not to touch — so you don't have to repeat them in every prompt.

The easiest way to create one is to type /init in the Claude Code panel. Claude will read your project folder and generate a draft CLAUDE.md automatically — inferring file structure, languages, and conventions. Review and edit the result before using it. Alternatively, create the file manually. A useful starting point for an empirical project:

# Project: [Paper title]

## Data
- Unit of analysis: firm-year panel, 2010–2022
- Raw data lives in data/raw/ — never edit these files
- Main dataset after cleaning: data/clean/panel.dta (N ≈ 47,000)

## Code conventions
- R scripts are numbered (01_clean.R, 02_analysis.R, ...)
- All regressions cluster SEs at the firm level
- Use fixest for all panel regressions

## LaTeX
- Main file: paper/main.tex
- Tables generated by 03_tables.R go to paper/tables/
- Use \widehat{} not \hat{} for estimators

CLAUDE.md files work at two levels, and both are loaded simultaneously:

When you open a project, Claude reads both files. The folder-level file takes precedence if there's any conflict.

Skills — Reusable Workflows as Slash Commands

Skills are markdown instruction files that become /slash-commands you can invoke in the Claude panel. When you type /skill-name, Claude loads the instructions and executes them — useful for repetitive tasks you run across projects, like a standard robustness check or reformatting a regression table.

Skills live in one of two places:

Each skill is a folder containing a SKILL.md file. The folder name becomes the slash command. To create a /robustness command:

mkdir -p ~/.claude/skills/robustness
# then create ~/.claude/skills/robustness/SKILL.md

A minimal SKILL.md for an economist:

---
name: robustness
description: Run standard robustness checks on the main regression.
---

For the main specification in this project:
1. Re-run with alternative clustering (industry-year instead of firm)
2. Re-run dropping the top and bottom 1% of the outcome variable
3. Re-run on the pre-2020 subsample only
4. Produce a summary table comparing coefficients across all specs

Downloading Skills Others Have Written

There is a growing community library of shared skills. Copy any SKILL.md folder into ~/.claude/skills/ and it becomes available immediately as a slash command.

For economists specifically, a few starting points that are already in use in the field:

Version Control

Git Integration

Claude Code is git-aware. When you open a project that is a git repository, Claude can see your commit history, staged changes, and diffs — without you having to paste anything. This makes it useful for version control tasks alongside coding ones.

Write a commit message for my staged changes.

Summarise what changed between the last two commits 
in 02_analysis.R.

I have a merge conflict in 03_tables.R. Read both 
versions and resolve it, keeping the more recent 
variable names.

Performance

Managing the Context Window

Claude can only hold a limited amount of text in its working memory at once — roughly equivalent to a few hundred pages. For large empirical projects with many scripts and data files, this fills up quickly, and performance degrades as it approaches the limit. A few habits help:

@02_analysis.R fix the clustering in the main spec.