VS Code

VS Code

Not a reasoning tool — the workbench everything else runs inside of.

What it is

• Free code editor (code.visualstudio.com).
• The Dev Containers extension gives the whole class an identical, reproducible environment — same Python, R, Node and Claude Code CLI versions for everyone.

Where it fits

Hosts every stage as the container/editor — D1·1 (setup) and D2·3 (editing analysis scripts) are where you actually interact with it directly.

Access & cost

Free. Download at code.visualstudio.com.

Option A — Dev Containers

• Open the repo, “Reopen in Container.”
• postCreateCommand installs Python/R packages and the Claude Code CLI automatically.
• Recommended — most reproducible across everyone's laptops.

Option B — GitHub Codespaces

Push the repo to GitHub, then “Code → Create codespace on main.” Same .devcontainer config, runs in the browser — nothing to install locally.

Option C — local, no containers

Install Python 3.11+, R 4.x, Node 18+ yourself, then pip install -r requirements.txt and npm install -g @anthropic-ai/claude-code. Less reproducible — the fallback if Docker isn't available.

D1·1 — environment check

The integrated terminal is where you run python scripts/check_setup.py, set ANTHROPIC_API_KEY, and start claude or codex for the first time.

Throughout Day 1 — the files are just files

AGENTS.md, every SKILL.md, and all six agent .md files are opened, read, and edited right here — there's no separate authoring tool for any of them.

D2·3 — the data lab

analysis.py / analysis.R are opened, edited (filling in the robustness TODOs) and run from VS Code's integrated terminal.

What it is not

VS Code does none of the reasoning itself. It's the workbench; Claude Code, Codex, and the browser-based literature tools do the actual work inside or alongside it.

Recap

Back to the home page for the full pipeline.