Agent setup
Start with VS Code + GitHub Copilot. The detailed setup covers installation, account access, cloning, language environments, and the first plot. The main lessons use Copilot Chat, Explorer, and Source Control throughout.
Prefer Claude Code? Its VS Code extension supports the same prompts and visual Git workflow. Other clients below are optional references. For a complete terminal workflow, use the CLI appendix.
| Client or task | Read next |
|---|---|
| VS Code + GitHub Copilot (default) | Detailed setup · Controls and optional features |
| Claude Code in VS Code (alternative) | Extension setup |
| Codex (optional) | Setup |
| Cursor | Setup |
| Gemini CLI | Setup |
| Instructions, skills, and configuration folders | Portable context |
| Modes and agent workflows | Working with coding agents |
| One-line R formatting request | Optional R refactoring skill |
Platform references reviewed: September 5, 2026, with later scoped dates on refreshed pages. Pages link to current installation/sign-in guidance; actual features depend on client, version, account, and local versus hosted execution. Inclusion here is not approval for university data. Check the lab data guidance.
Use the lesson sequence
| Your next step | Copilot role in a Local session |
|---|---|
| Understand the code | Ask: explain the source and cite files |
| Specify the repair | Plan: propose changes and checks for review |
| Implement and verify | Agent: edit, run checks, and show their results |
Select the actual role before each lesson prompt; writing “make a plan” does not itself show which role is active. Session target, role, model, and permissions are separate controls. Keep the Local target for the main path so edits appear in the open workspace. Other clients use their own planning and permission controls. VS Code session targets and roles
Check the agent’s understanding
With the example repository open, select Local → Ask in Copilot Chat and paste:
Read AGENTS.md and REPO_MAP.md. I'm using R. Where is the plotting function,
and how do I run and check it? Cite the files; don't edit or execute anything.
Replace R with Python if needed. Open the cited files and compare the proposed checks with your setup. Check the client’s loaded-instructions display where available; a plausible answer is not proof that the instructions loaded.
Let the first lesson guide your setup
Continue with Python Lesson 1 or R Lesson 1. Add the review skill when useful. Comparing clients, adding a reviewer agent, and connecting external tools are optional.
For a reproducible handoff, record the client/version, model, execution environment, and actual checks. Keep credentials and personal settings out of Git.