Before the day

Allow thirty minutes before the day, and use the setup clinic if anything fails. You work on your own laptop with your own agent, and your partner does the same. Once the agent is installed, it does the rest for you.

Setup for the hour

1. GitHub. Create a free account and accept the invitation to your pair’s repository. The invitation goes to the username you gave at sign-up, and it also shows at github.com/notifications.

2. Git. On a Mac, open Terminal, paste the line below and press Return. It is the one line you have to type anywhere in the preparation.

   xcode-select --install

   On Windows, install Git for Windows from gitforwindows.org with the default options.

3. VS Code. Install it from code.visualstudio.com and sign in to GitHub inside VS Code, with the Accounts icon at the bottom left. That lets you clone, commit and push without typing a password or a token.

4. Python 3.11 or later. Install it from python.org. On Windows, tick “Add python.exe to PATH” in the installer. The repository’s checks are small Python scripts.

5. One coding agent in VS Code. Claude Code (with a Claude Pro or Max subscription) or Codex (with ChatGPT Plus or Pro). Install the extension and sign in. If you have neither, you are paired with someone who has one: the first part of the hour needs no agent, and in the second part you use your partner’s agent under your own name.

6. Clone, then let the agent set you up. In VS Code, choose “Clone Git Repository” and paste your pair’s repository link. Pick a folder that is not inside iCloud, OneDrive or Dropbox. Open it, open your agent and say:

   Say to your agent

   Read AGENTS.md. Set my Git name to Your Name and my email to [email protected], set pull.rebase so the shared history stays in one line, then run the workflow check and tell me what it says.

   It should report that the mechanical checks passed. If it says Python is too old, go back to step 4.

7. Make a readiness commit.

   Say to your agent

   Read AGENTS.md. Create workspace/sandbox/<A or B>/ready.md with my name and today’s date, then commit it and push.

   If the agent shows you a draft record and stops, read it and say finalize. When the push succeeds you are ready, and your file shows on GitHub.

Without an agent

If your agent is not installed yet, open a terminal in VS Code (Terminal, New Terminal) and run these four lines instead of step 6. Step 7 waits until you have an agent.

git config --global user.name "Your Name"
git config --global user.email "[email protected]"
git config --global pull.rebase true
python3 scripts/check_workflow.py --out workflow-check.json

Setup for the six parts

Only needed if you want to try the six parts after the hour. Each part says what it needs, and none of them is needed for the hour itself.

OpenRouter, for the second model in Part A and the library search in Part B. Sign up at openrouter.ai, add a small amount of credit (a few dollars is far more than the workshop uses) and create a key. In VS Code, copy .env.example to .env and put the key after OPENROUTER_API_KEY=. That file is ignored by Git, so keep the key there and out of the chat.

Zotero, for Part B. Install Zotero and import the workshop’s reference pack (File, Import, then references/workshop-pack.ris in your clone), or use a library you already have. In Settings, Advanced, turn on the local API so your agent can read the library while Zotero is running.

Say to your agent

Read AGENTS.md. Check that you can read my Zotero library, and build its index.

LaTeX, for Part D. On a Mac install BasicTeX; on Windows install Basic MiKTeX. Then ask your agent to check that it can build a LaTeX page.

Overleaf, for Part E. You need an Overleaf plan with the Git integration, which means a paid plan or your university’s licence. In Account Settings, under Git integration, generate a token and put it after OVERLEAF_GIT_TOKEN= in .env. One of you creates a blank project named after your pair and shares it with the other.

Note

Nothing here installs a hook, and no check blocks a commit. The repository reports what it finds and keeps the states of the work apart. Bring the laptop you did this on.