Prompt Lab
Reading about prompting teaches you the vocabulary. Writing a prompt, running it against a real model, and seeing it graded teaches you the skill.
1. Your API key
This lab calls the Anthropic API directly from your browser using
your key. The key is held in this page only, is sent to
api.anthropic.com and nowhere else, and is never transmitted to AI School.
You pay for your own usage, which is typically a fraction of a cent per attempt.
Read this before pasting a key. Anthropic requires an explicit
opt-in header for browser-side calls precisely because a key in a web page is easier to
leak than one on a server. Use a key created for practice, give it the lowest spend limit
your account allows, and revoke it when you are done. Never use a production key here.
"Remember" stores it in this browser's local storage, which any script running on this
origin could read - leave it off on a shared machine.
2. The task
Pick an exercise. Each one has a rubric that a second model call grades your attempt against.
3. Your prompt
Write the prompt you would send. It is used verbatim as the user turn - the lab adds nothing to it.
Why bring your own key. AI School is free and has no budget to pay for
other people's model calls, and proxying them through our server would mean us holding a
key that anyone could spend. Your key, your account, your spend limit - and the practice is
against the same API you will use at work rather than a simulation of it. The grading call
uses structured outputs so the rubric result is a validated object rather than prose we
have to parse.