Level 1 — Get a Real Result in Twenty Minutes

8 questions, drawn from a bank of 33 across 22 topics. Pass at 80%. Every wrong option carries a note explaining why it is tempting — those notes are the point of the quiz, so read them whether you pass or not.

attempt 1

1 / 8 · Choose one
Your organisation requires model traffic to stay within its existing cloud provider. Which is a supported way to run Claude Code?
Your organisation requires model traffic to stay within its existing cloud provider. Which is a supported way to run Claude Code?
Point it at a self-hosted open-weights model of your choiceRoute it through Amazon Bedrock, Google Cloud's Agent Platform, or Microsoft FoundryRun it behind a VPN so requests appear to originate inside the networkThere is no supported route — Claude Code only talks to Anthropic directly
2 / 8 · Choose one
You configure a status line that prints context_window.used_percentage, so every turn now ends with a context reading at the bottom of the screen. A colleague warns you that a readout redrawn that often is itself eating the window it reports on. What is true?
You configure a status line that prints context_window.used_percentage, so every turn now ends with a context reading at the bottom of the screen. A colleague warns you that a readout redrawn that often is itself eating the window it reports on. What is true?
It is free, but it costs you the footer: a configured status line takes the built-in bar’s place rather than adding a row, so the badges that were there are goneIt costs nothing to the window: the status line is a shell script Claude Code runs locally, and it does not consume API tokensIt costs a little, but only when the value changes — an unchanged status line is not re-sentYour colleague is right — the script’s output joins the conversation each turn the way a tool result does, so the meter inflates the number it prints
3 / 8 · Choose one
A session produced a poor result. Which move is most likely to produce a better one?
A session produced a poor result. Which move is most likely to produce a better one?
Change the inputs — narrow the scope, name the file, state the symptom rather than your diagnosis, or start a fresh sessionRe-run the same prompt — the route through the codebase differs every time, so a second attempt reaches the file the first one walked pastAccept the result and fix it by hand afterwardsSend the same request again with more force behind it — 'this really matters, please be careful, and give it your best effort' — so the ask arrives with the urgency it deserves
4 / 8 · Choose one
You are inside a running Claude Code session and want the checkup claude doctor gives. What do you run?
You are inside a running Claude Code session and want the checkup claude doctor gives. What do you run?
/help, which prints the diagnosticsNothing — you have to quit the session and run claude doctor from your shellclaude --version, which covers the same ground/doctor, which runs the same checkup and can offer to fix what it finds
5 / 8 · Choose one
It is your first day on the team. You have started claude at the root of a repository you have never read, and what you need to know is how logging works. What do the docs say you have to do before you can ask?
It is your first day on the team. You have started claude at the root of a repository you have never read, and what you need to know is how logging works. What do the docs say you have to do before you can ask?
Nothing. Ask it the way you would ask another engineer on the team — 'How does logging work?' — and let the session do the readingWrite it a short brief first — the stack, the folder layout, the conventions the team follows — so it is not answering from nothingDress it up as a prompt — give Claude a role, spell out the steps to follow, and ask it to reason through them before it answersTrack down the files that look like they handle logging and paste them into your first message, so they are in the window before you ask
6 / 8 · Choose one
Which install method adds a prerequisite the others do not?
Which install method adds a prerequisite the others do not?
WinGet — it needs a specific PowerShell versionHomebrew — it needs a compiler toolchain to build from sourceThe native installer — it needs curl and a POSIX shellnpm — it needs Node.js 22 or newer
7 / 8 · Choose one
claude --version prints command not found. What is the useful next step?
claude --version prints command not found. What is the useful next step?
Fix the install or your PATH — nothing else you type in a terminal will work until this prints a versionRe-authenticate with /login — a --version check contacts Anthropic to confirm the install, so an unauthenticated machine cannot answer itInstall the VS Code extension instead — it puts claude on your shell PATH as part of its setup, so the terminal command starts working tooRun claude doctor — inspecting install health and reporting a broken PATH is exactly what that command is for
8 / 8 · Choose one
A colleague opens a session in an unfamiliar repo with "investigate this codebase and tell me how it works", watches it read for ten minutes, and is now about to type the question they actually came for. What has that ten minutes done to the question they are about to ask?
A colleague opens a session in an unfamiliar repo with "investigate this codebase and tell me how it works", watches it read for ten minutes, and is now about to type the question they actually came for. What has that ten minutes done to the question they are about to ask?
Every file the investigation read is still in the window, so the real question gets asked into a context full of code neither of them needed — and performance degrades as the window fillsNothing that lands on the question: the bill was ten minutes of waiting and the tokens the reading spent, and both of those are already paid — a cost you have settled does not make the next answer any worseIt helped: the reading is in context now, so the real question gets answered from files already read instead of from fresh readsIt cleared the way: the investigation’s reading was discarded when it finished, so the real question starts from an empty window

8 questions still to answer