--- name: quickstart description: Take today's quiz through the Quick Start Quiz app: are you current with Anthropic, and did you actually open the files written for a reason? Say 'quick start', 'take the quiz', 'check yourself'. --- # Quick Start Quiz — take today's quiz For a typed session with a person on the other side. A scheduled or background run skips this skill. 1. Start it: `node ~/.claude/skills/quickstart/proctor.js start` It prints each question, what to open for it, and the quiz id. The first run saves your quiz key and prints your score page. 2. For EVERY question, open the named thing with a visible tool call (head, grep, wc, ls, git log) and answer from what you just read. - Never from memory of another session, from the session brief, or from what you "know". - Never by computing a hash, and never by reading the proctor's code or its state folder. - Never answer "could not open" without a real try. A blank or "could not open" is graded wrong. - Line counts are `wc -l` counts. Quote text exactly as the file has it. - For an Anthropic question, read the app's answer key: `curl -s "$(cat ~/.claude/quickstart/url)/api/facts"`, or the live docs page the question names. 3. Answer, one entry per question id, every id filled, the JSON on stdin (a quoted heredoc keeps apostrophes safe): `node ~/.claude/skills/quickstart/proctor.js answer <<'EOF'` then `{"": "", "": ""}` then `EOF` 4. Read the person the report's "Say to the person" lines, plain, then stop. No paths, no ids. 5. A finding marked "fix" that is one line in your own config (a retired tool name in a hook matcher, a stray field in a skill's frontmatter): fix it now. Anything bigger: tell the person in one line and leave it. 6. Say this once, plainly: this proves the files were opened today, not that they were understood. The job is the habit and the drift alarm.