4 build prompts · 4 checks · run them in order and your second brain gains a coach that drills you, logs your misses, and tells you when to book
Every move from the closing demo lives here, consolidated and cleaned up. Each one says what it is for and what you should see when it lands. Open a terminal in your second brain repo, type claude, and paste them in order. Rephrasing in your own words works just as well. If a result does not match, ask Claude what it did differently, fix that step, and keep going. Before you start, put the official exam guide PDF in a study folder at the root of your repo; the first prompt reads it, and everything after depends on the file it writes.
The exam is written from the guide's task statement bullets, so the coach drills from those bullets and nothing else. This turns the PDF into one file with thirty sections the coach can read in a second. The guide's own wording stays; a summary would lose the exact phrases the exam reuses.
read the exam guide pdf in my study folder and write study/task-statements.md: one section per task statement, numbered the way the guide numbers them, with the knowledge bullets, the skill bullets, and any anti-patterns the guide names. keep the guide's wording; do not summarize
What you should see: study/task-statements.md with thirty numbered sections from 1.1 to 5.6, each carrying a Knowledge list and a Skills list in the guide's words. Spot check one: open 5.2 and confirm the line about sentiment and self-reported confidence being unreliable proxies is there verbatim. If sections are missing or paraphrased, ask Claude to reread the domain in question and rewrite those sections.
A skill that fires on "drill me," picks five task statements weighted toward your misses, asks one scenario question at a time in the guide's style, waits for your answer, then explains and cites the bullet. Every miss goes to a log with its reason. The waiting is the mechanism: recalling an answer under a little pressure is what makes it stick, and a coach that reveals the answer before you try skips the one step that makes it stick, the effort of trying to recall it yourself.
create a skill called exam-coach. when i say drill me, quiz me, or name a domain or task statement, it reads study/task-statements.md and study/mistakes.md, picks five task statements weighted toward ones i have missed, and asks me one scenario question at a time in the exam guide's style: a production symptom, four options, one correct. it waits for my answer before revealing anything. after each answer it says correct or wrong, explains in three sentences, and cites the task statement bullet. every wrong answer gets one line in study/mistakes.md: date, task statement, what i chose, why it is wrong, the bullet. create the mistakes file with a header
What you should see: .claude/skills/exam-coach/SKILL.md with a name and description in the frontmatter and a numbered procedure below, and a new study/mistakes.md with a header and no entries. Read the description: it should name the trigger phrases, because the description is the only part Claude sees at session start.
A second skill that reads only the log, ranks task statements by misses, and drills the top three with fresh questions. A statement answered correctly twice in a row across sessions is marked recovered and leaves the top three. It stays in the file. The file is your study history, and three weeks from now the exam tests exactly what is in it.
create a skill called weak-spots. when i say weak spots or drill my misses, it reads study/mistakes.md, counts misses per task statement, takes the top three, and runs the exam-coach procedure on those three only, two fresh questions each, never a question already in the log. a statement answered correctly twice in a row across sessions gets a recovered note in the log and leaves the top three
What you should see: .claude/skills/weak-spots/SKILL.md, short, referring to the exam-coach procedure rather than restating it. If the log is still empty, the skill should say so and offer a normal drill instead; that is the correct behavior, and the third check below fills the log.
The skill that decides whether you book. It takes five domain percentages, reads the log for statements missed twice without a recovery, and writes a readiness file with a GO or NO GO per domain against 85 percent, an overall verdict that is GO only when every row is, and the exam day checklist. The exam is weighted by domain and does not grade on your average; neither does this file.
create a skill called readiness. when i say am i ready or paste my domain scores, it takes the five percentages, reads study/mistakes.md for statements missed twice or more without a recovered note, and writes study/readiness.md: each domain with its score and GO or NO GO against 85 percent, the open statements, an overall verdict that is GO only if every domain is GO and no statement is open, and an exam day checklist. if NO GO, name what to drill and stop
What you should see: .claude/skills/readiness/SKILL.md. Nothing else yet; the file it writes appears in the last check. The checklist it will produce should include: booked, identification ready, room and connection tested for online proctoring, sleep, two minutes a question, flag and move on.
Fresh matters because skills load their descriptions at session start. Close Claude Code, open it again inside your repo, and run these in order; the second one gets an answer wrong on purpose.
The coach's whole value is in asking before it tells. Prove it holds.
drill me on domain 5
What you should see: one question, four options, and a stop. If Claude reveals the answer in the same message, or asks all five at once, the skill body is being skipped: ask Claude whether it used the exam-coach skill, and if the answer is no, check that the description in SKILL.md names "drill me" and that the file is at .claude/skills/exam-coach/SKILL.md.
Answer the next question with the option you know is wrong, then open the log.
show me the last line of study/mistakes.md
What you should see: one line with today's date, the task statement number, the option you chose, a reason, and the bullet it tests. If the line is missing, the logging step did not run; ask Claude to append it now and to confirm step five of the skill says "every wrong answer."
Miss two more questions across different statements first, so the log has something to rank. Then:
drill my weak spots
What you should see: a short count of misses per task statement, then questions only on the top three statements, two each, and none of them a question you have already seen. A question on a statement with no misses means the skill ran the general coach instead; ask it to reread the log and restrict itself.
Paste scores with one domain under the line and a strong average, and read the verdict.
am i ready? domain scores: 1: 78, 2: 91, 3: 88, 4: 86, 5: 90
What you should see: study/readiness.md with a five row table, domain one marked NO GO, the open statements from your log, an overall NO GO, and a short list of what to drill. The average of those scores is above 85; a GO verdict means the skill is averaging, and the exam does not. Fix the skill's step three so the verdict requires every row.
Back to the Module 10 lesson, or straight to the mock exam.