Quiz — Post-Work Comprehension Check (Stay in the Loop)

This skill serves what the user does not know and you do, and surfaces the user's gaps.

When you and the user both seem to understand the request, do it without this method: if its result carries weight (it ships, is sent or is hard to undo), first read back in 2–3 lines the request as you understand it, the assumptions you are making and what outcome counts as done, as the whole reply, and do it once the user confirms; a self-evident request (fix a typo) needs no read-back. When the read-back exposes a mismatch, route by who does not know: the user → the teach-me skill; you → the interview skill; neither of you → the blindspot skill. A misread weighty request costs the work built on it; a read-back costs one exchange.

Even when the work was delegated, the user must be able to explain the result: they are the one who presents it, answers for it and acts on it. A request for a quiz on work you largely produced is never clear to both of you by itself, because whether the user can explain the work is what the quiz finds out, so run the Procedure.

Procedure — Part 1: Explain First

Before the quiz, explain concisely, anchoring each explained part with Where: its location in the work (file:line for code; section, page or row otherwise):

1. Overall structure of the work (what was created or changed, where)
2. Top 3 design decisions and why
3. Most likely failure points (if something is wrong, where it surfaces first)
4. What the user must verify themselves (what you cannot guarantee)

Without a tool, do the step in the conversation and say what could not be checked.

Procedure — Part 2: Six-Question Gate

1. Prepare 6 questions; pass = all 6 correct.

   Question priority, highest first:

   Priority	Question type	Example
   1	failure response	"X goes wrong — where do you look first, what do you do"
   2	design rationale	"why B, not A"
   3	behavior prediction	"given this input, what result"
   —	rote recall	avoid

   Each question is a call the user would have to make when something goes wrong or someone questions the work, not trivia.

   If the deviation log (IMPLEMENTATION_NOTES.md by default for code work) exists as a file, or is restated in this conversation, make at least 1 of the six questions about its recorded deviations.

   a. Five are multiple choice. Number their options, and keep answer keys hidden until the user has answered: format every option the same way, with no bold, tick or other mark on the correct one. Give every option of a question similar length and detail, so the key cannot be read from the options' form: when the correct option carries a cause, a consequence or a remedy, give each distractor one too.

   b. The sixth is a teach-back in plain text, not a choice: "Explain to a reviewer, in your own words, <the work's riskiest decision>."

   c. Commit to the 2–3 points (and change sites) a correct teach-back explanation must mention before reading the answer: write them to .unknowns/<YYYY-MM-DD>-quiz-<slug>-key.md before asking. Keep the key out of the questions and reply until the user has answered.

   Picking the right option only shows recognition; saying it unprompted is what the user will have to do in the handoff, the meeting or the PR. Writing the key file before asking and quoting it when grading makes the key provably predate the answer.

2. Ask the prepared questions. Ask the five multiple-choice questions as 4 + 1, splitting a batch across calls when the question tool accepts fewer questions. Ask the teach-back separately in plain text. Ask for the answers by question and option number, with no sample answer line: a sample made of option numbers can match the key.

3. Grade answers.

   a. Grade the teach-back against the 2–3 points in the key file from step 1c, and quote that file when grading.

   b. For each wrong or blank answer, name the exact change site to re-read (file:line, section or page). If the change site is unavailable, mark it unverified.

4. For each wrong or blank answer, re-explain only those parts, then offer a retry: one new question of the same type on each missed part only, prepared, asked and graded by the rules of steps 1–3, not a new six-question set. A retry that gets every missed part right brings the score to 6/6.

5. Close the gate by score:

   Score	What to do
   6/6	Passed. When the work needs reviewer approval, suggest the buy-in skill.
   Below 6/6	The result stays "not yet" — list the change sites to re-read and offer the retry from step 4.

Output

After closing the gate in step 5, assemble the answers + teach-back + final score + change sites to re-read as copyable text.