Quiz Me

quiz-me

One-shot quiz. Write a deck, render it, open it. Nothing is saved.

quiz-me-full is the sibling that adds spaced repetition, a progress file, and an adaptive difficulty ladder - it calls the exact scripts in this skill to do the rendering, so improve the UI here and both get it.

The two steps

1. Write a deck file. Markdown, one file, no scaffolding:

# Deck: HTTP Caching
id: http-caching

## Why does `Cache-Control: no-store` exist when `no-cache` already does? [2]
- [ ] They are synonyms
- [ ] `no-cache` forbids storage; `no-store` forbids reuse
- [x] `no-cache` allows storage but forces revalidation; `no-store` forbids storage
> `no-cache` is a revalidation rule, not a storage ban - the response may sit on
> disk as long as it is re-checked before reuse. `no-store` is the one that keeps
> it off disk, which is what you want for anything sensitive.
@ RFC 9111 §5.2.2

[2] is difficulty 1-5 (default 3). - [x] is the correct option. > lines are the explanation shown after answering. @ is the citation. That is the format.

2. Render it.

python3 ~/.claude/skills/quiz-me/scripts/render.py <deck.md>
python3 ~/.claude/skills/quiz-me/scripts/render.py <deck.md> --cards 3,7,9

It writes a self-contained HTML file and opens it. No server, no dependencies.

Writing questions that actually gauge intuition

The one rule that matters most: never test what grep answers

Do not ask where something lives. "Which file holds the Skill model?", "which module imports this?", "what is this function called?" - all worthless. A reader recovers any of that in seconds, so testing it measures lookup speed, not understanding, and knowing it confers nothing.

Ask instead what a person cannot recover by grepping:

If a card can be answered by someone who has never thought about the system but is fast with a search box, cut it.

The five rungs

1. Purpose - why does this exist at all; what would be worse without it
2. Mechanism - how does it work end to end; what calls what, in order
3. Factors - what feeds a decision, and what deliberately does not
4. Consequence - what breaks, and what the trade-off buys
5. Design - a concrete proposed change, and what is wrong with it

The best single question to build a deck around is "how does X actually work?" for each major subsystem. The second best is "what is the point of Y?" - and Y can be a doc, a convention, or a process, not only code.

What to mine for

• Where the plausible answer fails. "Which store loses the bytes on unpublish?" Everyone says all of them. Only one does.
• The reason behind a rule, not the rule. Anyone can read the rule.
• Blast radius. "You delete X tomorrow - what breaks?" This separates knowing the diagram from understanding it.
• Duplication and drift. Anything stored twice, and which copy wins on read.
• Absences that carry meaning - a gate that is deliberately soft, a table that was dropped, a check that is deliberately missing. Not "this column does not exist", but "why was it removed".

Rules that keep a deck honest

• Every explanation carries a citation (@ file.py:120, an ADR, a doc section). A wrong answer has to be falsifiable, or the quiz is just your opinion.
• Cite from the code you actually read, not from the docs alone. Docs go stale; a file:line is checkable in seconds.
• Distractors must be things a reasonable person would believe. Four options where three are obviously silly tests nothing.
• The explanation must teach, not just confirm. A reader who missed the card should come away with the model, not just the answer.
• Spread the difficulty. All-hard is demoralizing, all-easy is flattering.
• 12-16 cards per deck unless it is a reference deck built for repetition.

What the result screen reports

• Score, and accuracy broken out by stated confidence.
• A calibration read - the interesting output. Confidently-wrong answers are named separately because they are the expensive ones: a confident wrong model is what gets designed against. Lucky guesses are named too, in the other direction.
• Every missed question repeated with its explanation and citation.

Where things go

Rendered HTML lands in ~/Desktop/quiz-progress/.rendered/. Decks are conventionally kept in ~/Desktop/quiz-progress/decks/, but render.py takes any path.

Files

The template renders both modes. quiz-me-full passes scheduled: true plus per-card interval previews; without those it is a plain quiz. The scheduler is never implemented in JavaScript - Python ships six pre-computed outcome strings per card so the two cannot drift.