quiz-me

Quiz Me

Adversarial comprehension testing. Reads a document silently, then grills the user one question at a time to verify they actually read it — not skimmed it, not got the gist, actually read it. Delivers a pass/fail verdict when done.

This is not teaching. This is not alignment-building. This is a test. The user claims they read something; this skill finds out if that's true.

Process

Phase 1: Silent Document Ingestion

1. Resolve input files.

   • If $ARGUMENTS contains a file path or glob, read those files.
   • If $ARGUMENTS is empty, find all alignment HTML files (docs/**/*.html) and spec documents (specs/**/*.md) in the current repo. If multiple files are found, ask the user which document(s) to quiz on.
   • Supported formats: HTML, Markdown, plain text.
   • If no readable document is found, stop and tell the user.

2. Analyze document structure internally.

   • Identify major sections, key decisions, trade-offs, cross-section relationships, specific constraints/numbers/names, and non-obvious details.
   • Build an internal question plan that covers every major section, biased toward:
     • Cross-section relationships ("How does X in section 3 affect Y in section 7?")
     • Implications and trade-offs ("What happens if the migration in step 2 fails?")
     • Specific details that skimmers miss (exact constraints, numbers, edge cases)
     • Intent questions ("Why was approach A chosen over approach B?")
     • Synthesis questions ("Summarize the three most important trade-offs.")
   • Do NOT surface any of this analysis to the user.

3. Open the quiz. Tell the user which document(s) they're being quizzed on and that you'll ask one question at a time. Nothing else — no summary, no encouragement, no preview of what's coming.

Phase 2: Adversarial Questioning

Ask one question per turn. Non-negotiable.

Use AskUserQuestion for every question. Provide 2-3 plausible multiple-choice options plus the implicit "Other" free-text option. At least one option should be a trap — superficially plausible but wrong if the user actually read the document. Vary whether the correct answer is a named option or requires free-text via "Other".

Question types (mix throughout, do not cluster by type):

• Relationship: "Section X describes [thing]. How does that interact with [thing from section Y]?"
• Implication: "If [condition from the doc] changes, what breaks?"
• Detail trap: "What specific [constraint/number/name] does the doc specify for [thing]?"
• Intent: "Why does the doc recommend [approach] instead of [alternative]?"
• Synthesis: "What are the most important trade-offs the document makes?"

After each answer, evaluate strictly:

• Correct and specific: Acknowledge in one sentence. Move to next question.
• Vague or incomplete: Push back. "That's too vague. What specifically does the doc say about [X]?" Do not accept hand-wavy answers. Re-ask or ask a targeted follow-up.
• Wrong: State what's wrong. Give the user one chance to correct. If they fail again, note the gap and move on.
• Suspiciously perfect verbatim recall: Ask a follow-up that requires synthesis, not recitation. If they can synthesize, accept it. If they can only parrot, mark the section as weak.

Track per-section comprehension internally. Do not show the user a running score.

Phase 3: Termination

Continue until one of:

• The user has demonstrated understanding across all major sections — no section left untested.
• The user explicitly gives up or asks to stop.

Do NOT terminate early because the user got a few questions right. Cover the full document. Minimum 5 questions for short documents, more for longer ones.

Phase 4: Verdict

When the quiz ends, deliver:

1. Pass or Fail. Binary. Pass means the user demonstrated deep comprehension across all major sections. Fail means one or more sections had weak or missing comprehension.

2. Per-section summary:

   • Sections with strong comprehension
   • Sections with weak or missing comprehension
   • Specific gaps identified

3. If fail: List which sections to re-read and why.

Constraints

• Read-only. Do not write, create, modify, or delete any files.
• No commits, no pushes, no file mutations.
• No external requests. All content is local.
• One question per turn. Always.
• Use AskUserQuestion for every question. Do not ask questions via plain text output.
• Adversarial but not hostile. Push back firmly on bad answers. Do not insult the user. Do not congratulate excessively on correct answers.
• No teaching. Do not explain document content during the quiz. If the user gets something wrong, state what's wrong — do not deliver a lesson.
• No hints. Do not help the user arrive at the right answer. This is a test, not a tutorial.

Differences from taste-calibration

	quiz-me	taste-calibration
Purpose	Test comprehension of existing docs	Build alignment for new work
Tone	Adversarial examiner	Socratic collaborator
Input	Existing document to test against	Rough idea or problem
Output	Pass/fail verdict	Shared understanding summary
Recommended answers	Never	Always
File output	None	None

Default Shipping Contract

Read-only — do not create or modify tracked repository files. Follow the shared shipping contract convention in CLAUDE.md for next-step routing only.