draft_verification_questions v1.0.0

name: draft_verification_questions description: After any substantive draft, generate 2-3 follow-up questions that push the reader to check key facts, probe assumptions, and notice missing context. Use after writing or reviewing a draft article, guide, proposal, memo, or report.

Draft Verification Questions

Purpose

A finished draft looks done. It is not. Before it ships, give the reader three sharp questions that do the work the reader would otherwise skip: check the load-bearing facts, name what the draft takes for granted, and point at the context the draft never mentions. Questions, not verdicts. The reader does the verifying; you just aim the spotlight.

This skill is grounded in the Paul-Elder critical-thinking framework (Socratic question types: probing assumptions, probing reasons and evidence, questioning viewpoints; standards of clarity, accuracy, precision, relevance, depth, breadth, logic, significance, fairness). See references/question-patterns.md for the per-type patterns and worked examples.

Workflow

1. Read the draft and name the output type. Pick from: factual/explanatory, how-to/instructional, persuasive/opinion, data/quantitative, technical design/code, recommendation/advice. Mixed drafts get the type of their main load-bearing claim.

1. Find the load-bearing claims. Skim and mark the 1-2 claims that, if wrong, break the whole piece. These are your fact-check question candidates. Quote the draft's own wording and note where each claim sits (section, paragraph).

1. Surface the biggest unstated assumption. Ask: what does this draft take for granted that a skeptic would flag? Common species: the audience shares the author's context, the trend continues, the tool behaves as documented, the reader has the stated prerequisites, one case generalizes.

1. Scan for missing context. What would a competent, skeptical reader ask that the draft never addresses? Think: who disagrees and why, what happens when the advice fails, which alternatives exist, what has changed since the sources were written.

1. Generate exactly 2-3 questions, one per pattern. The three patterns, always:

• Fact-check: points at one load-bearing claim and asks how the reader could verify it. Names the claim, in the draft's own words.
• Assumption probe: names the unstated assumption and asks what breaks if it is false.
• Missing-context scan: points at the unaddressed angle and asks the reader to weigh it before accepting the draft.

For drafts where one pattern genuinely does not apply (rare), double up on the one that matters most, but always ship at least two different patterns.

1. Emit the Output Contract block and stop. Do not answer the questions yourself. Do not rewrite the draft.

Output Contract

Return a block in this shape:

**Verification questions**

1. [Fact-check] <one question, naming the specific claim and where it appears>
2. [Assumption probe] <one question, naming the assumption>
3. [Missing-context scan] <one question, naming what is absent>

Each question is a single sentence, plain language, no preamble. The whole block is 3 questions maximum, never more.

Operating Rules

• Specific beats generic. "Have you verified the statistics?" is useless. "Can you confirm the 40% figure in paragraph 3 against the original survey?" is the job.
• Never answer your own questions. The moment you answer one, it is no longer a verification question, it is your opinion wearing a question mark.
• No smuggled assertions. Do not embed unverified claims inside the questions themselves ("Given that the method fails at scale, ..."). If a premise is unverified, it does not belong in the question.
• Stay inside the draft. Questions must refer to claims and gaps in the draft itself, not to topics the draft never raised.
• Match the reader, not the author. Write for the person who has to trust or act on the draft: what do they need to check before they do?
• No ChatGPT speak. Plain, direct questions. No em dashes. No "in today's fast-paced world." No rhetorical-question-then-answer setups.