# Good Question > Use when a researcher is choosing, framing, refining, or stress-testing a research question, hypothesis, thesis topic, project idea, grant direction, paper angle, or stalled research direction. - Skill: `lingwei-zheng/good-question` (Agent Skill, multi-file: 48 files) - Install (CLI): `npx skillmds@latest add lingwei-zheng/good-question` - Raw SKILL.md: https://api.skillmd.com/api/skills/lingwei-zheng/good-question/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Research & Search - Author: lingwei-zheng (https://skillmd.com/u/lingwei-zheng) - Updated: 2026-09-21 - Page: https://skillmd.com/skills/lingwei-zheng/good-question --- # Good Question Turn a vague interest, literature gap, rough idea, failed project, or proposal draft into a question that is important, tractable, falsifiable, and defensible. ## Core Rules - Prefer one sharp question over many decorative ideas. - Separate the topic, problem, hypothesis, and project plan. - Treat novelty as insufficient unless the question also matters. - Separate novelty from contribution. Direct replication, regional extension, and modest data, measurement, implementation, or method improvements are valid when their value and venue fit are stated honestly. - Do not reject a question merely because it resembles prior work. Identify what becomes more credible, transferable, measurable, or useful. - Make hidden assumptions explicit. Name competing explanations when the question is explanatory or causal and the alternative could materially change the decision; do not invent rivals for descriptive or bounded incremental work. - Do not let a preferred method create the research problem. - Treat first-principles reasoning as a calibration lens for constraints, assumptions, rivals, falsifiers, and evidence boundaries. It cannot override field evidence, source audit, domain norms, or competing explanations. - Never turn "I do not know of work on X" into "nobody has studied X." - Do not recommend a mature question without a stake, falsifier, feasible pilot, and a clear advantage hypothesis: what this study could make more credible, useful, measurable, transferable, or efficient. - Do not make the strongest reviewer objection a mandatory author-side output. Surface it in `reviewer` or `grant` mode, for high-cost irreversible choices, or when a material flaw changes the decision. - Respond in Chinese when the user writes in Chinese unless asked otherwise. Use the Research Continuity record in `../shared/advantage-led-research-narrative.md`: preserve question, advantage state, fair comparison, claim/experiment IDs, evidence version, and retellable sentence across stages. Update only fields affected by the current task; reuse project records rather than starting another ledger. ## Boundaries - Use `good-question` to decide what should be asked, tested, falsified, or stopped. - Use `good-story` when the question is defensible and the task is to organize existing evidence, figures, results, or a manuscript into a scientific story. - Use `academic-advisor` for an integrated proposal or pre-submission decision that includes evidence grounding and journal strategy. - Do not create a persuasive story to rescue an unfalsifiable question. ## Modes | Mode | Use when | Emphasis | |---|---|---| | `mentor` | Direction is early or uncertain | Compare options and identify the most promising advantage | | `reviewer` | The user wants criticism or stress testing | Rejection risks and repair paths | | `collaborator` | Data or resources are ready | Pilot, milestones, and decision gates | | `grant` | Proposal, fund, or pitch | Audience, success criteria, and kill criteria | ## Workflow 1. **Check information sufficiency.** Read [references/question-workflow.md](references/question-workflow.md) and `../shared/research-calibration.md` plus `../shared/advantage-led-research-narrative.md`. Retrieve current evidence before field-specific ideation when novelty, literature consensus, reviewer expectations, journal fit, or technical feasibility affects the recommendation. When the user invokes first principles, fundamental assumptions, or root constraints, also read [references/first-principles-lens.md](references/first-principles-lens.md). 2. **Diagnose the starting point.** Identify the field, current idea, available resources, target output, hard constraints, and largest uncertainty. Ask at most one short clarifying question when an essential input is missing. 3. **Generate candidates.** Use a small mix of importance, assumption challenge, strong inference, boundary probing, structural analogy, simplicity, and stakeholder lenses. Keep candidates comparable and falsifiable. 4. **Converge.** Score importance, feasibility, falsifiability, evidence leverage, contribution value, venue fit, and downside learning. Classify the contribution level instead of demanding maximal originality. Drop questions only when they fail a fatal rule. 5. **Test finalists.** Name the advantage hypothesis, the fair comparison or decision arena, a discriminating test, and what evidence would kill the idea. Add competing hypotheses when the claim requires them. Add the strongest reviewer objection only when the active mode or a material decision requires it. 6. **Deliver a decision.** Produce one to three Good Question Cards and, when the user wants execution, a short pilot with a decision gate. ## Evidence Discipline - Use `Source-backed`, `Inference`, and `Unknown` when field claims influence the choice. - If retrieval is unavailable, provide a claim-to-verify list and label proposed questions as provisional. - Load [references/source-audit.md](references/source-audit.md) when a decisive claim depends on current literature, a target journal, or reviewer norms. - Do not attach citations that do not directly support the claim they accompany. ## Reference Router - [references/question-workflow.md](references/question-workflow.md): information gate, scoring, kill rules, card schema, onboarding, and response shape. - [references/alon-problem-choice.md](references/alon-problem-choice.md) and [references/fischbach-problem-picking.md](references/fischbach-problem-picking.md): problem choice, research taste, and method-first traps. - [references/platt-strong-inference.md](references/platt-strong-inference.md): mechanisms, competing hypotheses, and decisive tests. - [references/first-principles-lens.md](references/first-principles-lens.md): distinguish constraints, assumptions, evidence, inference, and unknowns without bypassing field evidence or source audit. - [references/problematization.md](references/problematization.md): assumption challenges and theory-oriented questions. - [references/heilmeier-catechism.md](references/heilmeier-catechism.md): grants, proposals, milestones, and risk. - [references/hamming-nielsen-research-taste.md](references/hamming-nielsen-research-taste.md) and [references/peters-question-development.md](references/peters-question-development.md): broad direction and literature-to-question development. - [references/orchestra-lenses.md](references/orchestra-lenses.md): rapid high-value ideation lenses. - [references/domain-brief-template.md](references/domain-brief-template.md) and [references/domain-adapters.md](references/domain-adapters.md): current field-specific grounding. - [references/question-patterns.md](references/question-patterns.md): convert topics, gaps, methods, and activities into questions. - [references/editor-desk-reject.md](references/editor-desk-reject.md): conditional skeptical gate for `reviewer`, `grant`, pre-submission, or high-cost decisions; not a default final ritual. ## Output Lead with a brief diagnosis, then the evidence status when needed, ranked candidates, one to three Good Question Cards, and the next action. Include repair or rejection notes only when they affect the decision. Keep the tone constructive and opportunity-seeking without relaxing the falsifier or evidence standard.