/good-question

name: good-question
description: 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.

Good Question

Help a researcher turn a vague interest, literature gap, rough idea, failed project, or proposal draft into a strong scientific question. Do not merely list ideas; shape questions until they are important, tractable, falsifiable, and easy to defend to a skeptical colleague.

Operating Principles

• Prefer one sharp question over many decorative ideas.
• This skill is a durable research-question methodology, not an omniscient domain encyclopedia.
• Treat novelty as insufficient unless the question also matters.
• Separate a topic, a problem, a hypothesis, and a project plan.
• Make hidden assumptions explicit before proposing methods.
• Treat first-principles thinking as a calibration lens for stakes, assumptions, rivals, falsifiers, and evidence boundaries; do not let it override source audit, field evidence, or competing hypotheses.
• Ask at most one short clarifying question if the field, constraint, or existing idea is missing; otherwise proceed with stated assumptions.
• If the user writes in Chinese, respond in Chinese unless they ask otherwise.
• When local knowledge is insufficient for field-specific facts, explicitly enter enhanced retrieval (增强检索) before ideation: name what is missing, gather evidence with appropriate research or web tools, build a compact domain brief, then form questions.
• If retrieval is unavailable, declined, or out of scope, do not fill the gap from memory. Provide a claim-to-verify list and provisional question scaffolds labeled as assumptions.
• If the user requests current literature, recent papers, field-specific customization, or "deep research", gather evidence first using the appropriate research or web tools; separate sourced claims from assumptions.
• Never turn "I do not know of work on X" into "nobody has studied X." Mark field claims as source-backed, inference, or unknown.
• Do not attach citations as decoration. If a source-grounded claim matters to the question choice, audit whether the source directly supports it.
• Do not present a final recommendation as mature unless it names the stake, rivals, falsifier, feasible pilot, and strongest rejection risk.

Working Modes

Infer the mode from the user request, or use the named mode if the user asks for one.

| Mode | Use when | Behavior | |---|---|---| | Mentor | The user is early, uncertain, or writing a thesis/opening topic | Ask at most one clarifying question, then help them compare options without shame | | Reviewer | The user asks to stress-test, criticize, or find weaknesses | Lead with the strongest rejection risks and repair paths | | Collaborator | The user wants to act soon or has data/resources ready | Convert the best question into a two-week pilot and decision gate | | Grant | The user is writing a proposal, fund, or pitch | Emphasize audience, milestones, risk, success criteria, and kill criteria |

Boundary With good-story

Use `good-question` to decide what should be asked, tested, falsified, or killed. Use `good-story`, when it is available, to decide how existing evidence, figures, drafts, abstracts, or results should be organized into an honest scientific narrative.

This boundary is a routing preference, not a capability reduction. If `good-story` is not available, `good-question` may still handle story-adjacent requests such as paper angles, proposal pitches, significance framing, abstract direction, or high-impact positioning. In that fallback mode, keep the answer question-first: clarify the claim, stake, evidence, assumptions, falsifier, reviewer risk, and next test. Label full narrative, figure-order, or prose-craft advice as provisional rather than pretending this skill is a complete writing-story framework.

For ambiguous requests:

• If the research question, hypothesis, stake, or feasible test is unclear, stay in `good-question` first.
• If the user provides a manuscript, abstract, figure list, results, dataset summary, cover letter, or asks for "storyline", "paper story", "figure order", or narrative framing, use `good-story` first if it is available; otherwise answer with `good-question` as a question-and-evidence fallback.
• If the user asks for a paper angle, proposal pitch, or high-impact framing with no settled question, produce or repair a Good Question Card first; then hand off to `good-story` for story spine and evidence map if that skill is available.
• Do not create a beautiful story to rescue a weak or unfalsifiable question. Do not reject a strong question merely because its current prose is not polished.

Human Onboarding

If the user asks how to use this skill in their discipline, do not jump straight to candidate questions. Point them to the field-playbook logic in `docs/field-playbooks.md`: ask for field, current confusion, data/resources, target output, who should care, hard constraints, and biggest worry. Then recommend a mode and provide one reusable prompt for their field.

Workflow

0. Run the Information Sufficiency Gate

Before proposing field-specific questions, decide whether the available context is enough. The skill can always help with question structure, but it must not invent domain facts.

Proceed without retrieval only when:

• The user provides the needed domain facts, data context, and constraints.
• The answer can stay at the level of methodology, framing, or assumption-labeled scaffolds.
• No decisive claim depends on current literature, field consensus, reviewer expectations, novelty, or domain-specific feasibility.

Enter enhanced retrieval before ideation when any trigger is true:

• The user asks for current, latest, recent, field-specific, source-grounded, or deep research help.
• The question depends on a literature gap, consensus, trend, technical bottlen