1How It Works & Mechanism
Pure prompt skill: Rewrites vague questions into agent-reasoning-ready, critique-ready, and verifiable problem specifications, assessing the degree of automation feasibility.
skills/dbs-good-question/SKILL.md你是 dontbesilent 的好问题生成器。
No scripts inside the directory; behavior occurs entirely on the model output side.
Core mission is treating questions as reasoning constraints: good questions compress the search space, expose conflicts, and point toward testable explanations.
skills/dbs-good-question/SKILL.md**核心使命：让问题承担推理约束。**
The original text explains the significance: the clearer the question, the better an agent can generate candidate explanations that are 'hard to vary'; vagueness forces reliance on default assumptions.
Principle 1 requires pinning down the phenomenon first: rewrites broad questions like 'Why am I not doing well?' into observable, concrete phenomena.
skills/dbs-good-question/SKILL.md不要直接回答「为什么我做不好」「为什么没人买」「这个能不能做」这类大问题。先把它钉成一个可以观察的现象。
Example of a good question provided is an observation backed by metrics (e.g., 'Over the past 30 days, 80 people inquired in the private channel, but only 2 made a purchase').
Principle 2 treats 'conflict' as the source of a question's power, listing five common types of conflicts (data, behavior, expectation, resource, constraint).
skills/dbs-good-question/SKILL.md问题的力量来自冲突。没有冲突，Agent 只能泛泛分析。
No conflict = no discriminative power; this serves as the rationale for refusing to jump straight into explanation generation.
Principle 3 defines a constraint field across five categories (target, goal, variable, constraint, feedback) as the skeletal structure of the problem specification.
skills/dbs-good-question/SKILL.md1. 对象：到底分析谁、哪件事、哪个场景。
The Phase 4 specification template expands precisely on these five categories (problem to analyze, phenomenon, goal, core conflict, background facts, constraints, feedback entry point).
Phase 2 evaluates five inspection points on a 0–2 scale (target, goal, conflict, constraint, feedback), but hides the scoring table from the user by default.
skills/dbs-good-question/SKILL.md对外输出时，默认不要展示完整评分表。
Defaults to showing only 'Current clarity: Low / Medium / High' and 'Largest gap: {single sentence most impacting agent reasoning}'.
Phase 3 uses six dimensions to evaluate agent solvability, outputting one of four tiers A/B/C/D (Highly automated / Semi-automated / Assisted reasoning / Not suitable for now).
skills/dbs-good-question/SKILL.md- **A 档：可高度自动化**。Agent 可以直接执行大部分流程。
Six dimensions = clear boundaries, expressible variables, available feedback, testable explanations, executable actions, stable patterns.
Principle 6 requires providing actionable footholds before auditing: rather than scoring first, it highlights breakpoints and clarifies how strong an explanation can currently be provided.
skills/dbs-good-question/SKILL.md### 原则 6：先给抓手，再做审计
Even if information is incomplete, it permits providing 1–2 low-confidence candidate explanations marked as 'Pending verification'.
Hard constraints on candidate explanations: each must contain mechanism, observable signal, exclusions, and action changes, up to a maximum of 3.
skills/dbs-good-question/SKILL.md候选解释不超过 3 个。
'- Mechanism: How A leads to B' is one of its fixed fields; low-confidence tier only provides 1–2 items, avoids large tables, and makes no definitive conclusions.
Principle 5 'Do not fake certainty': When information is insufficient, state what is missing first, then provide minimal follow-up questions or minimal observation actions.
skills/dbs-good-question/SKILL.md信息不足时，不要硬凑解释。
Unknown items in the specification template are explicitly required to be labeled as 'Unknown' rather than fabricated ('Unknown items must be marked as "Unknown"; do not hallucinate settings merely for format completeness.').
Output formats are categorized into five tiers A–E (Default output, Strict audit, Solvability, Full specification, Candidate explanations & critique), selected based on user request.
skills/dbs-good-question/SKILL.md### 格式 B：严格问题质量审计
Default output is constrained within 5 sections ('6. Control length. Default output should not exceed 5 sections; expand the scoring rubric, full specification, or explanation comparison table only when the user follows up.').
2Core Capabilities
01Input classification (vague question / phenomenon / materials / automation request / mixed)
02Good question 5-point audit and 0–2 scoring (hidden by default)
03Agent solvability 6-dimensional evaluation with A–D 4-tier conclusions
04Problem specification rewriting (7 fields + 4-step instructions for Agent)
05Candidate explanations and 'hard to vary' critiques (mechanism / exclusions / verifiable signals / action changes)
06Low-confidence candidate explanations (providing footholds when info is incomplete but distinct breakpoints exist)
07Minimal next-step decision matrix (action proposals across loose / medium / clear / automated categories)
08Output length and language discipline (default <= 5 sections, maximum 3 questions)
6Verdict & Scenarios
Provides footholds before auditing: skips displaying scoring rubrics by default, delivering breakpoints and clarity levels first to avoid an interrogation-like experience.Structures problems into actionable deliverables: maps the five constraint field categories directly to specification fields, enabling direct downstream agent execution.Falsifiable candidate explanations: requires four elements (mechanism, observable signal, exclusions, action change) capped at 3 candidate options.Actionable automation assessments: six-dimensional scores map directly to A–D tiers, identifying human-in-the-loop and feedback loop placements.Rejects fake certainty: explicitly states missing information when data is incomplete, strictly banning hallucinations for unknown values.Best for: Ideal for 'having an issue but unable to articulate it clearly': high bookmarks but zero inquiries, stagnant course sales, or wanting to automate a workflow without knowing where to start. Use it to pin down confusion into metric-backed phenomena, rewrite them into constrained problem specifications with feedback loops, and determine whether the task can be delegated to an agent.
Not for: Not suitable for users who already possess a precise problem definition (it will stop at 'high clarity, providing candidate explanations'); not suitable for execution-only requests needing immediate direct conclusions (it offers candidate explanations and minimal validation actions, not end-to-end execution); not suitable for fact-checking or data collection scenarios (handled by host tools and other skills); not suitable for psychological or emotional venting queries.