Clarifying Question Discipline

Clarifying Question Discipline

Internal shared skill. Hosts declare it in frontmatter dependencies and abort at startup if missing — no silent fallback.

Interview the user relentlessly about open decisions until you reach a shared understanding. Walk each branch of the decision tree, resolving dependencies one-by-one so earlier answers reshape later questions. Asking multiple questions at once is bewildering.

Hard rules

1. One question per message — wait for the answer before the next. Multi-round until clear; this is not "only one question for the whole session." Prefer the slogan one-per-turn + multi-round-until-clear; do not use phrasing that reads as "ask then immediately solve."
2. Clarify first, answer later — do not rush solutions, conclusions, or full answers while critical unknowns remain. Proceed early only if the user explicitly asks to skip further clarification (record assumed defaults for the rest).
3. Recommended answer — when a reasonable default exists, every question includes a recommended option or stated default plus a one-line rationale. User may reply with just an option letter ("A", "B").
4. Facts vs decisions — if a fact is findable in the environment (repo, referenced files, prior messages, tools), look it up; do not spend the question slot. The decisions are the user's — put each one to them and wait.
5. Context before asking — before the first clarifying question (and when stuck), skim relevant files, docs, and recent commits. If the request spans multiple independent subsystems, help decompose scope first; do not burn rounds refining details of a project that must be split.
6. Shared-understanding gate — do not act on the clarified plan/design/implementation path until the user confirms you share an understanding, or explicitly skips further clarification.

Unconditional: this discipline always applies; it does not depend on optional enhancement skills being installed.

Picking the next question: follow decision-tree dependencies first; when several roots are open, prefer purpose → constraints → success criteria.

Question shape (platform-agnostic)

Describe intent; the agent picks native capability — do not hardcode a platform-specific tool or enumerate "platform X uses A, Y uses B".

• Prefer structured single-select (one question + options + recommended answer); open-ended is fine when fixed options would fake certainty.
• Fall back to plain prose when structured UI is unavailable.

[One sentence stating the question clearly]
- A option one
- B option two
Recommended: A (brief rationale)

Investigation-first

1. Investigate before advising — no investigation, no say.
2. Speak with evidence; avoid void hypotheses.

Host integration (three touchpoints only)

Referencing workflows keep exactly these touchpoints; the full discipline lives only here:

1. Prominent pointer — one tagged line naming this skill and one-per-turn + multi-round until clear + clarify-first. Prefer: ⚠️ Follow clarifying-question-discipline (one question per turn; multi-round until clear; clarify first, do not rush to answer). Hosts MAY add a locale-equivalent one-liner; do not restate the full decision-tree rules in the host.
2. Entry-point quantity — at each user-questioning step, state: ask only ONE most critical question this turn; ask the next only after the answer.
3. Red Flags — dumping multiple questions/open points at once; rushing to answer during clarification.

Skills that do not declare the dependency inline the full form themselves (pointer + quantity + Red Flags + platform-agnostic phrasing).