메뉴
Socratic problem exploration — identifies critical unknowns and gates on ≤20% ambiguity before proceeding
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Convert PDF/EPUB library to Markdown and generate Obsidian MOC notes
Hook-based compaction suggestions at logical task boundaries
Context window management — track spend, decide when to compact, preserve state
Session-start orientation — loads context, surfaces learnings, confirms registry
Quality and semantic review — catches what automated tools miss
Planner → Architect → Critic deliberation loop — produces a formally validated ADR
| name | deep-interview |
| description | Socratic problem exploration — identifies critical unknowns and gates on ≤20% ambiguity before proceeding |
| version | 0.1.0 |
| level | 3 |
| triggers | ["deep interview","explore this problem","before we spec","understand the problem","/deep-interview"] |
| pipeline | ["deep-interview","prd"] |
| context_files | ["context/user.md","context/project.md"] |
| steps | [{"name":"Problem Statement","description":"Ask the user to state the problem in one sentence. Restate it back. Confirm alignment before proceeding."},{"name":"Critical Unknown Identification","description":"Identify the full set of critical unknowns — questions whose answers would change the approach. Score ambiguity."},{"name":"Socratic Questioning","description":"Ask the most impactful unknown first. One question at a time. Recalculate ambiguity score after each answer."},{"name":"Ambiguity Gate","description":"When ambiguity score reaches ≤20%, the gate opens. State the score and offer to proceed to PRD."},{"name":"Handoff","description":"Summarize what was learned. Invoke the prd skill with the enriched problem understanding as context."}] |
Systematic Socratic exploration before requirements are written. The output is a understood problem, not a solution. PRD comes after.
Without deep-interview, PRD starts from the user's initial framing. The initial framing is almost always incomplete — it describes symptoms rather than root causes, assumes constraints that may not exist, and omits the edge cases that make the problem hard. PRD built on an incomplete framing produces requirements that solve the wrong problem precisely.
Ambiguity score = unanswered critical questions ÷ total critical questions identified
Target gate: ≤ 20% (at least 80% of critical questions answered)
A question is critical if its answer would change either the approach or the scope. Not all questions are critical.
| Category | Examples |
|---|---|
| User / Stakeholder | Who experiences this problem? Who decides success? |
| Scope | What is explicitly in and out? Where does this end? |
| Constraints | What cannot change? What must remain backwards-compatible? |
| Success criteria | What does success look like? How will it be measured? |
| Failure modes | What are the top 3 ways this solution fails? |
| Prior attempts | What has been tried? Why did it fail? |
| Integration | What does this touch? What is upstream and downstream? |
Ask: "State the problem in one sentence."
Restate it back in different words. Ask: "Is this right?" Do not proceed until the user confirms.
Before asking any questions, internally identify the full set of critical unknowns across all 7 categories. Count them. This is the denominator for ambiguity scoring.
State to the user: "I've identified [N] critical unknowns. I'll work through them — we need ≤20% unanswered before proceeding."
One question at a time. After each answer:
Do not ask multiple questions at once. The point is depth of understanding, not breadth of coverage.
When the score reaches ≤ 20%, state:
Ambiguity score: [N]% ([X] of [Y] critical questions answered)
Gate: OPEN — ready to proceed to PRD.
Summary of what we know:
- [Key finding 1]
- [Key finding 2]
- [Key finding 3]
Proceed to PRD? (yes / continue interviewing)
The user may choose to continue interviewing past the gate. The gate is a floor, not a ceiling.
Pass the enriched problem understanding to the prd skill. The PRD should reference the interview findings, not re-derive them.
Do not ask questions you already know the answers to from context/project.md or context/user.md. Read those first.
Do not ask for information that does not change the approach. Only critical unknowns count toward the score.
Do not rush to the gate. If the score drops to 20% but a major category (e.g., failure modes) is completely unanswered, ask one more question from that category.
Do not paraphrase the user's answer back as a new question. Each exchange should reduce ambiguity, not circle.