onejaejae/ideation

Ideation

Turn vague ideas into concrete plans through structured divergent-then-convergent exploration.

Core principle: Explore the problem BEFORE solving it. Generate many approaches BEFORE choosing one.

CRITICAL: AskUserQuestion 턴 분리 규칙

AskUserQuestion은 이 스킬이 로드된 턴(같은 assistant turn)에서 절대 호출하지 마세요.

Skill tool로 이 스킬이 로드되면, 같은 턴에서 AskUserQuestion을 호출할 경우 사용자에게 질문 UI가 표시되지 않고 빈 응답으로 자동 처리됩니다 (Claude Code 플랫폼 제약).

필수 절차:

1. Phase 0 (Seed Capture)을 수행한다 — 사용자의 아이디어를 분석한다
2. 아이디어 요약, 타입, 내포된 가정을 텍스트로 출력한다
3. 반드시 STOP하고 사용자 응답을 기다린다
4. 사용자가 응답한 다음 턴에서 AskUserQuestion으로 Phase 1 Discover를 시작한다

이 규칙을 어기면 사용자가 질문을 볼 수 없고, 아이데이션이 진행되지 않습니다.

Process Flow

digraph ideation {
    rankdir=TB;

    "User presents idea" [shape=doublecircle];
    "Capture seed verbatim" [shape=box];
    "Explored problem space?" [shape=diamond];
    "Ask probing questions (Discover)" [shape=box];
    "Reframe as HMW questions (Define)" [shape=box];
    "User agrees on framing?" [shape=diamond];
    "Generated 5+ different approaches?" [shape=diamond];
    "Generate diverse solutions (Ideate)" [shape=box];
    "STOP: No evaluation during generation" [shape=octagon, style=filled, fillcolor=red, fontcolor=white];
    "User selects direction" [shape=box];
    "Develop concrete plan (Design)" [shape=box];
    "Pre-mortem reality check" [shape=box];
    "Deliver output" [shape=doublecircle];

    "User presents idea" -> "Capture seed verbatim";
    "Capture seed verbatim" -> "Explored problem space?";
    "Explored problem space?" -> "Ask probing questions (Discover)" [label="no"];
    "Ask probing questions (Discover)" -> "Explored problem space?" [label="loop 2-4 rounds"];
    "Explored problem space?" -> "Reframe as HMW questions (Define)" [label="yes"];
    "Reframe as HMW questions (Define)" -> "User agrees on framing?";
    "User agrees on framing?" -> "Ask probing questions (Discover)" [label="no, go deeper"];
    "User agrees on framing?" -> "Generated 5+ different approaches?" [label="yes"];
    "Generated 5+ different approaches?" -> "Generate diverse solutions (Ideate)" [label="no"];
    "Generate diverse solutions (Ideate)" -> "STOP: No evaluation during generation";
    "STOP: No evaluation during generation" -> "Generated 5+ different approaches?";
    "Generated 5+ different approaches?" -> "User selects direction" [label="yes"];
    "User selects direction" -> "Develop concrete plan (Design)";
    "Develop concrete plan (Design)" -> "Pre-mortem reality check";
    "Pre-mortem reality check" -> "Deliver output";
}

Phase 0: Seed Capture

Capture the user's idea verbatim. Do NOT interpret, refine, or solve yet.

Note:

• What type? (architecture / feature / tech adoption / process / exploration)
• What excites them about it?
• What solution did they imply (if any)? — this is the anchor to challenge later

Scaling the Process

Not every idea needs all 4 phases at full depth. Scale to complexity:

| Complexity | Phases | Example | |---|---|---| | Light | Phase 0 → Phase 1 (1 round) → Phase 3 → done | "로깅 라이브러리 뭐 쓸까?" | | Medium | All phases, 2 rounds each | "캐시 시스템 도입하고 싶어" | | Deep | All phases, 3-4 rounds each + written output | "마이크로서비스 전환 검토" |

Minimum requirement: Even for "Light" ideas, always do at least Phase 0 + 1 round of Phase 1 questions. Never skip straight to solutions.

If the user says "질문 말고 그냥 아이디어 줘": Acknowledge the request, explain briefly why 1 round of questions will make the ideas much better, then ask just 2 focused questions. If they insist again, proceed to Phase 3 but flag assumptions made without exploration.

Phase 1: Discover (DIVERGENT — Expand Problem Space)

RULE: NO solutions in this phase. Only questions.

Pacing: 2-3 questions per round, max. Do NOT dump all questions at once — even under time pressure. Each round builds on previous answers.

Use AskUserQuestion (2-3 questions per round, 2-4 rounds):

| Question Type | Purpose | Example | |---|---|---| | Why | Uncover motivation | "What triggered this idea? What pain drives it?" | | Who | Identify stakeholders | "Who benefits most? Who is affected negatively?" | | What-if-not | Explore inaction | "What happens if we do nothing for 6 months?" | | Constraint removal | Expand space | "If [X] didn't exist, what would you build?" | | Constraint addition | Reveal priorities | "If you had to ship in 1 week, what's the core?" | | Inversion | Surface risks | "How would this fail spectacularly?" | | Scaling | Test boundaries | "What changes at 10x? At 100x?" |

Anti-anchoring: If the user names a specific solution (e.g., "마이크로서비스로 전환"), do NOT accept it as the answer. Explore the underlying problem first. The named solution is a hypothesis, not a conclusion.

Completion criteria: Move to Phase 2 when you understand:

• The real pain/opportunity (not just the surface request)
• Who cares and why
• Key constraints (real vs. assumed)
• What success looks like

Phase 2: Define (CONVERGENT — Reframe the Problem)

Synthesize Phase 1 into How Might We (HMW) questions. Present 3-5 framings via AskUserQuestion.

Example: "서비스가 커지고 있어서 마이크로서비스로 전환하고 싶어" →

• HMW: "팀이 서로 블로킹 없이 독립적으로 배포하려면?"
• HMW: "코드베이스 복잡성을 관리하면서 10배 성장하려면?"
• HMW: "배포 안정성을 높이면서 속도도 빠르게 하려면?"
• HMW: "새 팀원이 1주일 안에 생산적이 되려면?"

Each HMW opens a different solution space. Get user agreement on which 1-2 framings capture the real opportunity.

Phase 3: Ideate (DIVERGENT — Generate Solutions)

RULES:

1. Minimum 5 genuinely different approaches — NOT variations of one idea
2. Each approach must have a fundamentally different strategy (different trade-offs, different assumptions)
3. Include at least 1 unconventional/surprising approach
4. Use cross-domain analogies where helpful ("How does [unrelated field] solve this?")
5. NO evaluation during this phase — list ALL approaches neutrally

Present as a comparison table:

| # | Approach | Core Idea | Key Trade-off |
|---|----------|-----------|---------------|
| 1 | ...      | ...       | ...           |
| 2 | ...      | ...       | ...           |

Then ask the user to select 1-2 approaches for deeper exploration via AskUserQuestion.

Phase 4: Design (CONVERGENT — Concrete Plan)

For selected approach(es):

1. Evaluate — Quick assessment: Feasibility / Impact / Risk / Effort
2. Scope v1 — What's IN? What's explicitly OUT?
3. Key decisions — 2-3 critical decisions with trade-offs (ADR style: Options → Decision → Rationale)
4. Pre-mortem — "It's 6 months later, this failed. Why?" Surface 3+ failure modes
5. Next steps — Concrete action items

Output Format

Adapt to the depth of exploration:

| Depth | When | Output | |---|---|---| | Quick | User wants ideas, not a plan | Conversation: idea list + insights | | Medium | Comparison needed | Comparison table + recommended direction with rationale | | Deep | Architecture/major decision | Written document: plans/{topic-slug}.md with ADR structure |

For written documents, include: Context, Decision Drivers, Options Considered (with pros/cons), Recommendation, Consequences, Next Steps.

Anti-patterns — What This Skill Prevents

| Anti-pattern | Symptom | Fix | |---|---|---| | Premature convergence | Jumping to solutions in Phase 1 | Enforce phase separation. NO solutions until Phase 3 | | Anchoring | Accepting user's implied solution as given | Treat named solutions as hypotheses. Explore the problem first | | Shallow diversity | 5 variations of the same idea | Each approach needs different fundamental trade-offs | | Evaluation during ideation | "This won't work because..." in Phase 3 | Save ALL evaluation for Phase 4 | | Questions last | Giving advice, then asking "what's your situation?" | ALWAYS explore first (Phase 1-2), suggest later (Phase 3-4) | | Agreement bias | "Great idea! Here's how..." | Ask "What if the opposite were true?" in Phase 1 | | Question dump | 15+ questions in one round under time pressure | Strict 2-3 per round. Depth over breadth per round | | Skipping Phase 2 | Jumping from questions straight to solutions | HMW reframing ensures you're solving the RIGHT problem | | "빨리 시작해줘"에 STOP 생략 | 로드 턴에서 AskUserQuestion 호출 | 플랫폼 제약은 우회 불가 — 급해도 Seed Capture 후 STOP 필수 |