ask-deep

name	ask-deep
description	Detailed, multi-round planning interview for complex services, apps, websites, education products, AI workflows, or ambiguous projects where premature implementation would create rework. Use when the user wants many questions, deep clarification, or a full planning pass before implementation.

Ask Deep

Overview

Use this skill when the idea is important enough to deserve a careful interview. Work like a product strategist: ask, summarize, challenge ambiguity, and turn the result into a precise implementation prompt.

Principles

• Use the user's language.
• Ask in rounds, not one huge questionnaire.
• Ask 3-5 questions per round.
• Summarize after each round and identify what is still unclear.
• Treat the conversation itself as the planning artifact: each answer should become a decision, an assumption, an open question, or an implementation constraint.
• Probe vague words such as "simple", "pretty", "smart", "easy", "AI-powered", "dashboard", or "personalized".
• Keep confirmed decisions, assumptions, open questions, and non-goals separate.
• Stop when the user says to proceed, stop asking, or that the plan is enough.

Opt-Out

If the user explicitly declines the interview even for a complex project, do not insist on the full interview. First check whether the request is concrete enough to guess safely.

• If the goal and primary user are already clear enough from context, state the assumptions you would otherwise have asked about, mark them clearly as assumptions rather than confirmed decisions, note in one sentence that skipping raises rework risk, and proceed.
• If the goal or primary user is unclear enough that a wrong guess would produce a fundamentally different result, ask exactly one question to resolve that single biggest unknown — even a complex project deserves at least this much — then proceed without further questions.

Persisting State

Long, multi-round interviews are vulnerable to lost context: chat history can be summarized, sessions can restart, and work can span multiple days. After each round, write or update a PLAN.md file in the project root with the current Confirmed Decisions, Assumptions, Open Questions, and Non-Goals. Mention the file path the first time you create it. Treat that file, not the chat history, as the source of truth for what was decided.

Conversation Loop

Repeat this loop until the plan is clear enough to hand off:

1. Ask 3-5 focused questions.
2. Read the user's answers for decisions, preferences, constraints, and contradictions.
3. Summarize what became clearer.
4. Name what is still ambiguous.
5. Ask the next round based on the user's answers, not a fixed checklist.
6. When enough is known, convert the conversation into a prompt with ask-prompt.

Do not skip the summary step. The summary is how the user catches misunderstandings before implementation.

Interview Rounds

1. Intent: goal, audience, current pain, desired outcome.
2. Product shape: service type, core workflow, screens, content, data.
3. Scope: MVP, later work, explicit non-goals.
4. Users: personas, context, devices, accessibility needs.
5. Experience: user flow, states, edge cases, onboarding, errors.
6. Visual direction: route to ask-visual when UI or image style matters.
7. Domain depth: route to ask-educator for learning products or education content.
8. Risk: route to ask-risk for assumptions, contradictions, privacy, or feasibility.
9. Acceptance: success criteria, QA checks, demo path.
10. Handoff: route to ask-prompt for a final implementation prompt.

Deep Follow-Up Patterns

When the user gives a broad answer, convert it into specific choices:

• If they say "for students", ask grade, subject, device, and learning context.
• If they say "AI analysis", ask input, output, confidence, review, and privacy boundaries.
• If they say "dashboard", ask primary decision, top three metrics, filters, and drill-down path.
• If they say "beautiful UI", ask reference apps, tone, color, density, and platform.
• If they say "MVP", ask what can be removed while preserving the core value.

Output

Return, using the shared Ask Me Plan headers so state carries cleanly into the next skill:

• Interview Summary
• Confirmed Decisions (include scope)
• Assumptions
• Open Questions
• Non-Goals
• Implementation Prompt
• Optional next step: name the best next Ask skill (ask-visual, ask-educator, ask-risk, ask-prompt, or ask-build)