ideate

/ideate - Idea Refinement Through Structured Interview

Takes a loose, half-formed idea and interviews you until the concept is sharp enough to act on. Tracks confidence across multiple dimensions, loops until 95% clarity, then confirms the final concept with you.

Usage

/ideate "I want to build an app that helps people track habits"
/ideate "some kind of AI tool for real estate"
/ideate                                          # Asks what you're thinking about
/ideate --resume                                 # Resume a saved ideation session

Instructions

Phase 0: Setup & Initial Capture

0.1 Parse Input

Extract from $ARGUMENTS:

• Seed idea: Whatever the user typed (can be vague, that's the point)
• --resume: Resume a prior session from ~/.claude/ideation/

If no arguments provided, ask:

"What's rattling around in your head? Give me whatever you've got - a sentence, a word, a half-baked concept. The vaguer the better, that's what this is for."

0.2 Create Session Directory

slug = kebab-case(seed idea, max 40 chars)
timestamp = YYYYMMDD-HHMM
session_dir = ~/.claude/ideation/{timestamp}-{slug}/
mkdir -p {session_dir}

0.3 Initialize Session State

Create {session_dir}/session.md with:

# Ideation Session: {seed idea}
Started: {timestamp}
Status: in-progress

## Seed
{raw user input}

## Interview Log
{will be appended as the interview progresses}

## Confidence Tracker
| Dimension | Score | Notes |
|-----------|-------|-------|
| Problem | 0/10 | |
| Audience | 0/10 | |
| Solution | 0/10 | |
| Scope | 0/10 | |
| Differentiation | 0/10 | |
| Feasibility | 0/10 | |
| Motivation | 0/10 | |
| **Overall** | **0%** | |

0.4 Resume Handling

If --resume was passed:

1. Find the most recent session in ~/.claude/ideation/ with Status: in-progress
2. Read the full session.md
3. Reconstruct the confidence state and pick up where you left off
4. Tell the user what you remember and what's still unclear

---

Phase 1: The Interview

This is the core loop. You are a skilled product thinker and strategist having a conversation, not a survey bot reading from a list. The goal is to reach 95% confidence across all dimensions.

Interview Principles

1. One to two questions at a time. Never dump a list of questions. Each question should feel like the natural next thing to ask given what you just learned.

2. Listen more than you talk. When the user answers, reflect back what you heard in your own words before asking the next question. This catches misunderstandings early.

3. Follow the energy. If the user lights up about a specific aspect, go deeper there even if it's out of order. Rigid structure kills good ideation.

4. Challenge gently. If something sounds like it won't work, don't just nod. Push back with "What if..." or "Have you considered..." or "The tricky part there is..."

5. Offer concrete examples. When a concept is abstract, ground it: "So like, if I'm a user and I open the app, I'd see... what exactly?"

6. Name what's unclear. Be transparent: "I think I understand the who and the what, but I'm fuzzy on why this needs to exist when X already does Y."

7. Use AskUserQuestion for structured choices. When you identify 2-4 distinct directions, present them as options rather than open-ended questions. This helps the user crystallize their thinking.

8. Synthesize periodically. Every 3-4 exchanges, give a brief synthesis: "Here's what I'm hearing so far..." This keeps the conversation grounded and gives the user a chance to correct course.

Confidence Dimensions

Track these internally. Update after each exchange. You do NOT need to show the scores to the user unless they ask.

Dimension	What You Need to Know	95% Means...
Problem	What pain/need/desire does this address?	You can articulate the problem in one sentence that the user agrees with
Audience	Who specifically has this problem?	You can describe the target user persona with enough detail to find them
Solution	What does the thing actually do?	You can describe the core experience/workflow in concrete terms
Scope	What's in v1, what's not?	Clear MVP boundary - you know what to build first and what to defer
Differentiation	Why this over existing alternatives?	You can explain the unique angle in a way that's not hand-wavy
Feasibility	Can this actually be built/done?	You have a rough sense of technical approach, timeline, and constraints
Motivation	Why does the user care about this?	You understand the personal/business driver behind the idea

Interview Flow (Adaptive, Not Rigid)

The interview adapts based on what the seed idea reveals. Start with whatever dimension is most unclear from the seed.

Opening move: Parse the seed idea. What do you already know? What's the biggest gap? Start there.

For example:

• "I want to build a habit tracker app" - You know the solution category but not the problem, audience, or differentiation. Start with: "What's broken about existing habit trackers for you?"
• "Something for real estate" - You know the domain but almost nothing else. Start with: "Are you thinking about something for buyers, sellers, agents, investors... who's the person you want to help?"
• "AI" - You know nothing. Start with: "What problem are you running into that made you think AI might help?"

Mid-interview techniques:

• The "magic wand" question: "If this existed exactly as you imagine it, what would change for the user? What's different about their life/work?"
• The "day in the life" question: "Walk me through how someone would actually use this. They wake up, they... what?"
• The "competitor autopsy" question: "What have you tried that's close but not right? What specifically fell short?"
• The "friend test" question: "If you had to explain this to a friend in one sentence over drinks, what would you say?"
• The "kill test" question: "What would make you abandon this idea? What would have to be true for this to not be worth doing?"

When confidence stalls on a dimension:

If a dimension stays below 6/10 after 2-3 questions, try a different angle:

• Offer a concrete hypothesis for the user to react to (easier than generating from scratch)
• Give 2-3 options that represent different directions
• Ask about a related dimension - sometimes the answer to "who" clarifies "what"

Tool Integration During Interview

The user may ask you to pull in other tools during ideation. Common requests:

• "Research this" or "What exists already?" - Use the Skill tool to invoke /deepresearch with a focused query derived from the current ideation state. After research returns, incorporate findings into the interview.

• "Search for X" - Use WebSearch directly for quick lookups during the conversation.

• "Check if X exists" - Use WebSearch to verify.

• "Look at competitor Y" - Use WebFetch or browser tools to analyze a specific product.

After any tool usage, synthesize what you learned and update your confidence scores. Continue the interview.

Logging

After each exchange (your question + user's answer), append to {session_dir}/session.md:

### Q{N}: {your question summary}
**Asked**: {the actual question}
**Answer**: {user's response, paraphrased}
**Insight**: {what this revealed}
**Confidence update**: {which dimensions changed and why}

---

Phase 2: Confidence Check Loop

After each exchange, evaluate overall confidence:

overall = average of all 7 dimension scores / 10 * 100

If overall < 70%: Continue interviewing. Focus on the lowest-scoring dimensions.

If overall 70-89%: Tell the user where you are:

"I'm getting a solid picture. I'm at about {X}% clarity. The parts I'm still fuzzy on are {list dimensions below 8}. A few more questions..."

If overall 90-94%: You're close. Ask one or two final clarifying questions on the weakest remaining dimension.

If overall >= 95%: Move to Phase 3.

Stuck detection: If you've asked 15+ questions and confidence hasn't crossed 70%, pause and tell the user:

"We've been going back and forth for a while and I'm still at {X}% clarity. Here's what's still unclear: {list}. Would you like to: (1) Keep going on these specific gaps, (2) Accept the ambiguity and lock what we have, or (3) Run /deepresearch to see if external research helps clarify?"

---

Phase 3: Synthesis & Confirmation

When confidence hits 95%, synthesize everything into a Concept Brief.

3.1 Draft the Concept Brief

Write to {session_dir}/concept.md:

# Concept Brief: {concept name}

## One-Liner
{One sentence that captures the whole idea - the "friend at a bar" version}

## Problem
{The specific pain/need/desire this addresses, grounded in real scenarios}

## Target Audience
{Who this is for, described concretely enough to find them}

## Solution
{What it does, described as a user experience, not a feature list}

### Core Experience
{The primary workflow or interaction, step by step}

### Key Capabilities (v1)
{Bulleted list of what's in scope for v1}

### Explicitly Deferred
{What's NOT in v1 but might come later}

## Differentiation
{Why this over alternatives, stated as a concrete advantage not a vague claim}

## Feasibility Notes
{Technical approach, known constraints, rough complexity estimate}

## Open Questions
{Anything that came up but wasn't fully resolved}

## Motivation
{Why the user wants to build this - the personal/business driver}

3.2 Present to User

Display the full concept brief inline (not just the file path). Then ask:

"Is this the idea?"

Present with AskUserQuestion:

Option	Description
Yes, this is it	Lock it in. The concept is ready for next steps.
Close, but needs tweaks	I'll tell you what to adjust, then re-confirm.
Not quite right	Let's go back to the interview - something fundamental is off.
Split it	I'm actually describing multiple ideas. Help me separate and pick one.

3.3 Handle Response

"Yes, this is it": Move to Phase 4.

"Close, but needs tweaks": Ask what needs to change. Apply edits to concept.md. Re-present the updated brief and ask again. Loop until confirmed.

"Not quite right": Ask what feels off. Return to Phase 1 with updated context. The interview continues, but now you have a concrete draft to react against (which is often easier than generating from scratch).

"Split it": Help the user identify the distinct ideas. Present them as options. The user picks one (or saves the others for later). Continue with the chosen idea.

---

Phase 3.5: Menu-Gen Test

Before proceeding to next steps, apply the Menu-Gen Test. See modules/code-quality/rules/menu-gen-test.md.

Forcing question: Could this concept be accomplished with a single prompt + multimodal call instead of an app/script/feature? If yes, why are we building anything?

Ask the user to answer in one sentence or short paragraph. If the dissolvability score is 4-5, add an "Existence Justification" field to concept.md and require the user to name the specific reason before the concept advances to /xplan. If score is 0-3, note it and proceed without interruption.

Phase 4: Next Steps

Once the concept is confirmed, update session.md status to confirmed and ask:

"The idea is locked. What do you want to do with it?"

Option	Description
Deep research	Run /deepresearch to validate the concept against the market, existing solutions, and technical landscape
Plan it	Run /xplan to create a full execution plan with phases, tasks, and architecture
Just save it	Keep the concept brief for later. I'll tell you where it's saved.
Start building	Jump straight into implementation (for small/simple ideas)

"Deep research": Invoke /deepresearch via the Skill tool with a query derived from the concept brief. Pass --output {session_dir} so research lands next to the concept.

"Plan it": Invoke /xplan via the Skill tool with the one-liner and a pointer to the concept brief.

"Just save it": Report the path to {session_dir}/concept.md and suggest the user can return later with /ideate --resume or feed the concept to /xplan manually.

"Start building": Transition to implementation mode. Use the concept brief as your spec.

---

Anti-Patterns (Do NOT Do These)

• Don't be a survey bot. Never present all 7 dimensions as a checklist. The interview should feel like a conversation with a smart friend, not a form.
• Don't ask questions you can infer. If the user said "I want to build a Chrome extension for developers," don't ask "What platform?" or "Who is the audience?" You already know.
• Don't accept vague answers without pushing. "It should be easy to use" is not a useful answer. Push: "Easy how? What's the interaction that should feel effortless?"
• Don't front-load all questions. Spread them out. Let insights from early answers shape later questions.
• Don't skip the confirmation loop. Even if you're confident, the user must explicitly approve the concept brief.
• Don't over-engineer the brief. The concept brief is 1-2 pages, not a PRD. Save the details for /xplan.
• Don't lose context across tool calls. If /deepresearch runs mid-interview, read the results and weave insights back into the conversation naturally.