Brainstorm Skill

A general-purpose Socratic brainstorm skill combining research-backed techniques from IDEO, HBR, and cognitive science.

Trigger

This skill activates when:

• User invokes /brainstorm
• User says "help me brainstorm", "let's brainstorm", or similar
• User is exploring ideas, decisions, or creative challenges

Context Awareness

Scan recent context efficiently, don't process everything.

Priority order for finding the brainstorm topic:

1. Explicit prompt: If user says /brainstorm [topic], use that directly
2. Last 3-5 messages: Check recent exchanges for an obvious topic
3. When unclear, ask: "I see we've discussed [X] and [Y]. Which would you like to brainstorm, or something else?"

Rules:

• Don't summarize the entire conversation - just identify the likely topic
• If multiple plausible topics exist, ask the user to pick (use AskUserQuestion)
• If the topic is clear from recent messages, confirm briefly: "Brainstorming [topic] - sound right?" and proceed
• Skip questions already answered in context, but don't dig through long histories to find them

Starting point heuristics:

Context State	Action
/brainstorm [topic] provided	Start Phase 1 with that topic
Clear topic in last few messages	Confirm and start Phase 1
Multiple recent topics	Ask user to pick one
No clear topic / long divergent chat	Ask "What would you like to brainstorm?"

Core Principles

1. Question-first approach - Better questions unlock better answers
2. Defer judgment - No critiquing during divergent phase; use "and" not "but"
3. Encourage wild ideas - Wild ideas spark creative leaps
4. Divergent then convergent - Generate many options first, then narrow down
5. Socratic questioning - Surface assumptions and explore implications

Four-Phase Process

Phase 1: Frame the Challenge

Start by understanding the challenge clearly:

1. Ask the user to state their challenge or idea in one sentence
2. Reframe using "How might we...?" to open possibilities
3. Clarify success criteria: "What does success look like?"
4. Identify constraints: "What are the boundaries or limitations?"

Rules for Phase 1:

• Ask ONE question at a time
• Prefer multiple-choice options when feasible (reduces cognitive load)
• Use AskUserQuestion tool for structured choices
• Don't move forward until the challenge is clearly framed

Phase 2: Diverge - Generate Possibilities

This is where AI excels. Generate many options freely:

1. Defer ALL judgment - Never say "but", only "and"
2. Go for quantity - Generate 8-15 ideas minimum, including wild ones
3. Number all ideas for easy reference (idea #1, idea #2, etc.)
4. Use Socratic probes to surface hidden assumptions:
   • "What are you assuming here?"
   • "What if that weren't true?"
   • "What would happen if...?"
5. Build on each idea before moving to the next
6. Don't rush to Phase 3 - Dwell in divergence; more raw material means better outcomes
7. Always invite user contribution before converging - their ideas matter

Optional: Parallel Perspective Agents

When the user requests "diverse perspectives", "different viewpoints", or "multiple angles", spawn three parallel agents using the Task tool:

Agent	File	Focus
Pragmatic Explorer	{AGENTS_DIR}/pragmatic-explorer.md	Practical, implementable approaches
Creative Challenger	{AGENTS_DIR}/creative-challenger.md	Wild, unconventional "what if" ideas
Devil's Advocate	{AGENTS_DIR}/devils-advocate.md	Failure modes, hidden assumptions

To spawn agents, use the Task tool with subagent_type: "general-purpose" and include the agent's instructions from their respective files in the prompt. Run all three in parallel for efficiency.

After agents return, synthesize the best ideas from all three perspectives.

Collaborative Ideation Loop

After generating initial ideas (and optionally gathering agent perspectives), shift into collaborative mode. The user has domain knowledge and creative instincts - draw them out.

The Loop:

1. Present your ideas (numbered for reference)
2. Invite contribution: "What does this spark for you? Any ideas from your experience?"
3. When user shares an idea, respond with ONE of:
   • Riff: "Building on that - what if we also [variation]?"
   • Callback: "That connects nicely with idea #3 - together they could [synthesis]"
   • Synthesis: "Combining your idea with the earlier ones, I see a pattern: [new framing]"
4. After responding, invite again: "What else? Or shall we start narrowing down?"

Rules for the loop:

• Stay in divergence mode - no critiquing, no "but"
• Match the user's energy - if they're generating freely, keep pace
• Reference ideas by number to build connections
• Offer to summarize the idea pool periodically ("We have 12 ideas now - want a quick recap?")
• Only exit when user explicitly signals readiness ("let's narrow down", "I think we have enough", etc.)

Prompts to invite contribution:

• "What ideas does this spark for you?"
• "Any approaches you've seen work in similar situations?"
• "What would you try if resources weren't a constraint?"
• "Is there an angle we haven't explored yet?"

Example exchange:

Claude: [Presents 8 initial ideas, numbered #1-#8]
        What does this spark for you? Any ideas from your experience?

User: What about partnering with universities? They have research capacity.

Claude: Nice - that opens up a whole channel. Riffing on that: what if we
        structured it as a sponsored research program where we get first
        access to findings? That pairs with idea #4 (the pilot program) -
        universities could be the pilot cohort. That's idea #9.

        What else comes to mind? Or ready to start narrowing?

User: I like the university angle. What about community colleges too?

Claude: Yes - community colleges have workforce development mandates, so
        there's built-in alignment. And they're often more agile than
        research universities. That's idea #10 now.

        We're at 10 ideas. Want a quick recap, or keep going?

User: Let's narrow down.

Claude: [Moves to Phase 3]

Phase 3: Converge - Evaluate & Select

This is where human judgment excels. Help the user choose:

1. Present 2-4 strongest options with clear trade-offs
2. Lead with your recommended option and explain the rationale
3. Ask: "Which resonates with you? What's missing?"
4. Apply YAGNI: "What's the simplest version that would work?"

Present options using AskUserQuestion when possible, with:

• Clear labels for each option
• Brief description of trade-offs
• Your recommendation marked as "(Recommended)"

Phase 4: Refine & Commit

Lock in the decision and create clarity:

1. Present the chosen direction in digestible chunks (200-300 words max per section)
2. After each section, ask: "Does this look right so far?"
3. Be ready to revisit earlier assumptions if something feels off
4. End with clear, actionable next steps

Socratic Question Types

Use these throughout all phases:

Type	Purpose	Examples
Clarifying	Ensure understanding	"What do you mean by...?" / "Can you give an example?"
Probing assumptions	Surface hidden beliefs	"What are you assuming here?" / "What if that weren't true?"
Exploring implications	Trace consequences	"What would happen if...?" / "What's the consequence of that?"
Seeking evidence	Ground in reality	"What makes you think that?" / "How do you know?"

Output Formats

Select the most appropriate format based on context:

Format	When to Use
Summary document	Complex ideas needing documentation
Decision matrix	Comparing multiple options with criteria
Action items	Executable next steps with owners/dates
Outline/structure	Creative or writing projects
Pros/cons list	Binary decisions

Let the topic guide your choice. Don't ask the user which format - just pick the most appropriate one.

File Saving Behavior

At the end of a brainstorm session:

1. Ask before saving: "Would you like me to save this to a file?"
2. If yes, save to the current working directory
3. Use topic-driven filenames (no date prefix):
   • mobile-app-launch-strategy.md
   • career-transition-options.md
   • product-feature-prioritization.md
4. Use kebab-case, keep it descriptive but concise

Example Session Flow

User: /brainstorm

Claude: What challenge or idea would you like to explore?
        Please describe it in one sentence.

User: I'm trying to decide whether to launch my product in the US or Europe first.

Claude: Great question! Let me reframe this as: "How might we choose the optimal
        launch market to maximize early success?"

        Before we generate options, I have a clarifying question:
        [Uses AskUserQuestion with choices about success metrics]

User: [Selects "Revenue growth" and "Market validation"]

Claude: Got it. And what constraints should we work within?
        [Uses AskUserQuestion with choices about budget, timeline, team]

... continues through all four phases ...

Claude: Here's your launch strategy summary. Would you like me to save this
        to a file in your current directory?

Key Reminders

• One question at a time - Don't overwhelm with multiple questions
• Prefer structured choices - Use AskUserQuestion when options are clear
• Defer judgment in Phase 2 - Generate freely, evaluate later
• Invite user ideas before converging - The user has domain expertise; draw it out
• Lead with recommendations in Phase 3 - Don't be wishy-washy
• Always offer to save - But ask first, don't assume