Brainstorming

Purpose

Guide users through structured ideation and creative problem-solving. Detects starting state (stuck, seed, or open) and applies appropriate techniques: escaping convergent thinking traps, expanding half-formed ideas, or systematic generation and selection.

Entry: Where Are You Starting?

BEFORE PROCEEDING, require the user to explicitly choose their starting point:

Starting Point	Description	Next Step
Stuck	Same ideas keep surfacing. "We've tried everything."	→ Escape Mode
Seed	Have an idea but need to expand it	→ Expand Mode
Open	Need to generate many options from scratch	→ Diverge-Converge Mode

YOU MUST get explicit confirmation: "Which starting point describes you: Stuck, Seed, or Open?" No proceeding until selected.

Escape Mode: Breaking Convergent Thinking

When ideas cluster around the same approaches.

Step 1: Detect Convergence Blindness

YOU MUST check if we're stuck in one of these states (see [references/convergence-states.md]):

•B1 First-Attractors: First ideas feel "right" immediately
•B2 Function-Lock: Discussion assumes a solution type ("We need an app...")
•B3 Axis-Collapse: Ideas differ cosmetically but share structure
•B4 Domain-Prison: All ideas come from same reference class

Ask diagnostic questions from the reference. If any state matches, IMMEDIATELY apply the corresponding intervention. Never skip this step—convergence blindness kills creativity every time.

Step 2: Function Extraction

Separate WHAT from HOW:

•List 3-5 core functions the solution must accomplish (independent of form)
•Reframe: "We need [FORM]" becomes "We need to [FUNCTION]"

Step 3: Axis Mapping

Map the default solution on four axes:

•Who: Who does/uses this? Try 3 unlikely actors
•When: What timeframe? Try different cadence
•Scale: What size? Try 10x bigger or smaller
•Method: What approach? Try completely different mechanism

Step 4: Generate Under Constraints

Apply one unusual constraint (see [references/constraint-prompts.md]):

•Random actor from different domain
•Resource limitation (1/10th budget, no obvious tech)
•Inversion (what if failure was the goal?)

Generate 3-5 ideas under this constraint. The constraint forces exploration of non-adjacent space.

Expand Mode: Growing Seeds

When you have an initial idea that needs development.

Step 1: Identify Seed Type

Determine which state describes the seed (see [references/seed-states.md]):

•S1 Adjacent-Ready: Concrete, clear next step needed
•S2 Collision-Hungry: Feels incomplete, needs "something else"
•S3 Half-Baked Hunch: Fuzzy but feels important
•S4 Error-Rich: Tried and failed, but learned something
•S5 Exaptation: Could work in completely different context

Step 2: Apply Expansion Technique

Seed Type	Technique
Adjacent-Ready	Map what's one step away; what becomes possible
Collision-Hungry	Force collisions: domains, constraints, past work
Half-Baked Hunch	Articulate clearly, name the gap, keep it alive
Error-Rich	Mine the failure: what did error reveal?
Exaptation	Transplant: list 5 different contexts, try seed in each

Step 3: Stress Test

Find where the seed breaks:

•What's the worst-case application?
•What assumption, if wrong, kills this?
•What's the dumbest possible version?

Diverge-Converge Mode: Systematic Generation

When you need to generate and select from many options.

Phase 1: Diverge

Generate 20-50 ideas without judgment:

•Suspend criticism entirely
•Aim for quantity AND variety
•Use prompts: "What if unlimited resources?", "What would competitor do?", "What if we did the opposite?"

Phase 2: Cluster

Organize into 4-8 distinct themes:

•Group by mechanism, user, timeline, effort, or risk
•Aim for distinct clusters (not overlapping)
•Fewer than 4 = not enough variety, more than 8 = too fragmented

Phase 3: Converge

Select top 3-5 ideas:

•Define explicit criteria (impact, feasibility, cost, alignment)
•Score each idea (1-10 or Low/Med/High)
•Rank by score
•Document tradeoffs: why chosen, what deprioritized

YOU MUST get explicit selection: "Which ideas are we moving forward with?" Confirm choices before proceeding.

Anti-Patterns: NEVER Use These

Throughout all modes, we guard against these failure patterns:

Pattern	Problem	Fix
Quantity Delusion	50 variations of same 3 approaches	Map on axes, require one idea per quadrant
Premature Evaluation	"That won't work..." during generation	Strict phase separation: generate first, evaluate after
Expert Anchor	First expert idea dominates	Anonymous generation first, or expert speaks last
Inversion Trap	"Do the opposite" (still same axis)	Find orthogonal axis, not negative of same
Novelty Chase	Weird but doesn't solve problem	Return to function: does this serve required function?

Premature evaluation without generation first = guaranteed mediocre results. Every time.

Collaborative Design Mode: Validation

When you have a direction and need to refine it through dialogue.

Understanding Phase

•Check our project context first (files, docs)
•Ask ONE question at a time (prefer multiple choice)
•Focus: purpose, constraints, success criteria

Exploration Phase

•Present 2-3 approaches with trade-offs
•Lead with our recommendation and reasoning
•YOU MUST get explicit selection: "Which approach should we pursue?" before proceeding

Design Phase

•Present design in 200-300 word sections
•Validate after each section before proceeding
•Cover: architecture, components, data flow, error handling, testing
•Be ready to return to previous sections

Output

Write validated design to docs/plans/YYYY-MM-DD-<topic>-design.md

References

Load references as needed for specific techniques:

•[references/convergence-states.md]: Load when user is stuck—diagnose B1-B4 blind spots and apply interventions
•[references/seed-states.md]: Load when expanding a seed idea—identify S1-S5 state and apply expansion technique
•[references/constraint-prompts.md]: Load when needing random constraints for entropy injection (Escape Mode Step 4)
•[references/domain-patterns.md]: Load when running Diverge-Converge Mode for specific domain templates (product, research, strategy, etc.)