brainstorm-diverge-converge

$npx mdskill add lyndonkl/claude/brainstorm-diverge-converge

Generate, cluster, and select the best ideas for complex problems.

  • Helps solve open-ended problems by creating many creative options.
  • Groups ideas into themes before evaluating them against criteria.
  • Decides on the strongest choices through systematic filtering.
  • Delivers a ranked list of refined solutions ready for action.
SKILL.md
.github/skills/brainstorm-diverge-convergeView on GitHub ↗
---
name: brainstorm-diverge-converge
description: Applies structured divergent-convergent thinking to generate many creative options, organize them into meaningful clusters, then systematically evaluate and narrow to the strongest choices. Balances creative exploration with disciplined decision-making. Use when exploring product ideas, solving open-ended problems, generating strategic alternatives, developing research questions, designing experiments, or when user mentions brainstorming, ideation, divergent thinking, generating options, or evaluating alternatives.
---

# Brainstorm Diverge-Converge

## Table of Contents

- [Workflow](#workflow)
  - [1. Gather Requirements](#1--gather-requirements)
  - [2. Diverge (Generate Ideas)](#2--diverge-generate-ideas)
  - [3. Cluster (Group Themes)](#3--cluster-group-themes)
  - [4. Converge (Evaluate & Select)](#4--converge-evaluate--select)
  - [5. Document & Validate](#5--document--validate)
- [Common Patterns](#common-patterns)
- [Guardrails](#guardrails)
- [Quick Reference](#quick-reference)

Three phases: Diverge (generate many ideas without judgment), Cluster (group into themes), Converge (evaluate against criteria and select).

**Quick Example:**

```markdown
# Problem: How to improve customer onboarding?

## Diverge (30 ideas)
- In-app video tutorials
- Interactive walkthroughs
- Email drip campaign
- Live webinar onboarding
- 1-on-1 concierge calls
- ... (25 more ideas)

## Cluster (6 themes)
1. **Self-serve content** (videos, docs, tooltips)
2. **Interactive guidance** (walkthroughs, checklists)
3. **Human touch** (calls, webinars, chat)
4. **Motivation** (gamification, progress tracking)
5. **Timing** (just-in-time help, preemptive)
6. **Social** (community, peer examples)

## Converge (Top 3)
1. Interactive walkthrough (high impact, medium effort) - 8.5/10
2. Email drip campaign (medium impact, low effort) - 8.0/10
3. Just-in-time tooltips (medium impact, low effort) - 7.5/10
```

## Workflow

Copy this checklist and track your progress:

```
Brainstorm Progress:
- [ ] Step 1: Gather requirements
- [ ] Step 2: Diverge (generate ideas)
- [ ] Step 3: Cluster (group themes)
- [ ] Step 4: Converge (evaluate and select)
- [ ] Step 5: Document and validate
```

**Step 1: Gather requirements**

Clarify topic/problem (what are you brainstorming?), goal (what decision will this inform?), constraints (must-haves, no-gos, boundaries), evaluation criteria (what makes an idea "good" - impact, feasibility, cost, speed, risk, alignment), target quantity (suggest 20-50 ideas), and rounds (single session or multiple rounds, default: 1).

**Step 2: Diverge (generate ideas)**

Generate 20-50 ideas without judgment or filtering. Suspend criticism (all ideas valid during divergence), aim for quantity and variety (different types, scales, approaches), and use creative prompts: "What if unlimited resources?", "What would competitor do?", "Simplest approach?", "Most ambitious?", "Unconventional alternatives?". Output: Numbered list of raw ideas. For simple topics → generate directly. For complex topics → Use `resources/template.md` for structured prompts.

**Step 3: Cluster (group themes)**

Organize ideas into 4-8 distinct clusters by identifying patterns, creating categories (mechanism, user/audience, timeline, effort, risk, strategic objective), naming clusters clearly, and checking coverage (distinct approaches). Fewer than 4 = not enough variety, more than 8 = too fragmented. Output: Ideas grouped under cluster labels.

**Step 4: Converge (evaluate and select)**

Define criteria (from step 1), score ideas on criteria (1-10 or Low/Med/High scale), rank by total/weighted score, select top 3-5 options, and document tradeoffs (why chosen, what deprioritized). Evaluation patterns: Impact/Effort matrix, weighted scoring, must-have filtering, pairwise comparison. See [Common Patterns](#common-patterns) for domain-specific approaches.

**Step 5: Document and validate**

Create `brainstorm-diverge-converge.md` with: problem statement, diverge (full list), cluster (organized themes), converge (scored/ranked/selected), and next steps. Validate using `resources/evaluators/rubric_brainstorm_diverge_converge.json`: verify 20+ ideas with variety, distinct clusters, explicit criteria, consistent scoring, top selections clearly better, actionable next steps. Minimum standard: Score ≥ 3.5.

## Common Patterns

**For product/feature ideation:**
- Diverge: 30-50 feature ideas
- Cluster by: User need, use case, or feature type
- Converge: Impact vs. effort scoring
- Select: Top 3-5 for roadmap

**For problem-solving:**
- Diverge: 20-40 solution approaches
- Cluster by: Mechanism (how it solves problem)
- Converge: Feasibility vs. effectiveness
- Select: Top 2-3 to prototype

**For research questions:**
- Diverge: 25-40 potential questions
- Cluster by: Research method or domain
- Converge: Novelty, tractability, impact
- Select: Top 3-5 to investigate

**For strategic planning:**
- Diverge: 20-30 strategic initiatives
- Cluster by: Time horizon or strategic pillar
- Converge: Strategic value vs. resource requirements
- Select: Top 5 for quarterly planning

## Guardrails

**Do:**
- Generate at least 20 ideas in diverge phase (quantity matters)
- Suspend judgment during divergence (criticism kills creativity)
- Create distinct clusters (avoid overlap and confusion)
- Use explicit, relevant criteria for convergence (not vague "goodness")
- Score consistently across all ideas
- Document why top ideas were selected (transparency)
- Include "runner-up" ideas (for later consideration)

**Don't:**
- Filter ideas during divergence (defeats the purpose)
- Create clusters that are too similar or overlapping
- Use vague evaluation criteria ("better", "more appealing")
- Cherry-pick scores to favor pet ideas
- Select ideas without systematic evaluation
- Ignore constraints from requirements gathering
- Skip documentation of the full process

## Quick Reference

- **Template**: `resources/template.md` - Structured prompts and techniques for diverge-cluster-converge
- **Quality rubric**: `resources/evaluators/rubric_brainstorm_diverge_converge.json`
- **Output file**: `brainstorm-diverge-converge.md`
- **Typical idea count**: 20-50 ideas → 4-8 clusters → 3-5 selections
- **Common criteria**: Impact, Feasibility, Cost, Speed, Risk, Alignment
More from lyndonkl/claude