---

name: brainstorming description: "Use before any creative work - creating features, building components, adding functionality, or modifying behavior. Triggers on /brainstorm command, when exploring ideas before planning, when user describes a vague goal or feature request, or when design decisions need collaborative exploration. Explores user intent, requirements and design before implementation."

Brainstorming Ideas Into Designs

Overview

Brainstorming transforms vague ideas into validated designs through structured collaborative dialogue. It bridges the gap between "I want X" and a concrete design that can be handed to the planning skill. This enhanced version integrates with the self-learning skill to build on known project context and avoid re-asking previously answered questions.

Announce at start: "I'm using the brainstorming skill to explore this idea and create a design before implementation."

Trigger Conditions

• User describes a new feature idea or goal
• User asks "how should we build X?"
• Creative work requires design exploration before planning
• /brainstorm command invoked
• Transition from a vague requirement to structured design is needed

---

Phase 1: Context Loading

Goal: Load everything known about the project before asking the user anything.

1. Check memory files (project-context.md, learned-patterns.md, decisions-log.md) for known patterns, stack, and conventions
2. Read relevant existing code, docs, and recent commits
3. Review existing designs in docs/plans/ for related prior work
4. Identify constraints from the existing architecture
5. Note what you already know — do NOT re-ask the user for information you can discover

Context Sources Priority

Source	What to Extract	Priority
Memory files	Stack, conventions, past decisions	Highest
CLAUDE.md	Project structure, hard gates	High
Existing code	Current patterns, architecture	High
Recent commits	Direction of development	Medium
docs/plans/	Prior designs and decisions	Medium

STOP — Do NOT proceed to Phase 2 until:

• Memory files have been checked
• Relevant codebase areas have been explored
• You can summarize what you already know about this domain

---

Phase 2: Idea Exploration

Goal: Understand the user's intent through focused, one-at-a-time questions.

1. Ask ONE question per message — never multiple questions
2. Prefer multiple choice questions when possible
3. Focus on understanding purpose, constraints, and success criteria
4. Build on context loaded in Phase 1 — skip questions you can already answer
5. Convert vague answers into specific, testable criteria

Question Flow Decision Table

What You Need to Know	Question Type	Example
Core purpose	Open-ended	"What problem does this solve for the user?"
Scope boundaries	Multiple choice	"Should this handle: (A) only logged-in users, (B) all users, (C) depends on role?"
Technical constraints	Yes/No + follow-up	"Does this need to work offline?"
Priority trade-offs	Forced ranking	"Rank these: speed, correctness, simplicity"
Success criteria	Measurable target	"What does 'working' look like? Can you describe the happy path?"
Non-goals	Explicit exclusion	"What should we explicitly NOT build in this iteration?"

Question Rules

Rule	Rationale
One question per message	Prevents cognitive overload
Multiple choice preferred	Faster to answer, reduces ambiguity
Research before asking	Respect user's time
Build on memory	Do not re-ask known things
Testable outcomes	Vague success criteria lead to vague designs

STOP — Do NOT proceed to Phase 3 until:

• You understand the core purpose and who it serves
• You know the constraints (technical, timeline, scope)
• You have identified success criteria and non-goals
• No critical ambiguities remain about intent

---

Phase 3: Approach Exploration

Goal: Propose 2-3 distinct approaches with trade-offs and a clear recommendation.

For each approach, present:

Section	Content
Name	Short descriptive label
Architecture summary	2-3 sentences
Key trade-offs	Pros and cons
Complexity	Low / Medium / High
Risk	What could go wrong
Your assessment	Why you do or do not recommend it

Approach Comparison Template

## Approach A: [Name] (Recommended)
**Summary:** [2-3 sentences]
**Pros:** [list]
**Cons:** [list]
**Complexity:** [Low/Medium/High]
**Risk:** [What could go wrong]
**Why recommended:** [1-2 sentences]

## Approach B: [Name]
...

## Approach C: [Name] (if needed)
...

Lead with your recommended approach and explain why it is the best fit given the constraints discussed in Phase 2.

STOP — Do NOT proceed to Phase 4 until:

• You have proposed at least 2 approaches
• Each approach has explicit trade-offs
• You have made a clear recommendation
• User has indicated which approach to pursue

---

Phase 4: Design Presentation

Goal: Present the detailed design in sections, getting approval incrementally.

1. Present the design in logical sections scaled to complexity
2. Ask after each section whether it looks right so far
3. Be ready to go back and revise any section based on feedback
4. Cover all relevant design dimensions

Design Sections by Complexity

Complexity	Required Sections	Optional Sections
Simple (1-3 tasks)	Architecture, Components	—
Medium (4-10 tasks)	Architecture, Components, Data Flow, Error Handling	Testing Strategy
Complex (10+ tasks)	Architecture, Components, Data Flow, Error Handling, Testing Strategy, Performance, Security	Migration Plan

Section Presentation Order

1. Architecture — High-level structure and key decisions
2. Components — What pieces exist and how they relate
3. Data Flow — How data moves through the system
4. Error Handling — What can go wrong and how to handle it
5. Testing Strategy — How to verify correctness
6. Performance — Only if relevant constraints exist
7. Security — Only if handling sensitive data or auth

Present one section at a time. Wait for user confirmation before proceeding.

STOP — Do NOT proceed to Phase 5 until:

• All relevant design sections have been presented
• User has approved each section (or revisions were made)
• The complete design is coherent and addresses all requirements

---

Phase 5: Documentation and Transition

Goal: Persist the design and hand off to the planning skill.

1. Write the validated design to docs/plans/YYYY-MM-DD-<topic>-design.md
2. Commit the design document
3. Update self-learning memory:
   • memory/decisions-log.md — any architectural decisions made
   • memory/learned-patterns.md — any new conventions discussed
4. Invoke the planning skill to create a detailed implementation plan

Design Document Template

# [Topic] Design Document

**Date:** YYYY-MM-DD
**Status:** Approved
**Approach:** [Which approach was chosen]

## Problem Statement
[What problem this solves and for whom]

## Design

### Architecture
[High-level structure]

### Components
[Key pieces and their relationships]

### Data Flow
[How data moves through the system]

### Error Handling
[What can go wrong and how it is handled]

## Decisions Made
- [Decision 1]: [Rationale]
- [Decision 2]: [Rationale]

## Non-Goals
- [What was explicitly excluded and why]

## Next Steps
Invoke planning skill to create implementation plan.

STOP — Do NOT proceed until:

• Design document is saved to docs/plans/
• Memory files are updated with new decisions/patterns
• User has confirmed the design is complete

---

Anti-Patterns / Common Mistakes

Anti-Pattern	Why It Fails	Correct Approach
"This is too simple to brainstorm"	Every project needs a design, even simple ones	Design can be brief, but must exist
"The user already knows what they want"	Users know WHAT, not HOW	Explore the HOW through approaches
"I can just start coding"	Code without design is technical debt from line 1	Design first, code second
"We don't have time to brainstorm"	We don't have time to rebuild from poor assumptions	Brainstorming prevents costly rework
"The requirements are clear"	Requirements are not design — you still need approaches	Explore trade-offs even with clear requirements
Asking 5 questions at once	Overwhelms the user, gets shallow answers	One question per message
Skipping context loading	Re-asks things already known	Check memory files first
Not proposing alternatives	Anchors on first idea, misses better options	Always propose 2-3 approaches
Presenting entire design at once	Too much to review, user skims	Section by section with approval
Not persisting decisions	Same discussions repeat in future sessions	Update memory files

---

Anti-Rationalization Guards

If you catch yourself thinking:

• "Let me just scaffold the project first..." — No. Design first.
• "The design is obvious..." — Then it will be quick to document. Do it.
• "The user seems impatient..." — Poor design wastes more time than brainstorming.

---

Integration Points

Skill	Relationship	When
self-learning	Upstream — provides project context from memory	Phase 1: context loading
planning	Downstream — receives approved design	Phase 5: transition to planning
task-decomposition	Complementary — breaks design into work breakdown	When design reveals complex scope
spec-writing	Complementary — can formalize design into specs	When formal specifications are needed
verification-before-completion	Downstream — verifies design completeness	Before claiming design is done

---

Concrete Examples

Example: Simple Feature Brainstorm

User: "I want to add dark mode to the app"

Phase 1: [Check memory — React app, Tailwind CSS, no current theme system]
Phase 2: "Should dark mode (A) follow system preference automatically,
          (B) be a manual toggle only, or (C) both with manual override?"
Phase 3: Approach A: CSS variables + Tailwind dark: prefix
         Approach B: Theme context provider with CSS-in-JS
         Recommend A — aligns with existing Tailwind usage
Phase 4: Section 1: Architecture — Tailwind dark mode with class strategy
         Section 2: Components — ThemeToggle component, layout wrapper
Phase 5: Save design doc, invoke planning skill

Example: Transition to Planning

Design approved and saved to docs/plans/2026-03-15-dark-mode-design.md.
Updated memory/decisions-log.md with theme system decision.
Invoking planning skill to create implementation plan.

---

Key Principles

• One question at a time — Do not overwhelm
• Multiple choice preferred — Easier to answer
• YAGNI ruthlessly — Remove unnecessary features
• Explore alternatives — Always propose 2-3 approaches
• Incremental validation — Present and approve in sections
• Build on context — Use self-learning memory to avoid re-asking known things

---

Skill Type

RIGID — Follow this process exactly. Every idea goes through all five phases. The design can be brief for simple projects, but it must be documented and approved before any implementation begins.