brainstorming
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.
- Installs
- 78
- Last 24h
- —
- Trending rank
- —
- Security
- Not audited
Install
Works with every agent the skills CLI supports.
npx skills add https://github.com/pixel-process-ug/superkit-agents --skill brainstorming- 01
Install
Run the command in your project. The CLI asks which agents to install for.
- 02
Describe the task
Ask your agent for something this skill covers. It loads brainstorming automatically when relevant.
- 03
Review the result
The agent follows the SKILL.md steps, scripts and references bundled with the skill.
SKILL.md
The instructions your agent reads when the skill is triggered.
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
/brainstormcommand 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.
- Check memory files (
project-context.md,learned-patterns.md,decisions-log.md) for known patterns, stack, and conventions - Read relevant existing code, docs, and recent commits
- Review existing designs in
docs/plans/for related prior work - Identify constraints from the existing architecture
- 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.
- Ask ONE question per message — never multiple questions
- Prefer multiple choice questions when possible
- Focus on understanding purpose, constraints, and success criteria
- Build on context loaded in Phase 1 — skip questions you can already answer
- 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.
- Present the design in logical sections scaled to complexity
- Ask after each section whether it looks right so far
- Be ready to go back and revise any section based on feedback
- 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
- Architecture — High-level structure and key decisions
- Components — What pieces exist and how they relate
- Data Flow — How data moves through the system
- Error Handling — What can go wrong and how to handle it
- Testing Strategy — How to verify correctness
- Performance — Only if relevant constraints exist
- 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.
- Write the validated design to
docs/plans/YYYY-MM-DD-<topic>-design.md - Commit the design document
- Update self-learning memory:
memory/decisions-log.md— any architectural decisions madememory/learned-patterns.md— any new conventions discussed
- Invoke the
planningskill 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
<HARD-GATE> Do NOT invoke any implementation skill, write any code, scaffold any project, or take any implementation action until you have presented a design and the user has approved it. This applies to EVERY project regardless of perceived simplicity. </HARD-GATE>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.
FAQ
What does the brainstorming skill do?
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.
How do I install brainstorming?
Run npx skills add https://github.com/pixel-process-ug/superkit-agents --skill brainstorming in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/brainstorming/ (personal) or .claude/skills/brainstorming/ (project).
Which AI agents can use brainstorming?
Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.
Is brainstorming safe to install?
No partner security audit is available yet. Skills can run commands and scripts, so read the SKILL.md and any bundled files before installing.
Who maintains brainstorming?
It is published by pixel-process-ug in pixel-process-ug/superkit-agents. It has 78 recorded installs.
More from pixel-process-ug/superkit-agents
ui-ux-pro-max
pixel-process-ug/superkit-agents
Use when the user needs full UI/UX design intelligence — styles, palettes, fonts, UX guidelines, chart selection, and accessible, performant implementation across any supported stack. Triggers: user says "design", "UI", "UX", "color palette", "typography", "accessibility", "responsive design", "chart type", "style guide", building any user-facing interface.
frontend-ui-design
pixel-process-ug/superkit-agents
Use when designing UI components, creating component architectures, implementing responsive layouts, setting up design systems, or selecting state management solutions for frontend applications
reverse-engineering-specs
pixel-process-ug/superkit-agents
Use when onboarding to an existing codebase that lacks specifications — exhaustively traces code paths and produces implementation-free behavioral specifications for safe refactoring, feature addition, or legacy modernization
canvas-design
pixel-process-ug/superkit-agents
Use when the user needs HTML Canvas, SVG graphics, data visualization, generative art, or interactive visual elements with D3.js or similar libraries. Triggers: user says "canvas", "SVG", "chart", "data visualization", "D3", "generative art", "animation", "interactive graphics", drawing on screen.
autonomous-loop
pixel-process-ug/superkit-agents
Use when running Ralph-style iterative autonomous development. Triggers on /ralph or /loop commands, when autonomous iterative development is needed, when a project has specs and an implementation plan ready for iterative execution, or when deterministic context loading with subagent delegation and dual-condition exit gates is required. Orchestrates PLANNING, BUILDING, and STATUS cycles.
mobile-design
pixel-process-ug/superkit-agents
Open the skill to read its SKILL.md instructions.
Popular code quality & workflow skills
grill-me
mattpocock/skills
A relentless interview to sharpen a plan or design.
improve-codebase-architecture
mattpocock/skills
Scan a codebase for deepening opportunities, present them as a visual HTML report, then grill through whichever one you pick.
handoff
mattpocock/skills
Compact the current conversation into a handoff document for another agent to pick up.
domain-modeling
mattpocock/skills
Build and sharpen a project's domain model. Use when discussing codebase terminology, writing or editing a GLOSSARY.md, or recording or editing an ADR.
codebase-design
mattpocock/skills
Shared vocabulary for designing deep modules. Use when the user wants to design or improve a module's interface, find deepening opportunities, decide where a seam goes, make code more testable or AI-navigable, or when another skill needs the deep-module vocabulary.
diagnosing-bugs
mattpocock/skills
Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.