brainstorm-design（头脑风暴设计）

Skill: Brainstorm Design

Purpose

Transform rough ideas into validated, production-grade designs through systematic collaborative dialogue. Prevent premature implementation by exploring context, clarifying requirements, proposing alternatives with trade-offs, and obtaining explicit approval before any code is written.

Core Objective

Primary Goal: Produce a validated design document that serves as the single source of truth for implementation.

Success Criteria (ALL must be met):

1. ✅ Design document exists: Written to docs/design-decisions/YYYY-MM-DD-<topic>.md and committed to version control
2. ✅ User explicitly approved: User said "approved", "looks good", "proceed", or equivalent confirmation
3. ✅ Alternatives documented: At least 2-3 approaches considered with trade-offs analysis
4. ✅ YAGNI applied: Design focuses on minimum viable solution, unnecessary features removed
5. ✅ DRY applied: Design references existing patterns/components rather than reinventing
6. ✅ No code written: Zero implementation code exists (design only)

Acceptance Test: Can a developer with zero project context implement this design without asking clarifying questions?

Scope Boundaries

This skill handles:

• Rough idea → Validated design document
• Requirement clarification through dialogue
• Alternative exploration and trade-off analysis
• Design approval and documentation

This skill does NOT handle:

• Implementation planning (use writing-plans or similar)
• Code writing (use implementation skills)
• Testing strategy details (mention in design, detail in implementation plan)
• Deployment planning (out of scope)

Handoff point: When design is approved and documented, hand off to implementation planning or development workflow.

Use Cases

• Feature planning: User has rough idea for new functionality; needs help refining requirements and design approach.
• Architecture decisions: Team needs to explore multiple technical approaches with clear trade-offs before committing.
• Requirement clarification: Stakeholder request is vague or incomplete; extract concrete specifications through dialogue.
• Design validation: Existing design draft needs review and user approval before development.
• Best practices application: Apply domain-specific best practices (React, Next.js, Remotion, etc.) during design phase.

Behavior

Interaction Policy

• Defaults: Design path from project norms or docs/design-decisions/; present section by section
• Choice options: One question at a time; offer [A][B][C] for preferences
• Confirm: User must say "approved" or equivalent before handoff; no code until design approved

HARD-GATE: No Implementation Before Approval

DO NOT invoke implementation skills, write code, scaffold projects,
or take implementation actions until design is presented and approved.

This applies to EVERY project regardless of perceived simplicity.

Anti-Pattern: "This Is Too Simple To Need A Design"

Every project goes through this process. A todo list, single-function utility, config change — all require design. "Simple" projects are where unexamined assumptions cause most wasted work. Design can be short (few sentences for simple projects), but MUST be presented and approved.

Phase 1: Explore Context

Announce at start: "I'm using the brainstorm-design skill to refine this idea into a validated design."

1. Examine project state: Review relevant files, documentation, recent commits, existing architecture.
2. Identify constraints: Note technical limitations, dependencies, existing patterns, team conventions.
3. Check for best practices: If domain-specific (React, Next.js, etc.), note applicable best practices.
4. Surface assumptions: Document what is known vs. what needs clarification.

Phase 2: Clarify Through Dialogue

Core principle: One question at a time, building understanding incrementally.

1. Ask focused questions: Understand purpose, constraints, success criteria.
2. Prefer structured questions: Use multiple choice when possible; open-ended when exploration needed.
3. Build incrementally: Each answer informs next question; avoid overwhelming with multiple questions.
4. Validate understanding: Summarize key points periodically to confirm alignment.

Question patterns:

• Purpose: "What problem does this solve?" "Who is the user?"
• Constraints: "What are the technical limitations?" "What must we preserve?"
• Success criteria: "How will we know this works?" "What defines done?"

Phase 3: Explore Alternatives

Core principle: Always propose 2-3 approaches before settling.

1. Propose 2-3 distinct approaches: Present options with clear trade-offs.
2. Lead with recommendation: State your recommended option and explain reasoning.
3. Consider constraints: Ensure each approach addresses identified requirements and limitations.
4. Apply YAGNI ruthlessly: Remove unnecessary features from all designs; focus on minimum viable solution.
5. Invite feedback: User may prefer different approach or suggest hybrid solution.

Trade-off framework:

• Pros: What makes this approach strong?
• Cons: What are the downsides or risks?
• Best for: When is this the right choice?

Phase 4: Present Design

Core principle: Scale to complexity, validate incrementally.

1. Scale to complexity:

- Simple projects: Concise design (few sentences to 1 paragraph) - Medium projects: Structured sections (100-200 words each) - Complex projects: Detailed sections (200-300 words each)

1. Present incrementally: Show design in logical sections; validate each before proceeding.
2. Cover key aspects (adjust depth to complexity):

- Architecture: High-level structure, component relationships - Components: Key modules, their responsibilities - Data flow: How information moves through system - Error handling: How failures are managed - Testing strategy: How correctness is verified - Performance considerations: Optimization approach (if relevant) - Security considerations: Auth, data protection (if relevant)

1. Apply DRY principle: Don't repeat yourself; reference existing patterns.
2. Iterate as needed: Revise sections based on feedback; go back to clarify when needed.

Phase 5: Document and Transition

1. Resolve project norms: Check for .ai-cortex/artifact-norms.yaml or docs/ARTIFACT_NORMS.md per spec/artifact-norms-schema.md. If found, use project path for design; otherwise use default docs/design-decisions/YYYY-MM-DD-<topic>.md from spec/artifact-contract.md.
2. Write design document: Save to resolved path. Create directory if it does not exist.
3. Document structure (include YAML front-matter, see below): --- artifact_type: design created_by: brainstorm-design lifecycle: snapshot created_at: YYYY-MM-DD --- # [Feature Name] Design **Date:** YYYY-MM-DD **Status:** Approved **Approved by:** [User name or "User"] ## Goal [One sentence describing what this builds] ## Architecture [2-3 sentences about approach] ## Components [Key modules and responsibilities] ## Data Flow [How information moves] ## Error Handling [How failures are managed] ## Testing Strategy [How correctness is verified] ## Trade-offs Considered [Alternatives explored and why this approach chosen] ## Acceptance Criteria - [] [Specific, measurable criterion 1] - [] [Specific, measurable criterion 2] - [] [Specific, measurable criterion 3]
4. Commit to version control: Preserve validated design as project artifact.
5. Verify completion: Check all success criteria met:

- ✅ Design document exists and committed - ✅ User explicitly approved - ✅ Alternatives documented with trade-offs - ✅ YAGNI applied (unnecessary features removed) - ✅ DRY applied (references existing patterns) - ✅ No code written

1. Announce completion and handoff: "Design complete and approved. Saved to docs/design-decisions/YYYY-MM-DD-<topic>.md. Next steps: - For detailed implementation plan: Use writing-plans or similar skill - For immediate implementation: Proceed with development workflow - For review: Share design document with team Ready to proceed?"

Input & Output

Input

• Rough idea: User's initial concept, feature request, or problem statement.
• Project context: Existing codebase, documentation, constraints (discovered during exploration).
• User responses: Answers to clarifying questions, feedback on proposals, design approval.

Output

• Design document: Validated design specification covering architecture, components, data flow, error handling, testing.
• Documentation artifact: Markdown file committed to version control at agreed location.
• Implementation readiness: Clear handoff point where design is approved and ready for development.
• Trade-off analysis: Documented alternatives considered and reasoning for chosen approach.

Restrictions

Hard Boundaries

• No premature implementation: Do NOT write code, scaffold projects, or invoke implementation tools until design is approved.
• No assumption of simplicity: Every project goes through this process, regardless of perceived simplicity. "Simple" projects can have short designs, but must still be presented and approved.
• One question at a time: Do not overwhelm user with multiple questions in a single message.
• YAGNI ruthlessly: Remove unnecessary features and complexity from all designs; focus on minimum viable solution.
• DRY principle: Don't repeat yourself; reference existing patterns and components.
• Validate incrementally: Do not present entire design at once; get approval section by section for complex projects.
• No guessing: If you don't understand something, ask for clarification rather than assuming.

Skill Boundaries (Avoid Overlap)

Do NOT do these (other skills handle them):

• Implementation planning: Creating detailed task lists, file paths, exact code → Use writing-plans or similar
• Code writing: Writing actual implementation code → Use implementation/development skills
• Code review: Reviewing existing code for quality → Use review-code or similar
• Debugging: Investigating bugs or test failures → Use systematic-debugging or similar
• Testing: Writing test cases or test plans → Mention in design, detail in implementation phase
• Deployment: Planning deployment strategy → Out of scope for design phase

When to stop and hand off:

• User says "approved", "looks good", "proceed" → Design complete, hand off to implementation
• User asks "how do we implement this?" → Design complete, hand off to writing-plans or implementation
• User asks "can you write the code?" → Design complete, hand off to development workflow

Self-Check

Core Success Criteria (ALL must be met)

• Design document exists: Written to docs/design-decisions/YYYY-MM-DD-<topic>.md and committed
• User explicitly approved: User said "approved", "looks good", "proceed", or equivalent
• Alternatives documented: At least 2-3 approaches with trade-offs in design document
• YAGNI applied: Design focuses on minimum viable solution, unnecessary features removed
• DRY applied: Design references existing patterns/components rather than reinventing
• No code written: Zero implementation code exists (design only)

Process Quality Checks

• Context explored: Did I examine project state, constraints, and existing patterns before asking questions?
• Questions focused: Did I ask one question at a time, building understanding incrementally?
• Alternatives presented: Did I propose 2-3 distinct approaches with clear trade-offs and recommendation?
• Design validated: Did I present design in appropriately-sized sections and get user approval?
• Trade-offs documented: Did I document alternatives considered and reasoning for chosen approach?
• No premature code: Did I avoid writing any implementation code before design approval?
• No guessing: Did I ask for clarification when information was missing rather than assuming?

Acceptance Test

Can a developer with zero project context implement this design without asking clarifying questions?

If NO: Design is incomplete. Return to Phase 2 (Clarify) or Phase 4 (Present Design).

If YES: Design is complete. Proceed to handoff.

Examples

Example 1: Simple Feature - User Authentication

User: "I need to add login to my web app"

Agent announces: "I'm using the brainstorm-design skill to refine this idea into a validated design."

Agent explores context:

• Checks existing codebase (finds Express.js backend, React frontend)
• Notes no existing auth system
• Identifies database (PostgreSQL)

Agent asks clarifying questions (one at a time):

1. "What authentication method do you prefer? (A) Email/password, (B) OAuth (Google/GitHub), (C) Both"
2. "Do you need: (A) Basic login/logout only, (B) Password reset, (C) Email verification, (D) All of the above"
3. "For session management: (A) JWT tokens, (B) Server-side sessions, (C) No preference"

Agent proposes approaches:

• Option A (Recommended): Passport.js with local strategy + JWT. Pros: Well-tested, flexible, good docs. Cons: More setup than alternatives. Best for: Production apps needing flexibility.
• Option B: NextAuth.js. Pros: Simpler setup, built-in OAuth. Cons: Requires Next.js migration or separate auth service. Best for: Next.js projects.
• Option C: Custom implementation. Pros: Full control. Cons: Security risk, more maintenance. Best for: Unique requirements not covered by libraries.

Agent presents design (section by section, applying YAGNI):

1. Architecture: Passport.js middleware, JWT tokens, bcrypt for passwords
2. Components: Auth middleware, User model, Token service
3. Data flow: Client → POST /auth/login → Passport → JWT → Client stores token → Authenticated requests include token
4. Database: Users table with email, password_hash, created_at (YAGNI: no roles, no profiles yet)
5. API endpoints: POST /auth/register, POST /auth/login, GET /auth/me (YAGNI: no password reset yet)
6. Frontend: Auth context, protected routes, token storage in httpOnly cookies
7. Error handling: Rate limiting (5 attempts/15min), input validation, secure password requirements (min 8 chars)
8. Testing: Unit tests for auth logic, integration tests for endpoints

Agent documents: Writes to docs/design-decisions/2026-03-02-user-authentication.md with trade-offs section and commits.

Example 2: Complex Feature - Real-time Collaboration

User: "Add real-time collaboration like Google Docs"

Agent announces: "I'm using the brainstorm-design skill to refine this idea into a validated design."

Agent explores context:

• Examines existing document editor (CodeMirror-based)
• Notes current architecture (REST API, no WebSocket infrastructure)
• Identifies scale requirements (asks: "How many concurrent users per document?")

Agent asks clarifying questions:

1. "What collaboration features are essential? (A) Cursor positions only, (B) Live editing, (C) Comments/annotations, (D) All of the above"
2. "Expected concurrent users per document: (A) 2-5, (B) 5-20, (C) 20+, (D) Unknown"
3. "Conflict resolution preference: (A) Last-write-wins, (B) Operational Transform, (C) CRDT, (D) No preference"

Agent proposes approaches:

• Option A: Yjs (CRDT) + WebSocket. Pros: Proven, handles conflicts well, offline support. Cons: Learning curve, larger bundle (200KB). Best for: 5+ users, offline support needed.
• Option B (Recommended for 2-5 users): ShareDB (OT). Pros: Simpler than CRDT, good for small teams, smaller bundle (50KB). Cons: Doesn't scale as well. Best for: Small teams, simpler requirements.
• Option C: Custom WebSocket + simple locking. Pros: Lightweight (10KB). Cons: Poor UX with conflicts, hard to get right. Best for: Very simple use cases, single editor at a time.

Agent presents design (detailed sections, applying DRY):

1. Architecture: Yjs for CRDT, y-websocket for sync, y-codemirror for editor binding (DRY: reuse existing CodeMirror)
2. Infrastructure: WebSocket server (separate from REST API), Redis for presence (DRY: reuse existing Redis)
3. Components: Sync server, Presence service, Document store, Editor binding
4. Data model: Document state in Yjs format, periodic snapshots to PostgreSQL (every 5 min or 100 changes)
5. Data flow: Editor change → Yjs update → WebSocket → Server → Broadcast → Other clients → Apply update
6. Conflict resolution: Automatic via CRDT, no user intervention needed
7. Presence: Cursor positions, user colors (from existing user profiles), active user list
8. Performance: Lazy loading for large documents (>1MB), compression for network traffic (gzip), debounce snapshots
9. Error handling: Reconnection logic (exponential backoff), offline queue (max 1000 ops), sync conflict UI (rare, show merge dialog)
10. Testing: Multi-client simulation (2, 5, 10 users), network partition scenarios, data consistency checks, load testing (100 concurrent docs)

Agent documents: Writes to docs/design-decisions/2026-03-02-realtime-collaboration.md with architecture diagrams, trade-offs analysis, and commits.

Example 3: Edge Case - "Too Simple" Project

User: "Just need a function to format phone numbers"

Agent announces: "I'm using the brainstorm-design skill to refine this idea into a validated design."

Agent explores context:

• Checks if utility functions directory exists (finds src/utils/)
• Notes existing validation patterns in codebase (uses Zod for validation)

Agent asks clarifying questions:

1. "Which phone number formats? (A) US only, (B) International, (C) Specific countries"
2. "Input format: (A) Any string, (B) Already validated numbers, (C) User input from form"
3. "Output format: (A) Display format (e.g., (555) 123-4567), (B) Storage format (e.g., +15551234567), (C) Both"

Agent proposes approaches:

• Option A (Recommended): libphonenumber-js library. Pros: Handles international, well-tested, maintained. Cons: 100KB bundle size. Best for: International support needed.
• Option B: Simple regex for US only. Pros: Tiny (1KB), no dependencies. Cons: Breaks with international numbers, maintenance burden. Best for: US-only apps, bundle size critical.

Agent presents design (concise, applying YAGNI and DRY):

"Utility function using libphonenumber-js: accepts string input, validates format, returns formatted string or error. Handles US and international numbers. DRY: integrates with existing Zod validation patterns. YAGNI: no custom formatting rules, no phone number database, just format and validate. Includes unit tests for valid/invalid inputs and edge cases (empty string, partial numbers, invalid country codes, null/undefined)."

Agent documents: Writes brief design to docs/design-decisions/2026-03-02-phone-formatter.md with trade-offs (libphonenumber-js vs regex) and commits.

Example 4: Edge Case - Blocked by Missing Information

User: "Build a payment system"

Agent announces: "I'm using the brainstorm-design skill to refine this idea into a validated design."

Agent explores context:

• Checks existing codebase (finds e-commerce app)
• Notes no existing payment integration

Agent asks clarifying questions:

1. "Which payment providers? (A) Stripe, (B) PayPal, (C) Both, (D) Other"

User: "Stripe"

1. "What payment methods? (A) Credit cards only, (B) Credit cards + ACH, (C) All Stripe methods"

User: "Not sure, what do you recommend?"

Agent stops and asks for clarification:

"I need to understand your requirements better before proposing approaches. Key questions:

• Who are your customers? (B2B or B2C)
• What's your average transaction size?
• Do you need recurring payments (subscriptions)?
• Any specific compliance requirements (PCI, GDPR)?

Without this information, I can't recommend the right payment methods or architecture. Should we discuss these requirements first?"

Agent does NOT guess or proceed with assumptions.

Appendix: Output contract

This skill produces a document-artifact (design). Each output document MUST conform to: