architecting-solutions（架构解决方案）

Architecting Solutions

Analyzes requirements and creates detailed PRD documents for software implementation.

Description

Use this skill when you need to:

• Create PRD documents
• Design software solutions
• Analyze requirements
• Specify features
• Document technical plans
• Plan refactoring or migration

Installation

This skill is typically installed globally at ~/.claude/skills/architecting-solutions/.

How It Works

The skill guides Claude through a structured workflow:

1. Clarify requirements - Ask targeted questions to understand the problem
2. Analyze context - Review existing codebase for patterns and constraints
3. Design solution - Propose architecture with trade-offs considered
4. Generate PRD - Output markdown PRD to {PROJECT_ROOT}/docs/ directory

IMPORTANT: Always write PRD to the project's docs/ folder, never to plan files or hidden locations.

Workflow

Copy this checklist and track progress:

Requirements Analysis:
- [ ] Step 1: Clarify user intent and success criteria
- [ ] Step 2: Identify constraints (tech stack, timeline, resources)
- [ ] Step 3: Analyze existing codebase patterns
- [ ] Step 4: Research best practices (if needed)
- [ ] Step 5: Design solution architecture
- [ ] Step 6: Generate PRD document (must be in {PROJECT_ROOT}/docs/)
- [ ] Step 7: Validate with user

Step 1: Clarify Requirements

Ask these questions to understand the problem:

Core Understanding

• Problem Statement: What problem are we solving? What is the current pain point?
• Success Criteria: How do we know this is successful? Be specific.
• Target Users: Who will use this feature? What are their goals?

For Refactoring/Migration:

• Why Refactor?: What's wrong with current implementation? Be specific.
• Breaking Changes: What will break? What needs migration?
• Rollback Plan: How do we revert if something goes wrong?

Step 2: Identify Constraints

• Technical Constraints: Existing tech stack, architecture patterns, dependencies
• Time Constraints: Any deadlines or phases?
• Resource Constraints: Team size, expertise availability
• Business Constraints: Budget, external dependencies, third-party APIs

Step 3: Analyze Existing Codebase

# Find similar patterns in the codebase
grep -r "related_keyword" packages/ --include="*.ts" --include="*.tsx"

# Find relevant directory structures
find packages/ -type d -name "*keyword*"

# Check existing patterns
ls -la packages/kit/src/views/similar-feature/

Critical for Refactoring:

• Find ALL consumers of the code being changed
• Identify ALL state/data flows
• Trace ALL entry points and exit points
• Look for existing mechanisms that might solve the problem already

# Find all imports/usages of a module
grep -r "useFeatureContext" packages/ --include="*.ts" --include="*.tsx"
grep -r "refreshSignalRef" packages/ --include="*.ts" --include="*.tsx"

CRITICAL: Before proposing a refactoring, ask:

1. Is there an existing mechanism that can be extended?
2. What's the simplest possible solution?
3. Can we solve this with minimal changes?
4. Does my solution actually connect the dots? (e.g., empty callbacks won't work)

Look for:

• Architectural patterns: How are similar features implemented?
• State management: What state solution is used? (Jotai, Redux, Context, Refs)
• Component patterns: How are components organized?
• API patterns: How are API calls structured?
• Type definitions: Where are types defined?

Step 4: Research Best Practices

For unfamiliar domains, search for best practices.

Step 5: Design Solution Architecture

CRITICAL: Consider Multiple Solutions

Before settling on a solution, ALWAYS present multiple options:

1. Minimal Change Solution - What's the absolute smallest change that could work?
2. Medium Effort Solution - Balanced approach with some refactoring
3. Comprehensive Solution - Full architectural overhaul

Example:

Problem: Data doesn't refresh after operation

Option 1 (Minimal): Hook into existing pending request count decrease
  - Changes: 1-2 files
  - Risk: Low
  - Selected: ✓

Option 2 (Medium): Add refresh callback through existing shared context
  - Changes: 3-5 files
  - Risk: Medium

Option 3 (Comprehensive): Migrate to a centralized state-store pattern
  - Changes: 10+ files, new atoms/actions
  - Risk: High
  - Time: 2-3 days

Ask user BEFORE writing PRD:

• Which option do you prefer?
• Are you open to larger refactoring?
• What's your tolerance for change?

Architecture Design Principles

1. Simplicity First: Choose the simplest solution that meets requirements
2. Progressive Enhancement: Start with MVP, extend iteratively
3. Separation of Concerns: UI, logic, and data should be separated
4. Reusability: Design components that can be reused
5. Testability: Design for easy testing

Document Trade-offs

For each major decision, document:

Step 6: Generate PRD Document

IMPORTANT: Always write PRD to the project's docs/ directory, never to plan files or hidden locations.

Output location: {PROJECT_ROOT}/docs/{feature-name}-prd.md

Example:

• If project root is /Users/user/my-project/, write to /Users/user/my-project/docs/feature-name-prd.md
• Use kebab-case for filename: data-refresh-logic-refactoring-prd.md

Step 7: Validate with User

Before finalizing:

1. Review success criteria - Do they align with user goals?
2. Check constraints - Are all constraints addressed?
3. Verify completeness - Can another agent implement from this PRD?
4. Confirm with user - Get explicit approval before finalizing

PRD Quality Checklist

Content Quality

• Problem statement is clear and specific
• Success criteria are measurable
• Functional requirements are unambiguous
• Non-functional requirements are specified
• Constraints are documented
• Trade-offs are explained

Implementation Readiness

• Architecture is clearly defined
• File structure is specified
• API contracts are defined (if applicable)
• Data models are specified
• Edge cases are considered
• Testing approach is outlined

Agent-Friendliness

• Another agent can implement without clarification
• Code examples are provided where helpful
• File paths use forward slashes
• Existing code references are accurate

Root Cause Analysis Checklist (CRITICAL)

For bugs and refresh issues, ALWAYS verify:

• Existing mechanism already exists - Does a working solution exist elsewhere?
• Why existing solution doesn't work - Timing? Scope? Not connected?
• Each hook/component instance is independent - They don't share state unless explicitly connected
• Callback chain is complete - Trace from trigger to effect, every link must work
• Empty callbacks are called - If onRefresh is provided, is it actually implemented?
• Polling/refresh timing - What are the intervals? When do they fire?

Common Root Cause Mistakes:

• Assuming hooks share state (they don't - each instance is independent)
• Empty callback implementations that do nothing
• Not tracing the full call chain from trigger to effect
• Not understanding when events fire (e.g., revalidateOnFocus requires actual focus change)

Migration Scope Completeness

• ALL existing state is accounted for: List every piece of state being migrated

- What states are being migrated? (e.g., items, summary, isLoading, filters, pendingRequests) - What's the migration strategy for each? (direct move / transform / deprecate)

• ALL consumers are identified: Find every file that uses the code being changed # Must run: grep -r "import.*ModuleName" packages/ # Must run: grep -r "useHookName" packages/
• Provider usage points are covered: Every file using the Provider is updated

- Root Provider → Mirror Provider migration - All pages/components using the provider

State/Data Flow Validation

• No orphaned state: Every piece of state has a clear source and consumer
• No dead state: Every new state/state variable has a defined purpose and consumer
• No undefined references: All imports/references resolve to existing code
• Complete call chain documented: From trigger → callback → effect, show every step
• All related operations covered: If module has Create/Edit/Delete/Import/Export, test all of them

React/Hook Rules Compliance

• No conditional hooks: Never call hooks conditionally (e.g., isAdvancedMode? useHook(): null)

- Hooks MUST be called at the top level, unconditionally - If conditional logic is needed, use early return or conditional rendering

• Ref usage is correct: If using ref pattern, access via .current

- Check: useFeatureActions().current not useFeatureActions()

Provider Pattern Completeness

• Root Provider is defined: Main Provider component exists
• Mirror Provider is defined: Mirror Provider for modal/overlay contexts exists
• All usage points wrapped: Every page/component using the provider is wrapped # Must verify: Each page that uses the context has the Provider wrapper

Auto-mount/System Integration

• Enum registration: Added to appropriate enum (e.g., EContextStoreNames)
• Switch case registration: Added to auto-mount switch statement
• Store initialization: Store initialization logic is complete
• No duplicate registrations: Verify no conflicts with existing entries

Backward Compatibility

• Existing consumers work: Code using the old pattern still works during migration
• Migration path is clear: How do consumers migrate to the new pattern?
• Deprecation timeline: When is the old pattern removed?

Code Examples

• Before/After comparisons: Show code changes clearly
• Type definitions are accurate: TypeScript types match the implementation
• Import paths are correct: All imports use correct workspace paths

Common Anti-Patterns to Avoid

Over-Engineering Warning (Critical Lesson)

The Problem with Jumping to Complex Solutions

Real Case Study:

• PRD Proposed: Full shared state-store migration (10+ files, 2-3 days)
• Actual Solution: Hook into existing pending request count decrease (1-2 files, 1 hour)
• Lesson: Always look for the simplest solution first

Signs You Might Be Over-Engineering

• ❌ Proposing new patterns when existing ones could work
• ❌ Creating new state management before exhausting current options
• ❌ Multiple new files when one file change could suffice
• ❌ "Best practice" justification without considering practicality

Questions to Ask Before Writing PRD

1. Is there an existing mechanism that does 80% of what we need?
2. Can we extend/modify existing code instead of creating new patterns?
3. What's the absolute minimum change to solve THIS problem?
4. Does the user actually want a major refactor?
5. Does my solution's callback actually do something? (Empty callbacks are bugs!)
6. Have I traced the complete call chain? (Trigger →... → Effect)

When Comprehensive Solutions ARE Appropriate

• Current architecture is fundamentally broken
• Technical debt is blocking all new features
• Team has explicitly decided to modernize
• Problem will recur if not properly addressed

Key: Comprehensive solutions should be a CHOICE, not the DEFAULT.

Patterns for Common Scenarios

New Feature Implementation

1. Read similar feature implementations
2. Identify reusable patterns
3. Design component hierarchy
4. Define state management approach
5. Specify API integration points
6. List all new files to create
7. List all existing files to modify

Refactoring Existing Code

1. Analyze current implementation
2. Find ALL consumers (grep -r imports)
3. Identify pain points and technical debt
4. PROPOSE MULTIPLE SOLUTIONS (minimal → comprehensive)
5. GET USER APPROVAL on approach
6. Plan migration strategy (phased vs big-bang)
7. Define rollback approach
8. List migration checklist

# CRITICAL: Start with the simplest solution!
# Only propose comprehensive refactoring if user explicitly wants it.

Bug Fix Investigation

1. Understand expected vs actual behavior
2. Locate root cause in code
3. Identify affected areas
4. Design fix approach
5. Specify testing for regression prevention

Reference Materials

• PRD Template: Look at existing PRDs in the project's docs/ folder
• Similar Implementations: Reference similar features/modules in the codebase

Tips for Effective PRDs

1. Be Specific: "Improve performance" → "Reduce API response time from 2s to 500ms"
2. Show Context: Explain why a decision was made, not just what was decided
3. Include Examples: Show code snippets for complex patterns
4. Think About Edge Cases: What happens when API fails? User has no data?
5. Consider Migration: For refactoring, how do we move from A to B safely?
6. List ALL Changes: For refactoring, list every file that changes
7. Validate Imports: Verify all import paths exist and are correct
8. Check Hook Rules: Ensure no conditional hooks, proper hook dependencies

Accuracy & Completeness (Critical Lessons from Real PRD Reviews)

Technical Terms - Be Precise

Call Chain Documentation - Don't Skip Steps

Bad: "onRefresh triggers data refresh" Good:

onRefresh() → navigation.goBack() → Dashboard focused
  → usePromiseResult (revalidateOnFocus: true) fires
  → refreshItems() → handleRefresh()
  → fetchItems() + refreshSummary() + refreshMetrics()

Include file paths and line numbers for each step!

Test Coverage - Cover ALL Operations

If module has 5 operations (Create/Edit/Delete/Import/Export), test all 5. Don't just test the 2 you're focused on.

Timeline Analysis for Refresh/Timing Issues

Draw out the timeline:

0s  ---- Modal opens, user starts Edit
10s ---- Action submitted, pending: 0→1
15s ---- Modal closes
        └─ Dashboard hook last polled at 5s
        └─ Next poll at 35s (25s away!) ❌

This shows WHY it doesn't work.

Common PRD Mistakes