minimal-abstractions（最小抽象）

Minimal Abstractions

Quick Start

Before adding a new abstraction (interface, abstract class, wrapper, layer), ask:

1. Does a similar abstraction already exist in this project?
2. Are there 2+ concrete use cases requiring this abstraction RIGHT NOW?
3. Is the complexity cost justified by actual flexibility needs?

If any answer is "no" or "maybe" → Don't add it. Use the simplest solution that works.

Table of Contents

1. When to Use This Skill
2. What This Skill Does
3. Core Philosophy
4. Abstraction Evaluation Checklist
5. Detection Patterns (Red Flags)
6. Examples: Good vs Over-Engineered
7. Integration with Architecture Validation
8. Expected Outcomes
9. Red Flags to Avoid

When to Use This Skill

Explicit Triggers (User asks):

• "Is this abstraction necessary?"
• "Are we over-engineering this?"
• "Too many layers in this code"
• "Simplify this architecture"
• "Reduce complexity"
• "Do we need this interface/wrapper/layer?"
• "Should we use [design pattern]?"

Implicit Triggers (Autonomous invocation):

• Reviewing pull requests with new interfaces/abstract classes
• Architecture proposals introducing new layers
• Code reviews where new patterns are introduced
• Refactoring tasks aimed at simplification
• When a new abstraction is proposed and only one implementation exists

Debugging/Problem Triggers:

• Maintenance burden is high due to abstraction complexity
• Team members confused by excessive indirection
• Tests are difficult to write due to layering
• Simple changes require touching multiple abstraction layers

What This Skill Does

This skill provides a systematic framework for:

• Evaluating whether new abstractions are warranted
• Detecting over-engineering and unnecessary complexity
• Guiding developers to use existing project abstractions
• Preventing abstraction proliferation
• Simplifying code by removing unneeded layers

Instructions

When evaluating an abstraction:

1. Search for existing abstractions - Use Grep/Read to find similar patterns in codebase
2. Apply the evaluation checklist - Run through all 5 questions in section 4
3. Check for red flags - Scan code for patterns in section 5
4. Recommend simpler alternative - Provide concrete code example
5. Document decision - Explain why abstraction is/isn't needed

Core Philosophy

The Abstraction Principle

Abstractions are not free - every interface, wrapper, layer, or pattern adds:

• Cognitive load (developers must understand the abstraction)
• Maintenance burden (more code to test, debug, refactor)
• Indirection cost (harder to trace execution flow)
• Rigidity (abstractions create assumptions that resist change)

When abstractions ARE valuable:

• Multiple concrete implementations exist or are planned imminently
• Decoupling is critical (e.g., domain from infrastructure in Clean Architecture)
• Testability requires dependency injection
• Third-party integrations need isolation

When abstractions are NOT valuable:

• "We might need it someday" (YAGNI - You Aren't Gonna Need It)
• "It's a best practice" (without understanding why)
• Only one implementation exists and no others are planned
• Wrapping for wrapping's sake

Prefer Existing Abstractions

Before creating a new abstraction:

1. Search the codebase for similar patterns
2. Extend or adapt existing abstractions
3. Reuse project-standard patterns (e.g., Repository, Service, Handler)
4. Only create new abstractions when existing ones truly don't fit

Abstraction Evaluation Checklist

Use this checklist before adding ANY new abstraction (interface, abstract class, wrapper, layer):

Question 1: Does this abstraction already exist?

• Searched codebase for similar interfaces/abstractions
• Checked project architecture docs for standard patterns
• Reviewed existing layers (domain, application, infrastructure)
• Considered extending existing abstractions instead

Action: If similar abstraction exists → Use it. Don't create a new one.

Question 2: Do we have 2+ concrete implementations RIGHT NOW?

• Count current implementations (not hypothetical future ones)
• Verify implementations are actually different (not just copy-paste)
• Check if "multiple implementations" are really needed

Action: If <2 implementations → Skip the abstraction. Use concrete implementation directly.

Question 3: Is there a concrete business/technical requirement?

• Can you name the specific requirement driving this abstraction?
• Is the requirement current (not speculative)?
• Would removing this abstraction make the code harder to maintain?

Action: If requirement is speculative → Don't build it yet. Wait for actual need.

Question 4: What is the complexity cost?

• Count files touched to add this abstraction
• Estimate lines of code added (interface + implementations + tests)
• Consider cognitive load for new team members
• Evaluate impact on debugging and tracing

Action: If cost > benefit → Simplify. Use direct solution.

Question 5: Can we solve this with simpler patterns?

• Would a simple function work instead of an interface?
• Could dependency injection handle this without abstraction?
• Is this trying to solve a problem we don't have?

Action: If simpler solution exists → Use it.

Detection Patterns (Red Flags)

Red Flag 1: Lonely Interface

# ❌ Over-engineered: Interface with only one implementation
class DataProcessor(Protocol):
    def process(self, data: dict) -> dict: ...

class JsonDataProcessor:  # Only implementation
    def process(self, data: dict) -> dict:
        return transform_json(data)

Why it's a red flag: No actual polymorphism. The interface adds indirection without benefit.

Better approach:

# ✅ Simple: Direct implementation
def process_json_data(data: dict) -> dict:
    return transform_json(data)

Red Flag 2: Wrapper with No Value

# ❌ Over-engineered: Wrapper that just forwards calls
class DatabaseWrapper:
    def __init__(self, db: Database):
        self._db = db

    def query(self, sql: str) -> list:
        return self._db.query(sql)  # Just forwarding

Why it's a red flag: No transformation, validation, or added behavior. Pure indirection.

Better approach: Use the database directly or add actual value (caching, retry, validation).

Red Flag 3: Layer for Layer's Sake

# ❌ Over-engineered: Unnecessary service layer
class UserRepository:  # Already exists
    def get_user(self, id: int) -> User: ...

class UserService:  # Adds nothing
    def __init__(self, repo: UserRepository):
        self.repo = repo

    def get_user(self, id: int) -> User:
        return self.repo.get_user(id)  # Just forwarding

Why it's a red flag: Service layer adds no business logic, validation, or orchestration.

Better approach: Use repository directly until business logic is needed.

Red Flag 4: Pattern for Pattern's Sake

# ❌ Over-engineered: Factory for single type
class UserFactory:
    @staticmethod
    def create_user(name: str, email: str) -> User:
        return User(name=name, email=email)

Why it's a red flag: Factory pattern used without variation or complexity justification.

Better approach:

# ✅ Simple: Direct construction
user = User(name="Alice", email="alice@example.com")

Red Flag 5: Premature Generalization

# ❌ Over-engineered: Generic solution for specific problem
class ConfigLoader(Generic[T]):
    def load(self, source: str, parser: Parser[T]) -> T: ...

class JsonParser(Parser[dict]): ...
class YamlParser(Parser[dict]): ...

Why it's a red flag: Generic abstraction built before knowing actual requirements.

Better approach: Start with simple JSON config loader. Generalize when second format is needed.

Examples: Good vs Over-Engineered

Example 1: Repository Pattern

Over-Engineered:

# Unnecessary: Abstract repository + generic base + implementation
class Repository(Protocol, Generic[T]):
    def get(self, id: int) -> T: ...
    def save(self, entity: T) -> None: ...

class BaseRepository(Generic[T]):  # Generic base
    def validate(self, entity: T) -> bool: ...

class UserRepository(BaseRepository[User]):  # Concrete
    def get(self, id: int) -> User: ...
    def save(self, user: User) -> None: ...

Right-Sized:

# Clean Architecture: Protocol in domain, implementation in infrastructure
# domain/repositories.py
class UserRepository(Protocol):  # Interface for dependency inversion
    def get_user(self, id: int) -> User: ...
    def save_user(self, user: User) -> None: ...

# infrastructure/repositories.py
class SqlUserRepository:  # Concrete implementation
    def get_user(self, id: int) -> User: ...
    def save_user(self, user: User) -> None: ...

Example 2: Service Layer

Over-Engineered:

# Unnecessary: Service that just forwards to repository
class UserService:
    def __init__(self, repo: UserRepository):
        self.repo = repo

    def get_user(self, id: int) -> User:
        return self.repo.get_user(id)  # No business logic!

Right-Sized:

# Use repository directly until business logic emerges
class AuthenticationHandler:
    def __init__(self, user_repo: UserRepository):
        self.user_repo = user_repo

    def authenticate(self, email: str, password: str) -> Result[User, AuthError]:
        user = self.user_repo.get_user_by_email(email)
        if not user:
            return Err(AuthError.USER_NOT_FOUND)
        if not verify_password(password, user.password_hash):
            return Err(AuthError.INVALID_PASSWORD)
        return Ok(user)

Example 3: Reusing Existing Abstractions

Over-Engineered:

# Project already has Repository pattern
# Adding NEW abstraction for similar purpose:
class DataAccessLayer(Protocol):  # Duplicates Repository!
    def fetch(self, id: int) -> Entity: ...
    def persist(self, entity: Entity) -> None: ...

Right-Sized:

# Use existing Repository pattern
class ProductRepository(Protocol):  # Follows project convention
    def get_product(self, id: int) -> Product: ...
    def save_product(self, product: Product) -> None: ...

Integration with Architecture Validation

This skill complements existing architecture validation skills:

Use with:

• architecture-validate-architecture - Check layer boundaries while avoiding unnecessary layers
• architecture-validate-layer-boundaries - Ensure layers are necessary and well-justified
• quality-code-review - Evaluate abstractions during PR review

Integration pattern:

1. Run architecture validation to check existing patterns
2. Use minimal-abstractions to evaluate NEW abstractions
3. Ensure new code follows project patterns (don't reinvent)

Expected Outcomes

Successful Simplification

Before:

src/
├── domain/
│   ├── interfaces/user_repository.py
│   ├── interfaces/user_service.py
│   ├── interfaces/user_validator.py
├── application/
│   ├── services/user_service.py (forwards to repo)
│   ├── validators/user_validator.py (just calls validate())
├── infrastructure/
│   ├── repositories/user_repository.py

After (applying minimal-abstractions):

src/
├── domain/
│   ├── repositories.py (UserRepository protocol)
│   ├── models.py (User with validation)
├── application/
│   ├── handlers.py (CreateUserHandler with actual business logic)
├── infrastructure/
│   ├── repositories.py (SqlUserRepository)

Metrics:

• 40% fewer files
• 60% less indirection
• Same functionality
• Clearer execution paths

Validation Output

Abstraction Evaluation: ProductService

✅ Checklist Results:
  ❌ Does abstraction already exist? YES - Repository pattern exists
  ❌ 2+ implementations? NO - Only one service planned
  ❌ Concrete requirement? NO - "We might need microservices later"
  ⚠️  Complexity cost: +3 files, +200 LOC, +2 layers indirection
  ✅ Simpler solution exists? YES - Use repository + handler directly

Recommendation: SKIP THIS ABSTRACTION
  - Use existing ProductRepository
  - Add business logic to ProductHandler
  - Wait for concrete multi-service requirement before abstracting

Red Flags to Avoid

Anti-Patterns

• ❌ "We might need it later" (YAGNI violation)
• ❌ Creating interfaces with only one implementation
• ❌ Wrappers that just forward calls without adding value
• ❌ Service layers that add no business logic
• ❌ Generic solutions for specific problems
• ❌ Design patterns used without understanding why
• ❌ Creating new abstractions when project patterns exist

Good Practices

• ✅ Use existing project abstractions first
• ✅ Wait for 2+ concrete implementations before abstracting
• ✅ Justify every layer with concrete requirements
• ✅ Prefer simple, direct solutions
• ✅ Question every new abstraction
• ✅ Measure complexity cost vs benefit
• ✅ Remove abstractions when they're no longer needed

Notes

Key Principle: Every abstraction must justify its existence with concrete, current requirements - not hypothetical future needs.

Balance: This skill advocates for minimal abstractions, but respects architectural patterns when they provide real value (e.g., Clean Architecture's dependency inversion).

When in doubt: Start simple. Add abstractions when pain points emerge, not before.