yagni（yagni 搜索）

YAGNI — You Aren't Gonna Need It

Before Applying

If .agents/stack-context.md exists, read it first. Apply this principle using idiomatic patterns for the detected stack. For framework-specific details, use context7 MCP or web search — don't guess.

Principle

Do not build for hypothetical future requirements. Build what is needed now, and refactor when actual requirements emerge.

Why This Matters in Production

Speculative code is the #1 source of accidental complexity. Every abstraction, configuration option, or extension point you add "just in case" has a real cost: it must be understood, tested, maintained, and debugged. Unused code paths are the most dangerous — they rot silently, give false confidence in test coverage, and create surface area for bugs.

Premature generalization is worse than duplication. Duplication is obvious and easy to fix later. A wrong abstraction is painful to undo because other code grows to depend on it.

Rules

1. Solve the problem in front of you. If you have one use case, write code for one use case. Not two. Not "what if later."
2. Three strikes, then abstract. The first time you write something, just write it. The second time, note the duplication. The third time, extract the pattern — now you have enough data to design the right abstraction.
3. Delete speculative code paths. If a feature flag has never been toggled, a configuration option has never been changed, or a parameter has never been passed anything other than the default — remove it.
4. Don't build plugin systems for one plugin. Interfaces, registries, and extension points are justified only when you have multiple concrete implementations today.
5. Prototype when uncertain. If you aren't sure whether something will be needed, spike it in a branch. Don't merge speculative infrastructure into main.

Anti-Patterns

• Adding parameters "for flexibility" that only ever receive one value
• Building an event system when you have two components that could just call each other
• Creating abstract base classes with a single concrete implementation
• Writing configuration files for values that never change
• Adding database columns "we might need later"
• Implementing caching before measuring whether there's a performance problem
• Building a microservice when a function call would suffice

Examples

-- YAGNI violation: generic "processor" for one operation
class DataProcessor:
    def __init__(self, strategy, validator, transformer, output_format):
        self.strategy = strategy
        ...

-- Actually needed: one function
def process_csv_upload(file):
    rows = parse_csv(file)
    validate_rows(rows)
    save_to_db(rows)

-- YAGNI violation: premature abstraction
interface INotificationService
class EmailNotificationService implements INotificationService
class SMSNotificationService implements INotificationService  // "we might need this"
class PushNotificationService implements INotificationService  // "just in case"

-- Actually needed: you only send emails today
def send_welcome_email(user):
    mailer.send(to=user.email, template="welcome")

Boundaries

• YAGNI does not mean ignore architecture. Good structure (separation of concerns, clear module boundaries) is not speculative — it makes future changes cheaper. YAGNI targets unused features, not good design.
• YAGNI does not mean skip error handling. Handling known failure modes (network errors, invalid input, disk full) is not speculative — those things will happen in production.
• YAGNI does not mean avoid extensibility at zero cost. If making code extensible costs nothing (e.g., using a map instead of a switch statement), do it. YAGNI targets costly abstractions.
• Tension with DRY: Sometimes YAGNI wins — it's better to have two similar-but-not-identical functions than to prematurely unify them behind the wrong abstraction.

Code Review Checklist

• Does this change introduce any code paths that aren't exercised by current requirements?
• Are there parameters, configs, or options that only have one possible value today?
• Could this abstraction be replaced by a direct implementation without loss of functionality?
• Is this interface/trait/protocol justified by multiple concrete implementations?
• Would a simpler approach work for the current scope?

Related Skills

• kiss: When the issue is complexity rather than speculation
• solid: When designing abstractions that are justified
• dry: When deciding whether to extract a pattern (YAGNI says wait for 3 occurrences)