oop-principles-solid

name	oop-principles-solid
description	SOLID principles for reducing coupling and improving testability — SRP, OCP, LSP, ISP, and DIP with annotated examples. TRIGGER when: user mentions SOLID, SRP, OCP, LSP, ISP, DIP, single responsibility principle, open-closed principle, Liskov substitution, interface segregation, dependency inversion, god class, fat interface, one reason to change, extend without modifying, depend on abstractions, interface too large, reduce coupling, improve testability.
user-invocable	false

SOLID Principles

Quick Reference

Principle	Rule	Signal it's violated	Fix
SRP — Single Responsibility	One reason to change	Class handles auth, hashing, and persistence	Split UserController → UserValidator, PasswordHasher, UserRepository
OCP — Open–Closed	Extend via subclass; don't modify proven code	Adding CSV export requires editing JsonFormatter	Add CsvResponseFormatter extends JsonResponseFormatter without touching the base
LSP — Liskov Substitution	Subtypes substitutable without surprising callers	instanceof checks in calling code; subclass throws "not supported"	Split ReadableCache / WritableCache interfaces instead of ReadOnlyCache extends Cache
ISP — Interface Segregation	Many focused interfaces over one fat one	Read-only client forced to implement write methods	Slice fat interface; don't force read-only clients onto write methods
DIP — Dependency Inversion	Depend on abstractions, not concretions	EmailService directly instantiates SendgridClient	Inject EmailService(transport: MailTransport), not new SendgridClient()

→ Full annotated examples for all five principles: references/solid.md

Benchmark

Scenario: .benchmarks/scenarios/solid-001-god-class.md · Run: 2026-06-26

Model	Without	With	Delta
claude-opus-4-8	100%	100%	+0%
claude-sonnet-4-6	100%	100%	+0%
claude-haiku-4-5	100%	100%	+0%

NEUTRAL (no regression). On the OrderService god-class all baselines already name the SRP/DIP violations, split by genuine responsibility, and — notably — do not over-split into anemic classes. The skill neither helps nor harms at this difficulty; the feared dogmatic-over-split regression did not materialise. Follow-up: author a harder borderline scenario (a class that only looks multi-purpose) to find where the skill is load-bearing. Gate per skill-optimizer/release-gates.md.