About go-interface-design
A Go interface-design skill for developers and coding agents, covering where to define interfaces, how to keep them focused, and common design pitfalls.
Key features
- 01Define interfaces at the consumer
- 02Keep interfaces small and composable
- 03Accept interfaces, return concrete types
- 04Verify implementation at compile time
- 05Avoid pointers to interfaces
- 06Recognize embedding and assertion pitfalls
Use cases
- 01Design package boundaries around consumer needs
- 02Add test seams without broad interfaces
- 03Review Go code for interface anti-patterns
Install
Copy the command and run it in a terminal.
npx skills add https:/ / github.com/ eduardo-sl/ go-agent-skills --skill go-interface-designAfter installing, type/go-interface-designin Claude Code to confirm it loaded.
Bundled filesreferences/embedding-and-assertions.md
Details
From the project READMEGo Interface Design
Go interfaces are implicit. This is the single most important design feature of the language, and most people coming from Java or C# get it wrong at first.
Community default. A project skill or agent-config rule that explicitly supersedes
go-interface-designtakes precedence over this one.
Detailed reference material, loaded on demand:
references/embedding-and-assertions.md— embedding vs named fields, interface-embedding test fakes, optional capabilities, sealed interfaces.
Read a reference file only when the summary below is not enough.
1. The Cardinal Rule: Define Interfaces at the Consumer
The consumer of a behavior defines the interface, NOT the provider:
// ❌ Wrong — producer defines interface (Java thinking)
// package store
type UserStore interface { // defined alongside implementation
GetByID(ctx context.Context, id string) (*User, error)
Create(ctx context.Context, user *User) error
// ... 15 more methods
}
type PostgresStore struct { ... }
func (s *PostgresStore) GetByID(...) { ... }
func (s *PostgresStore) Create(...) { ... }
// ✅ Right — consumer defines what it needs
// package service
type UserReader interface { // only what THIS service needs
GetByID(ctx context.Context, id string) (*domain.User, error)
}
type UserService struct {
store UserReader // depends on narrow interface
}
// package store (no interface defined here)
type PostgresStore struct { db *sql.DB }
func (s *PostgresStore) GetByID(ctx context.Context, id string) (*domain.User, error) { ... }
func (s *PostgresStore) Create(ctx context.Context, user *domain.User) error { ... }
// PostgresStore satisfies service.UserReader implicitly — no declaration needed
Why this matters:
- Consumer depends only on what it uses (Interface Segregation Principle).
- Producer can add methods without breaking consumers.
- Testing requires only the methods the consumer calls.
- No import cycle: consumer doesn't import producer's package.
2. Keep Interfaces Small
The bigger the interface, the weaker the abstraction.
// ✅ Good — focused, composable
type Reader interface {
Read(p []byte) (n int, err error)
}
type Writer interface {
Write(p []byte) (n int, err error)
}
type ReadWriter interface {
Reader
Writer
}
// ❌ Bad — kitchen sink interface
type FileManager interface {
Read(path string) ([]byte, error)
Write(path string, data []byte) error
Delete(path string) error
List(dir string) ([]string, error)
Move(src, dst string) error
Copy(src, dst string) error
Stat(path string) (os.FileInfo, error)
Watch(path string) (<-chan Event, error)
}
Guideline: 1-3 methods is ideal. If you need more, compose smaller interfaces.
3. Accept Interfaces, Return Structs
// ✅ Good — accepts interface, returns concrete type
func NewUserService(store UserReader, logger Logger) *UserService {
return &UserService{store: store, logger: logger}
}
// ❌ Bad — returns interface (hides the concrete type for no reason)
func NewUserService(store UserReader) UserServiceInterface {
return &UserService{store: store}
}
Return a concrete type so callers get full access to the type's methods. Returning an interface only makes sense when the function genuinely returns different concrete types based on input (factory pattern).
4. Verify Interface Compliance at Compile Time
Use the blank identifier assignment to catch broken contracts early:
// Verify *PostgresStore implements service.UserReader at compile time
var _ service.UserReader = (*PostgresStore)(nil)
// Verify LogHandler implements http.Handler
var _ http.Handler = (*LogHandler)(nil)
// For value receivers:
var _ fmt.Stringer = Status(0)
Place these immediately after the type declaration. They cost nothing at runtime and prevent silent contract breakage.
5. Don't Use Pointers to Interfaces
// ❌ Bad — pointer to interface is almost never correct
func process(r *io.Reader) { ... }
// ✅ Good — interface is already a pointer internally
func process(r io.Reader) { ... }
An interface value is internally two pointers (type + data). A pointer to an interface is a pointer to a pointer — needless indirection.
The only exception: when you need to replace the interface value itself (swap the implementation at runtime), which is extremely rare.
6. The Empty Interface
interface{} (or any in Go 1.18+) means you've given up on type safety.
Use it sparingly:
// ✅ Acceptable — generic container before generics / stdlib compatibility
func Marshal(v any) ([]byte, error)
// ✅ Better (Go 1.18+) — use generics instead of any
func Map[T, U any](slice []T, fn func(T) U) []U { ... }
// ❌ Bad — lazy interface design
func Process(data any) any { ... } // what does this even do?
7. Optional Configuration Is Not an Interface Problem
Optional constructor settings are not a reason to introduce an interface. Use functional options or a config struct — the go-design-patterns skill owns both and the trade-off between them.
8. Common Interface Anti-Patterns
Premature interfaces:
// ❌ Bad — interface defined before second implementation exists
type Processor interface {
Process(ctx context.Context, data []byte) error
}
type processor struct { ... } // only one implementation ever
// ✅ Good — use concrete type until you need the abstraction
type Processor struct { ... }
// Add interface when you have 2+ implementations or need testing seam
"Don't design with interfaces, discover them." — Rob Pike
Interface pollution:
// ❌ Bad — wrapping every struct in an interface "for testability"
type UserServiceInterface interface { ... }
type OrderServiceInterface interface { ... }
type PaymentServiceInterface interface { ... }
// 50 more interfaces with exactly one implementation each
// ✅ Good — define interfaces where they're consumed
// Each consumer declares only the methods IT needs
Sealed interfaces vs enums:
A sealed interface — an unexported marker method, so only this package can
add variants — is the right tool for a closed set of variants that carry
different data, consumed with a type switch. When the variants carry no
data, use a typed constant enum instead. Example:
references/embedding-and-assertions.md.
9. Embedding and Type Assertions
- Embedding promotes the whole method set. Embedding
sync.Mutexin an exported struct publishesLock/Unlock; use a named field unless the outer type truly is the inner one, extended. - An interface embedded in a struct satisfies it with methods that panic when called — fine for test fakes, a hidden runtime failure elsewhere.
- Assert to an interface for optional capabilities
(
w.(http.Flusher)), always with comma-ok. - Generic methods (Go 1.27+) never satisfy an interface method; keep the methods of anything you define as an interface non-generic.
Decision Checklist
- Do I need an interface here? — Only if you have 2+ implementations, need a testing seam, or are crossing a package boundary.
- Where should it be defined? — At the consumer, not the producer.
- How many methods? — Fewer is better. 1-3 is ideal.
- Am I returning an interface? — Probably shouldn't. Return concrete.
- Have I verified compliance? —
var _ Interface = (*Type)(nil) - Am I embedding for convenience? — named field unless every promoted method belongs on the outer type.