Go code organization
Skill v0lka/skills/development/idiomatic-go/go-code-organization
A set of AI agent skills for research and development tasks.
npx -y skills add v0lka/skills --skill go-code-organizationAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 13 stars13 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Guides the agent to write well-organized Go code and projects. Use when writing or reviewing Go code involving variable scoping, nested control flow, init functions, getters/setters, interfaces, generics, type embedding, functional options, package structure, utility packages, package naming, code documentation, or linter configuration. Covers the 16 most common code and project organization mistakes in Go.
SKILL.md
14.8 KB, as published. Nobody here has run it
Go Code and Project Organization — Rules and Patterns
1. Variable shadowing
A variable declared with := in an inner block shadows the outer variable of the same name. The outer variable remains unchanged, which is almost always a bug.
Wrong:
var client *http.Client
if tracing {
client, err := createClientWithTracing() // shadows outer client
if err != nil {
return err
}
log.Println(client)
} else {
client, err := createDefaultClient() // shadows outer client
if err != nil {
return err
}
log.Println(client)
}
// client is still nil here
Fix — pre-declare err, use = instead of :=:
var client *http.Client
var err error
if tracing {
client, err = createClientWithTracing()
} else {
client, err = createDefaultClient()
}
if err != nil {
return err
}
Alternative: assign := to a temporary variable (c), then client = c after the error check.
Detection: Run go vet -vettool=$(which shadow) (install with golang.org/x/tools/go/analysis/passes/shadow/cmd/shadow).
2. Unnecessary nesting
Keep the happy path left-aligned. Reduce nesting by returning early on errors and flipping conditions.
Rules:
- When an
ifblock returns, omit theelse. - If the non-happy path is in the
else, flip the condition. - Aim for a maximum of two indent levels inside a function body.
Wrong:
if s != "" {
// ... long happy path
} else {
return errors.New("empty string")
}
Correct:
if s == "" {
return errors.New("empty string")
}
// ... happy path at top level
Apply this recursively: replace every if ... { return } else { ... } chain with guard clauses that return early, keeping the happy path at the lowest indent level.
3. init functions
Avoid init when:
- The initialization can fail —
initcannot return errors, so the only signal islog.Fatal/panic, removing the caller's ability to retry or fall back. - It sets global mutable state — makes testing harder and any function in the package can alter the global.
- The side effect is not needed by every test in the file.
Acceptable uses of init:
- Registering static, infallible configuration (e.g.,
http.HandleFuncwith non-nil handlers). - Side-effect imports (
_ "image/png") for codec registration.
Wrong — DB connection in init:
var db *sql.DB
func init() {
d, err := sql.Open("mysql", os.Getenv("DSN"))
if err != nil {
log.Panic(err) // caller can't handle this
}
db = d
}
Correct — explicit constructor:
func NewDB(dsn string) (*sql.DB, error) {
db, err := sql.Open("mysql", dsn)
if err != nil {
return nil, err
}
if err = db.Ping(); err != nil {
return nil, err
}
return db, nil
}
4. Getters and setters
Go does not require getters and setters. Do not add them unless they provide value (validation, computed values, mutex wrapping, debugging interception, or forward compatibility).
Naming convention when used:
- Getter:
Balance()(notGetBalance()) - Setter:
SetBalance(v int)
currentBalance := customer.Balance()
if currentBalance < 0 {
customer.SetBalance(0)
}
The standard library exposes struct fields directly when appropriate (e.g., time.Timer.C). Follow the same pragmatism.
5. Interface pollution
Core principle: Abstractions should be discovered, not created. Do not define an interface until you have a concrete need.
Valid reasons to create an interface:
- Common behavior — multiple types share the same method set (e.g.,
sort.Interface). - Decoupling — swap implementations for testing or Liskov substitution.
- Restricting behavior — expose only a subset of a type's methods (e.g., read-only config getter from a read-write config struct).
Do not create interfaces preemptively "in case we need them later." If it is unclear how an interface improves the code, remove it.
Keep interfaces small. "The bigger the interface, the weaker the abstraction." (Rob Pike)
6. Interface on the consumer side
Interfaces should live in the package that uses them, not in the package that implements them.
- The producer exports the concrete struct.
- Each consumer defines only the interface it needs (possibly a single-method interface), keeping it unexported.
Wrong — producer-side interface:
// package store
type CustomerStorage interface { // forces all consumers into this abstraction
StoreCustomer(Customer) error
GetCustomer(id string) (Customer, error)
// ... 4 more methods
}
Correct — consumer-side interface:
// package client
type customersGetter interface { // unexported, minimal
GetAllCustomers() ([]store.Customer, error)
}
Exception: An interface on the producer side is acceptable when you know (not foresee) it will be used by many consumers (e.g., encoding.BinaryMarshaler). Keep it as small as possible.
7. Returning interfaces
Functions should return concrete types, not interfaces. Returning an interface:
- Creates a dependency from the implementation package to the client package.
- Forces every consumer into the same abstraction level.
Guideline (Postel's law applied to Go):
- Accept interfaces.
- Return structs.
Exceptions: The error interface (ubiquitous), and up-front abstractions proven to be universally useful (e.g., io.LimitReader returns io.Reader).
8. any says nothing
any (interface{}) discards all type information. Avoid it unless you genuinely need to accept every possible type.
Wrong:
func (s *Store) Get(id string) (any, error) { ... }
func (s *Store) Set(id string, v any) error { ... }
Correct — explicit per-type methods:
func (s *Store) GetContract(id string) (Contract, error) { ... }
func (s *Store) SetContract(id string, c Contract) error { ... }
func (s *Store) GetCustomer(id string) (Customer, error) { ... }
func (s *Store) SetCustomer(id string, c Customer) error { ... }
Legitimate uses of any: json.Marshal(v any), fmt.Println(a ...any), db.QueryContext(ctx, query, args ...any) — where any possible type is truly expected.
9. Generics
When to use
- Data structures — binary trees, linked lists, heaps parameterized by element type.
- Functions on slices/maps/channels of any type — e.g.,
merge[T any](ch1, ch2 <-chan T) <-chan T. - Factoring out behaviors — e.g., a generic
SliceFn[T]that implementssort.Interface.
When NOT to use
- Calling a method of the type argument — if the body calls
w.Write(b), just acceptio.Writerdirectly. - When it makes code harder to read — generics are never mandatory. If the generic version is not clearly simpler, keep the concrete version.
Rule of thumb: Do not use type parameters preemptively. Wait until you are about to write boilerplate code to consider generics.
Constraints
- Use
comparablefor map keys or equality checks. - Use
~int | ~string(union with~) to allow custom types whose underlying type matches. - Type parameters work on functions and type receivers, not on individual methods:
// Compile error:
func (Foo) bar[T any](t T) {} // methods cannot have type parameters
// OK — use a type-parameterized receiver:
type Foo[T any] struct{ val T }
func (f Foo[T]) Bar() T { return f.val }
10. Type embedding
Embedding promotes all fields and methods of the inner type. Use it only when promotion is desirable.
Do NOT embed when:
- It only saves typing (
Foo.Baz()vs.Foo.Bar.Baz()) with no semantic benefit. - It promotes fields or methods that should be private (e.g.,
sync.Mutex— clients should not callLock/Unlock).
Wrong — mutex embedded:
type InMem struct {
sync.Mutex // Lock/Unlock are now public
m map[string]int
}
// m := inmem.New(); m.Lock() — exposed to external callers
Correct — mutex as a named field:
type InMem struct {
mu sync.Mutex // unexported, invisible to clients
m map[string]int
}
Good use of embedding — forwarding methods intentionally:
type Logger struct {
io.WriteCloser // promotes Write and Close deliberately
}
Remember: Embedding is composition, not inheritance. The embedded type remains the method receiver.
11. Functional options pattern
Use this pattern when a constructor has optional configuration. It is the idiomatic Go approach and avoids the downsides of config structs (zero-value ambiguity) and builder patterns (empty struct boilerplate).
type options struct {
port *int
timeout time.Duration
}
type Option func(*options) error
func WithPort(port int) Option {
return func(o *options) error {
if port < 0 {
return errors.New("port should be positive")
}
o.port = &port
return nil
}
}
func WithTimeout(t time.Duration) Option {
return func(o *options) error {
o.timeout = t
return nil
}
}
func NewServer(addr string, opts ...Option) (*http.Server, error) {
var o options
for _, opt := range opts {
if err := opt(&o); err != nil {
return nil, err
}
}
// use o.port, o.timeout with defaults ...
}
Caller usage:
srv, err := httplib.NewServer("localhost",
httplib.WithPort(8080),
httplib.WithTimeout(time.Second),
)
// Default config — no extra args needed:
srv, err := httplib.NewServer("localhost")
Conventions:
- Option functions start with
Withprefix. - The options struct is unexported.
- Validation happens inside each
With*function, not deferred toBuild.
12. Project structure
There is no official Go project structure standard. When choosing a layout:
/cmd— main entry points (/cmd/foo/main.go)./internal— private packages that cannot be imported externally./pkg— public library code (optional; some teams skip this)./test— integration and public API tests.- No
/srcdirectory.
Package organization rules:
- Avoid premature packaging. Start simple, split when boundaries become clear.
- Avoid nano packages (1-2 files with no cohesion) and monolith packages.
- Name packages after what they provide, not what they contain.
- Package names: short, single lowercase word, concise.
- Minimize exports. When unsure, keep it unexported; export later if needed.
- Organize by context (domain) or by layer (hexagonal), but be consistent.
13. No utility packages
Do not create packages named utils, common, shared, or base. These names carry no meaning about what the package provides.
Wrong:
package util
func NewStringSet(...string) map[string]struct{} { ... }
func SortStringSet(map[string]struct{}) []string { ... }
set := util.NewStringSet("c", "a", "b")
fmt.Println(util.SortStringSet(set))
Correct — name after what it provides:
package stringset
type Set map[string]struct{}
func New(...string) Set { ... }
func (s Set) Sort() []string { ... }
set := stringset.New("c", "a", "b")
fmt.Println(set.Sort())
If common types are shared between a client and server package, consider merging them into one package rather than creating a common package.
14. Package name collisions
Do not use a variable name that shadows an imported package name. It makes the package inaccessible within the variable's scope and confuses readers.
Wrong:
redis := redis.NewClient() // redis package is now shadowed
v, err := redis.Get("foo") // is this the package or the variable?
Fix A — different variable name:
redisClient := redis.NewClient()
Fix B — import alias:
import redisapi "mylib/redis"
redis := redisapi.NewClient()
Also avoid shadowing built-in function names (copy, len, cap, new, make, close, delete, append, etc.).
15. Code documentation
- Every exported element (type, function, method, constant, variable) must have a doc comment.
- Comments start with the element name:
// Customer is a customer representation. - Each comment is a complete sentence ending with punctuation.
- Document what a function does and why, not how.
- Package comments:
// Package math provides basic constants and mathematical functions. - Place package comments in the relevant file or a dedicated
doc.go. - Deprecate with
// Deprecated: Use X instead. - A blank line between a copyright header and the package comment keeps the copyright out of godoc.
// DefaultPermission is the default permission used by the store engine.
const DefaultPermission = 0o644 // Need read and write accesses.
16. Linters
Always use linters. At minimum:
go vet— standard Go analyzer.shadow— detects variable shadowing.errcheck— catches unchecked errors.gocyclo— flags high cyclomatic complexity.goconst— finds repeated string constants.
Formatters: gofmt, goimports.
Use golangci-lint as a single entry point — it wraps many linters, runs them in parallel, and is configurable via .golangci.yml. Automate linting in CI or Git pre-commit hooks.
Quick-reference checklist
- No unintended variable shadowing (
:=in inner blocks re-checked). - Happy path left-aligned; no unnecessary
elseafterreturn. -
initonly used for infallible static setup; fallible init done in explicit constructors. - Getters named
Field(), notGetField(); only present when they add value. - Interfaces discovered from need, not created speculatively.
- Interfaces defined on the consumer side, not the producer side.
- Functions return concrete types, not interfaces (except
errorand proven up-front abstractions). -
anyonly used when truly any type is valid; signatures are explicit otherwise. - Generics used for data structures and collection utilities, not when a plain interface suffices.
- Embedded types only when promotion of all fields/methods is intended;
sync.Mutexnever embedded. - Functional options pattern used for optional constructor config.
- No
utils/common/basepackages; packages named after what they provide. - No variable names colliding with package names or built-in functions.
- All exported elements have doc comments starting with the element name.
-
golangci-lint(or equivalent) integrated in CI.