Google Go Style Guide

This skill enforces the Google Go Style Guide when writing or reviewing Go code. It covers naming, formatting, error handling, documentation, testing, and language patterns.

The guide is organized by priority: naming and errors matter most because they affect every reader. Formatting is handled by gofmt. The rest improves clarity and maintainability.

When writing Go code, follow these rules directly. When reviewing Go code, flag violations with the specific rule and a concrete fix.

Quick Reference — The Rules That Matter Most

Naming

MixedCaps everywhere. Use MixedCaps or mixedCaps, never snake_case. This applies to constants, variables, functions, methods — everything except filenames and flag names.

No stuttering. Don't repeat the package name in exported symbols:

• widget.New not widget.NewWidget
• db.Load not db.LoadFromDatabase

Short receivers. One or two letters, abbreviating the type. Consistent across all methods:

• func (s *Server) not func (self *Server) or func (server *Server)

No Get prefix. Use Counts() not GetCounts(). Use Compute or Fetch for expensive operations.

Variable length proportional to scope. Single-letter vars are fine in small scopes (loops, short functions). Larger scopes need descriptive names. Don't abbreviate by dropping letters — Sandbox not Sbx.

Don't encode types in names:

• users not userSlice
• count not numUsers or usersInt

Initialisms keep consistent case: URL or url, ID or id, HTTP or http. Never Url, Id, or Http.

Avoid uninformative package names: util, common, helper, base, model are banned. Package names should convey what they do.

Errors

Always return error as the last return value. Use the error interface type, never concrete error types in exported function signatures (avoids nil-interface bugs).

Error strings are lowercase, no punctuation:

// Good:
fmt.Errorf("something bad happened")

// Bad:
fmt.Errorf("Something bad happened.")

Handle every error. Don't discard errors with _ unless the function is documented to never fail (like bytes.Buffer.Write). When ignoring, add a comment explaining why.

Indent error flow — happy path at the left margin:

// Good:
if err != nil {
    return err
}
// normal code continues unindented

// Bad:
if err != nil {
    // error handling
} else {
    // normal code indented unnecessarily
}

No in-band errors. Don't return -1 or "" to signal failure. Use multiple returns:

// Good:
func Lookup(key string) (value string, ok bool)

// Bad:
func Lookup(key string) int // returns -1 on not found

Wrap errors with context using %w:

return fmt.Errorf("failed to load user: %w", err)

Documentation

All exported names get doc comments. Start with the name. Use full sentences:

// A Request represents a request to run a command.
type Request struct { ...

// Encode writes the JSON encoding of req to w.
func Encode(w io.Writer, req *Request) { ...

Comment line length ~80 chars. Not a hard limit, but wrap for readability on narrow screens. Don't break URLs.

Package comments go directly above package with no blank line. One per package. For main packages, describe the binary's purpose.

Formatting & Structure

gofmt is mandatory. Run gofmt -s to also simplify composite literals.

Keep function signatures on one line. Don't break parameter lists across lines — it causes indentation confusion. Factor out local variables to shorten call sites instead.

Don't break if conditions across lines. Extract boolean operands into named variables:

// Good:
inTransaction := db.CurrentStatusIs(db.InTransaction)
keysMatch := db.ValuesEqual(db.TransactionKey(), row.Key())
if inTransaction && keysMatch {
    // ...
}

// Bad:
if db.CurrentStatusIs(db.InTransaction) &&
    db.ValuesEqual(db.TransactionKey(), row.Key()) {
    // ...
}

switch/case on single lines. Don't break case lists across lines unless excessively long.

No redundant break in switch. Go cases don't fall through by default.

Variable on left in comparisons: if result == "foo" not if "foo" == result.

Imports

Group imports in order: (1) standard library, (2) third-party/project packages. Separate groups with blank lines.

Don't rename imports unless there's a collision or the name is truly uninformative (like v1). When renaming proto packages, use a pb suffix.

No dot imports (import . "pkg"). They obscure where things come from.

Blank imports (import _ "pkg") only in main or tests, never in libraries.

Interfaces

Interfaces belong in the consumer package, not the producer. This applies when writing code too — not just during reviews. When you create a service that depends on a database, do NOT define the DB interface in the same package. The interface should live where it's consumed:

// ❌ BAD: interface defined alongside its implementation (producer package)
package users
type DB interface { FindUser(id string) (*User, error) }
type Service struct { db DB }

// ✅ GOOD: consumer defines its own interface; producer returns a concrete type
package users
type Service struct { db *postgres.Client }

// If you need testability, define a minimal interface in the TEST file:
package users_test
type stubDB struct { ... }

When writing a new package with dependencies: accept a concrete type, not an interface you just invented. If you need to abstract the dependency for testing, define the interface in the test file or in the consuming package — never next to the implementation.

Return concrete types from constructors. Let consumers define interfaces for what they need.

Don't define interfaces before you need them. YAGNI applies strongly here.

Testing

No assertion libraries. Use cmp.Equal and standard t.Errorf/t.Fatalf with descriptive messages. Test failures should be diagnosable without reading the test source.

Table-driven tests: Use field names in struct literals. Omit zero-value fields when they're not relevant to the test case.

Test helpers call t.Helper() — but only helper functions, never Test functions themselves. t.Helper() goes in functions like assertUser(t, got, want), not in TestGetUser(t).

Use t.Fatal for setup failures, t.Error for test case validation (allows other cases to run).

Concurrency

Prefer synchronous functions. Let callers add concurrency if needed. Removing concurrency from an async API is much harder than adding it to a sync one.

Make goroutine lifetimes obvious. Use context.Context for cancellation. Use sync.WaitGroup to ensure goroutines don't outlive their parent function.

context.Context is always the first parameter. Don't store it in structs.

Other Rules

Prefer nil slices over empty literals (var s []string not s := []string{}). Don't design APIs that distinguish between nil and empty slices — use len(s) == 0 to check emptiness.

Use %q for string formatting in error messages and user-facing output. It handles empty strings and control characters gracefully.

Use any instead of interface{} in new code.

Don't panic for normal error handling. Reserve panic for truly impossible conditions (bugs). Use Must prefix for init-time helpers that panic on failure.

Don't copy structs with sync.Mutex or bytes.Buffer fields. Use pointer receivers and pointer parameters for types containing these.

Use crypto/rand for generating keys, never math/rand.

Generics: Use only when concrete types or interfaces don't suffice. Don't use generics to build DSLs or assertion frameworks.

Type aliases are rare. Prefer type definitions (type T1 T2) over aliases (type T1 = T2).

Struct Literals

Use field names for types from other packages — always. For local types with many fields — also use field names.

Omit zero-value fields unless the zero value is meaningful to the reader.

Matching braces: closing brace at same indentation as opening. Don't put closing brace on same line as last value in multi-line literals.

Omit repeated type names in slice/map literals:

// Good:
[]*Type{{A: 42}, {A: 43}}

// Bad:
[]*Type{&Type{A: 42}, &Type{A: 43}}

Common Mistakes

These are the violations agents commit most often, even when they know the rules:

Mistake	Why it happens	Fix
Define interface in same package as implementation	Feels "clean" to co-locate	Accept concrete types; define interfaces in consumer or test file
GetUser instead of User or ByID	Habit from other languages	Drop the Get prefix — Go getters don't use it
UserFilter in package users	Forget the package name is part of the API	Call it Filter — callers see users.Filter
Store context.Context in a struct	Convenient for "reuse"	Pass ctx as the first param to every method that needs it
t.Helper() in TestXxx functions	Confuse test functions with helper functions	Only call t.Helper() in non-Test helper functions
Id, Url, Http	Muscle memory	All-caps for initialisms: ID, URL, HTTP
map[string]*User{} instead of nil	Defensive instinct	Use var m map[string]*User unless you need a non-nil empty map

Review Checklist

When reviewing Go code, check for these violations in priority order:

1. Naming: stuttering, snake_case, Get prefix, uninformative names, wrong initialism casing
2. Error handling: discarded errors, in-band errors, capitalized error strings, missing wrapping
3. Documentation: missing doc comments on exported symbols, comments not starting with the name
4. Interfaces: defined in producer not consumer, premature interface definitions
5. Testing: assertion libraries, missing t.Helper(), unclear failure messages
6. Formatting: multi-line if conditions, broken function signatures, redundant break
7. Concurrency: context not first param, unclear goroutine lifetimes, context stored in struct

For detailed rules with more examples, read the relevant reference file:

• references/naming.md — MixedCaps, package names, receivers, getters, repetition, variable names
• references/commentary.md — Doc comments, comment sentences, examples, package comments
• references/imports.md — Import grouping, renaming, dot imports, blank imports
• references/errors.md — Returning errors, error strings, wrapping, in-band errors, indent error flow
• references/language-patterns.md — Literals, nil slices, formatting, conditionals, switch, copying, type aliases
• references/interfaces-and-types.md — Interfaces, generics, receivers, Must functions, panic
• references/concurrency.md — Goroutine lifetimes, contexts, synchronous functions, crypto/rand
• references/testing.md — Test failures, table-driven tests, cmp.Diff, test helpers, no assertion libraries