agentleFS
eduardo-sleduardo-sl/go-agent-skills/skills/(architecture)/go-architecture-review/SKILL.md

go-architecture-review

Skill85 starsChanged 18 days agoMITView on GitHub

Review Go project architecture: package structure, dependency direction, layering, separation of concerns, domain modeling, and module boundaries. Use when reviewing architecture, designing package layout, evaluating dependency graphs, or refactoring monoliths into modules. Trigger examples: "review architecture", "package structure", "project layout", "dependency direction", "clean architecture Go", "module boundaries". Not for: code-level style (go-coding-standards), endpoint design (go-api-design).

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 5 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read a reference file only when the summary below is not enough.

    line 30
  2. Distinctive:

    Run independent passes: (a) layout vs section 1, (b) dependency direction vs section 2, (c) wiring and config vs sections 3+5, (d) package …

    line 67
  3. Distinctive:

    If your environment supports delegating work to parallel sub-agents or tasks, assign each pass to one.

    line 70
  4. Very specific:

    Read config, build the logger, open connections, construct stores → services → handlers in that order, mount routes, run the server.

    line 130
  5. Very specific:

    Avoid dependency injection frameworks.

    line 134
  6. Very specific:

    If wiring gets complex, use Google's wire for compile-time DI code generation.

    line 135
  7. Distinctive:

    All config from environment variables (12-factor).

    line 182
  8. Very specific:

    No config scattered across packages — centralize in internal/config.

    line 184
  9. Distinctive:

    Never hardcode values. Not even "just for now."

    line 185
  10. Very specific:

    Avoid init(). It runs implicitly, makes testing harder, and creates hidden dependencies.

    line 189

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Go Architecture Review

Good architecture makes the next change easy. Bad architecture makes every change scary.

Detailed reference material, loaded on demand:

  • references/layout-and-wiring.md — the full reference directory tree and a composition-root main.go.

Read a reference file only when the summary below is not enough.

Operating Modes

Pick the mode that matches the request before starting:

  • Layout review (default) — assess an existing codebase against the sections below and report violations with severity.
  • Refactor plan — same assessment, but the deliverable is an ordered migration plan (smallest safe steps first), not just findings.
  • New service consultation — asked "how should I structure X": apply sections 1-3 as prescriptive guidance instead of review checks.

Match the Architecture to the Problem

Judge a codebase against the structure its problem needs, not the most elaborate one available. Over-architecture is a finding too.

SituationFitting structure
CLI, small service, one data store, little business logicFlat main + a few packages; no layers for their own sake
Typical service: HTTP/gRPC in front, a database behindHandler → service → store, interfaces at the consumer (sections 1-2)
Several adapters (HTTP, gRPC, queue, CLI) over the same rules, or infrastructure expected to changePorts and adapters: a pure core, adapters on the edge
Large domain, several teams, rules that differ by sub-domainBounded contexts as top-level packages, each layered inside, talking through explicit APIs or events

Signs of too much: interfaces with one implementation per layer, mapping structs copied unchanged through three layers, a "domain" package with no behavior. Signs of too little: SQL in handlers, business rules spread across transport code, packages that import each other in both directions.

Auditing Large Codebases

For repositories with many packages, build the dependency picture before judging it:

  1. Map the module: go list ./... for packages, then import statements to trace dependency direction.
  2. Run independent passes: (a) layout vs section 1, (b) dependency direction vs section 2, (c) wiring and config vs sections 3+5, (d) package design vs section 4.
  3. If your environment supports delegating work to parallel sub-agents or tasks, assign each pass to one; synthesize at the end — dependency findings often explain layout findings.
  4. Cite package paths and file.go:line in every finding.

1. Standard Project Layout

The reference layout: cmd/<binary>/main.go per binary, project-private code under internal/ (domain, service, store/<backend>, handler, config), pkg/ only for genuinely reusable code, plus api/ for OpenAPI/proto and migrations/. Full tree: references/layout-and-wiring.md. For scaffolding a new project, the go-project-layout skill owns the procedure.

Key Rules:

  • internal/ enforces encapsulation at the compiler level. Use it aggressively.
  • pkg/ is for genuinely reusable packages. When in doubt, use internal/.
  • cmd/ main packages should be thin — wire dependencies and call Run().
  • One main.go per binary, minimal logic inside.

2. Dependency Direction

Dependencies MUST flow inward. Domain core has zero external dependencies:

handlers → services → domain ← stores
    ↓          ↓                  ↓
  (net/http)  (pure Go)     (database/sql)

Rules:

  • domain/ imports NOTHING from the project. No store, no handler, no config.
  • service/ depends on domain/ types and interfaces, NOT on concrete stores.
  • handler/ depends on service/ interfaces.
  • store/ implements interfaces defined in service/ or domain/.
  • Circular dependencies are a 🔴 BLOCKER. The compiler catches them, but design should prevent them.
// ✅ Good — service defines the interface it needs
// internal/service/user.go
type UserStore interface {
    GetByID(ctx context.Context, id string) (*domain.User, error)
    Create(ctx context.Context, user *domain.User) error
}

type UserService struct {
    store UserStore // depends on interface, not postgres.Store
}

// internal/store/postgres/user.go
type Store struct { db *sql.DB }

// Implements service.UserStore without importing the service package
func (s *Store) GetByID(ctx context.Context, id string) (*domain.User, error) { ... }

3. Main Package Wiring

main.go is the composition root. Wire everything here:

Read config, build the logger, open connections, construct stores → services → handlers in that order, mount routes, run the server. No business logic. Example: references/layout-and-wiring.md.

Avoid dependency injection frameworks. Go's explicit wiring is a feature. If wiring gets complex, use Google's wire for compile-time DI code generation.

4. Package Design Principles

One package = one purpose

// ✅ Good — clear purpose
package orderservice  // business rules for orders
package postgres      // PostgreSQL data access
package httphandler   // HTTP transport layer

// ❌ Bad — grab-bag packages
package utils    // what ISN'T a util?
package common   // everything and nothing
package models   // types without behavior

Avoid package stuttering

// ❌ Bad — package name repeated in type
package user
type UserService struct{} // user.UserService

// ✅ Good
package user
type Service struct{} // user.Service

Package cohesion over size

A package with 20 related files is better than 20 packages with 1 file each. Split packages when they have distinct responsibilities, not when they get big.

5. Configuration

type Config struct {
    Addr        string        `env:"ADDR" envDefault:":8080"`
    DatabaseURL string        `env:"DATABASE_URL,required"`
    LogLevel    string        `env:"LOG_LEVEL" envDefault:"info"`
    Timeout     time.Duration `env:"TIMEOUT" envDefault:"30s"`
}

Rules:

  • All config from environment variables (12-factor).
  • Validate at startup, fail fast with clear messages.
  • No config scattered across packages — centralize in internal/config.
  • Never hardcode values. Not even "just for now."

6. Init Functions

Avoid init(). It runs implicitly, makes testing harder, and creates hidden dependencies.

// ❌ Bad — hidden side effects
func init() {
    db, _ = sql.Open("postgres", os.Getenv("DB_URL"))
}

// ✅ Good — explicit initialization
func NewStore(dsn string) (*Store, error) {
    db, err := sql.Open("postgres", dsn)
    if err != nil {
        return nil, fmt.Errorf("open db: %w", err)
    }
    return &Store{db: db}, nil
}

Exception: registering drivers or codecs is acceptable in init():

func init() {
    sql.Register("custom", &CustomDriver{})
}

Architecture Review Checklist

  • 🔴 No circular dependencies between packages
  • 🔴 Domain types have zero infrastructure dependencies
  • 🔴 No business logic in cmd/ main packages
  • 🔴 No init() with side effects (DB connections, HTTP calls)
  • 🟡 internal/ used for project-private packages
  • 🟡 Interfaces defined at the consumer, not the producer
  • 🟡 Configuration centralized and validated at startup
  • 🟡 Dependency direction flows inward (handlers → services → domain)
  • 🟢 Package names are short, singular, descriptive
  • 🟢 No utils/, common/, helpers/ packages
  • 🟢 Main package is a thin composition root
  • 🟢 Structure is proportional to the domain — no layers or interfaces without a second reason to exist

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.