| name | modular-architecture |
| description | Use when making dependency management decisions, drawing component boundaries, applying SOLID principles, designing layered architectures, or evaluating whether a structural decision keeps options open vs. locks them in. |
| version | 1.0.0 |
| author | Ayush Singh |
| license | MIT |
| platforms | ["macos","linux"] |
| metadata | {"hermes":{"tags":["architecture","solid","clean-architecture","boundaries","dependencies","components","swe-foundations"],"category":"swe-foundations","related_skills":["engineering-mindset","production-readiness","distributed-systems","agentic-swe-master"]}} |
Modular Architecture & Boundaries
When to Load
- Use this when a coding agent is about to choose a framework, database, or web server before domain logic exists
- Use this when agent-generated code imports ORM models, HTTP types, or framework objects inside business logic
- Use this when designing module, package, or component boundaries — deciding what belongs where
- Use this when a feature is getting slower to implement over time (productivity asymptote signal)
- Use this when an agent is asked to "just make it work quickly" and is about to skip structural discipline
- Use this when business rules are tangled with I/O, database calls, or HTTP handler code
- Use this when planning a microservices split — evaluating whether services truly decouple teams
- Use this when deciding how to pass data across a boundary (which side owns the data shape)
- Use this when evaluating a full system rewrite vs. incremental improvement
- Use this when an agent is about to couple a use case to a specific presenter, view, or delivery mechanism
- Use this when writing tests that are breaking because of structural coupling to production code
- Use this when team org chart is driving component boundaries
Core Rules
The goal of software architecture is to minimize the human resources required to build and maintain the required system.
The only way to go fast is to go well. Making messes is always slower than staying clean.
A good architect pretends decisions have not been made, and shapes the system such that those decisions can still be made later.
Source code dependencies must point only inward, toward higher-level policies. Nothing in an inner circle can know anything about an outer circle.
Good architects carefully separate details from policy, and decouple the policy from the details so thoroughly that the policy has no knowledge of the details.
The longer you wait to make a decision, the more information you have with which to make it properly.
A module should be responsible to one and only one actor — the single source of change that can demand it be modified.
Components should not be forced to depend on things they don't use (ISP). Depend on stable abstractions, not volatile concrete implementations (DIP).
Architecture should reveal intent — use cases, features, and actors should be visible in the top-level structure, not the framework choice.
The web is a detail. The database is a detail. The framework is a detail. Keep them on the outside.
The Dependency Rule (Most Critical)
All source code dependencies must point inward toward higher-level policy:
[Entities] <- [Use Cases] <- [Interface Adapters] <- [Frameworks & Drivers]
Violations to catch immediately:
- Use case importing a database model class
- Entity method calling an HTTP client
- Business rule knowing the name of a presenter or view
- SQL query in a use case layer
- Framework annotation on a domain object
SOLID at Architecture Scale
SRP: Group by actor (source of change), not by technical function.
OCP: Extend behavior by adding new components behind existing interfaces, not by modifying existing ones.
LSP: Subtypes must be fully substitutable. When they aren't, isolate dispatch behind a factory.
ISP: Don't force modules to depend on interfaces they don't use.
DIP: Depend on abstractions at every boundary. Factories isolate the crossing point.
Component Cohesion Principles
Cohesion (what goes together):
REP: Release as a unit. Nothing goes in a component that isn't released together.
CCP: Same actor, same component. Things that change together belong together.
CRP: Don't force users to redeploy for changes they don't care about.
Coupling (what can depend on what):
ADP: No cycles in the component dependency graph.
SDP: Depend in the direction of stability.
SAP: Stable components should be abstract.
Boundary Types (weakest to strongest)
- Source-level monolith — same process, compile-time separation
- Deployment component — dynamically linked, separately versioned
- Local process — same machine, separate address spaces
- Service — network call, highest latency, strongest isolation
Rule: Use the weakest boundary that satisfies actual deployment needs.
Do not use services for development convenience — they carry real operational cost.
Concept Map
Wiki root: $AGENTIC_SWE_WIKI_ROOT/clean-architecture/concepts/
| Concept | When to read |
|---|
| Dependency-Rule.md | Any boundary crossing decision |
| SOLID-Principles.md | Class or module design, interface design |
| Component-Cohesion.md | Deciding what belongs in the same release unit |
| Component-Coupling.md | Dependency graph has cycles, or stability direction is wrong |
| Humble-Object-Pattern.md | Writing tests that must cross an architectural boundary |
| Package-Organization.md | Deciding folder/package structure for a new service |
| Boundaries-and-Plugins.md | Choosing where a framework or DB attaches to the system |
| Screaming-Architecture.md | Top-level folder structure should reveal use cases, not framework |
Common Pitfalls
- Premature Framework Commitment: Framework chosen before domain model exists. Fix — treat frameworks as plugins; sketch domain use cases first, attach framework last.
- Architecture-Design Split: Treating high-level arch as separate from low-level design. Fix — they are one continuous discipline; design decisions are architectural decisions.
- Big-Rewrite Overconfidence: Full redesign replaces one mess with another. Fix — incremental improvement with enforced boundaries is safer.
- Service Architecture Fallacy: Microservices are inherently decoupled. Fix — services are deployment boundaries; coupling exists if cross-service data contracts are tight.
- Structural Test Coupling: Tests mirror production class structure one-to-one. Fix — test behavior, not structure; use a Testing API that bypasses structural detail.
- Passing Row Structures Inward: Returning a database row and passing it into use cases. Fix — define a data structure owned by the use case; adapter converts from row to that structure.
Verification Checklist
