SuperAPI AI Directory

architecture-review

General & other

View repository

Updated 10-16 00:05

About this skill

Architecture Review

Conduct rigorous architectural reviews that blend enterprise governance with modern, agent-friendly development patterns. Produce decisions that are documented, defensible, and immediately actionable.

Core Philosophy

Make decisions explicit. Every significant choice becomes an ADR. Guardrails over gates. Provide golden paths and reference architectures, not bureaucratic approval queues. Agent-native from day one. Structure projects so AI agents can understand, modify, and extend them reliably. Start right, stay lean. Enterprise-grade foundations without enterprise-grade overhead.


Quick Start: New Project Review

When starting a new project, run through these phases in order:

Phase 1: Strategic Alignment (5 min)

Answer these before touching code:

  1. What problem does this solve? One sentence.
  2. Who are the users? Primary persona + scale expectations.
  3. What are the constraints? Budget, timeline, team size, compliance requirements.
  4. What exists already? Existing services, data stores, or platforms to integrate with.
  5. What is the deployment target? Cloud provider, on-prem, edge, hybrid.

Phase 2: Architecture Decision Cascade

Work through these decisions in order. Each produces an ADR.

# Decision Reference
1 Architecture style architecture-patterns.md
2 Module/service boundaries architecture-patterns.md
3 API strategy api-and-data.md
4 Data architecture api-and-data.md
5 Security posture security-and-compliance.md
6 Observability stack devops-and-quality.md
7 Testing strategy devops-and-quality.md
8 CI/CD and deployment devops-and-quality.md
9 Agent workflow setup agentic-patterns.md

Phase 3: Scaffold and Ship

After decisions are made, generate the project structure. See project-scaffolding.md for the canonical directory layout and day-one checklist.


Architecture Decision Record Template

Every significant decision gets an ADR. Store in docs/adr/ with sequential numbering.

# ADR-NNNN: [Short Title]

## Status
[Proposed | Accepted | Deprecated | Superseded by ADR-XXXX]

## Context
What forces are at play? What is the problem?

## Decision Drivers
- Key factor 1
- Key factor 2

## Considered Options
1. **Option A** -- pros / cons
2. **Option B** -- pros / cons

## Decision
What we chose and why.

## Consequences
What becomes easier. What becomes harder. What risks remain.

Rules:

  • One decision per ADR. Never combine.
  • Immutable once accepted -- supersede, don't edit.
  • Keep under 2 pages. Alternatives considered is the most valuable section.
  • Link related ADRs to create a decision graph.

Quality Attributes Checklist

Evaluate every architecture against these attributes. Score each as: Addressed / Planned / Not Applicable / Risk.

Attribute Key Questions
Scalability Can it handle 10x load? Horizontal or vertical strategy?
Security Auth/authz defined? Data encrypted at rest and in transit? Threat model done?
Reliability Target uptime? Redundancy and failover?
Performance Latency budgets? Throughput targets?
Maintainability Modular? Clear boundaries? Can a new developer (or agent) be productive in a day?
Observability Structured logging? Distributed tracing? Alerting?
Testability Can components be tested in isolation? What is the test strategy shape?
Deployability CI/CD automated? Blue-green/canary capability? Rollback plan?
Cost efficiency Resource utilization optimized? Pay-per-use aligned?
Agent compatibility Can AI agents understand, modify, and extend this codebase?

Anti-Patterns to Flag

Reject architectures exhibiting these patterns:

Anti-Pattern Signal
Big Ball of Mud No clear module boundaries, circular dependencies
Over-Engineering Google-scale infra for startup workloads
Blind Pattern Following "We chose microservices because Netflix" without justification
Infrastructure Ignorance Designs assuming unlimited bandwidth or zero latency
Under-Modularization Monolithic components that can't deploy independently
Over-Modularization Hundreds of nano-services creating operational overhead
Document Theater Extensive paperwork that nobody reads or acts on
Domain Allergy Architecture that ignores business domain and language

Review Outcomes

Every review produces one of these decisions:

Decision Meaning
Approved Proceed as designed
Approved with conditions Proceed with specific modifications
Needs more info Specific questions must be answered first
Redesign required Does not meet quality standards

Reference Materials