review-software-architecture
À propos
Cette compétence examine l'architecture logicielle sous l'angle du couplage, de la cohésion, des principes SOLID, de la conception d'API, de l'évolutivité et de la dette technique. Elle fournit une évaluation au niveau système, examine les Archives de Décisions d'Architecture (ADR) et propose des recommandations d'amélioration. Utilisez-la pour évaluer la qualité, l'évolutivité, la sécurité ou l'état de préparation à la montée en charge de systèmes proposés ou existants.
Installation rapide
Claude Code
Recommandénpx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/review-software-architectureCopiez et collez cette commande dans Claude Code pour installer cette compétence
Documentation
Review Software Architecture
Evaluate software architecture at system level for quality attributes, design principles adherence, long-term maintainability.
When Use
- Evaluating proposed architecture before implementation begins
- Assessing existing system for scalability, maintainability, or security
- Reviewing Architecture Decision Records (ADRs) for project
- Performing technical debt assessment
- Evaluating whether system ready for significant scale-up or feature expansion
- Differentiating from line-level code review (which focuses on PR-level changes)
Inputs
- Required: System codebase or architecture documentation (diagrams, ADRs, README)
- Required: Context about system purpose, scale, constraints
- Optional: Non-functional requirements (latency, throughput, availability targets)
- Optional: Team size and skill composition
- Optional: Technology constraints or preferences
- Optional: Known pain points or areas of concern
Steps
Step 1: Understand System Context
Map system boundaries and interfaces:
## System Context
- **Name**: [System name]
- **Purpose**: [One-line description]
- **Users**: [Who uses it and how]
- **Scale**: [Requests/sec, data volume, user count]
- **Age**: [Years in production, major versions]
- **Team**: [Size, composition]
## External Dependencies
| Dependency | Type | Criticality | Notes |
|-----------|------|-------------|-------|
| PostgreSQL | Database | Critical | Primary data store |
| Redis | Cache | High | Session store + caching |
| Stripe | External API | Critical | Payment processing |
| S3 | Object storage | High | File uploads |
Got: Clear picture of what system does and what it depends on. If fail: Architecture documentation missing? Derive context from code structure, configs, deployment files.
Step 2: Evaluate Structural Quality
Coupling Assessment
Examine how tightly modules depend on each other:
- Dependency direction: Do dependencies flow in one direction (layered) or circular?
- Interface boundaries: Modules connected through defined interfaces/contracts or direct implementation references?
- Shared state: Mutable state shared between modules?
- Database coupling: Multiple services read/write same tables directly?
- Temporal coupling: Must operations happen in specific order without explicit orchestration?
# Detect circular dependencies (JavaScript/TypeScript)
npx madge --circular src/
# Detect import patterns (Python)
# Look for deep cross-package imports
grep -r "from app\." --include="*.py" | sort | uniq -c | sort -rn | head -20
Cohesion Assessment
Evaluate whether each module has single, clear responsibility:
- Module naming: Does name accurately describe what module does?
- File size: Files or classes excessively large (>500 lines suggests multiple responsibilities)?
- Change frequency: Do unrelated features need changes to same module?
- God objects: Classes/modules that everything depends on?
| Coupling Level | Description | Example |
|---|---|---|
| Low (good) | Modules communicate through interfaces | Service A calls Service B's API |
| Medium | Modules share data structures | Shared DTO/model library |
| High (concern) | Modules reference each other's internals | Direct database access across modules |
| Pathological | Modules modify each other's internal state | Global mutable state |
Got: Coupling and cohesion assessed with specific examples from codebase. If fail: Codebase too large for manual review? Sample 3-5 key modules and most-changed files.
Step 3: Assess SOLID Principles
| Principle | Question | Red Flags |
|---|---|---|
| Single Responsibility | Does each class/module have one reason to change? | Classes with >5 public methods on unrelated concerns |
| Open/Closed | Can behavior be extended without modifying existing code? | Frequent modifications to core classes for each new feature |
| Liskov Substitution | Can subtypes replace their base types without breaking behavior? | Type checks (instanceof) scattered through consumer code |
| Interface Segregation | Are interfaces focused and minimal? | "Fat" interfaces where consumers implement unused methods |
| Dependency Inversion | Do high-level modules depend on abstractions, not details? | Direct instantiation of infrastructure classes in business logic |
## SOLID Assessment
| Principle | Status | Evidence | Impact |
|-----------|--------|----------|--------|
| SRP | Concern | UserService handles auth, profile, notifications, and billing | High — changes to billing risk breaking auth |
| OCP | Good | Plugin system for payment providers | Low |
| LSP | Good | No type-checking anti-patterns found | Low |
| ISP | Concern | IRepository has 15 methods, most implementors use 3-4 | Medium |
| DIP | Concern | Controllers directly instantiate database repositories | Medium |
Got: Each principle assessed with at least one specific example. If fail: Not all principles apply equally to every architecture style. Note when principle is less relevant (e.g., ISP matters less in functional codebases).
Step 4: Review API Design
For systems that expose APIs (REST, GraphQL, gRPC):
- Consistency: Naming conventions, error formats, pagination patterns uniform
- Versioning: Strategy exists and is applied (URL, header, content negotiation)
- Error handling: Error responses structured, consistent, no leak internals
- Authentication/Authorization: Properly enforced at API layer
- Rate limiting: Protection against abuse
- Documentation: OpenAPI/Swagger, GraphQL schema, or protobuf definitions maintained
- Idempotency: Mutating operations (POST/PUT) handle retries safely
## API Design Review
| Aspect | Status | Notes |
|--------|--------|-------|
| Naming consistency | Good | RESTful resource naming throughout |
| Versioning | Concern | No versioning strategy — breaking changes affect all clients |
| Error format | Good | RFC 7807 Problem Details used consistently |
| Auth | Good | JWT with role-based scopes |
| Rate limiting | Missing | No rate limiting on any endpoint |
| Documentation | Concern | OpenAPI spec exists but 6 months out of date |
Got: API design reviewed against common standards with specific findings. If fail: No API exposed? Skip this step and focus on internal module interfaces.
Step 5: Evaluate Scalability and Reliability
- Statelessness: Can application scale horizontally (no local state)?
- Database scalability: Are queries indexed? Schema suitable for data volume?
- Caching strategy: Caching applied at appropriate layers (database, application, CDN)?
- Failure handling: What happens when dependency unavailable (circuit breaker, retry, fallback)?
- Observability: Logs, metrics, traces implemented?
- Data consistency: Eventual consistency acceptable or strong consistency required?
Got: Scalability and reliability assessed relative to stated non-functional requirements. If fail: Non-functional requirements undocumented? Recommend defining them as first step.
Step 6: Assess Technical Debt
## Technical Debt Inventory
| Item | Severity | Impact | Estimated Effort | Recommendation |
|------|----------|--------|-----------------|----------------|
| No database migrations | High | Schema changes are manual and error-prone | 1 sprint | Adopt Alembic/Flyway |
| Monolithic test suite | Medium | Tests take 45 min, developers skip them | 2 sprints | Split into unit/integration/e2e |
| Hardcoded config values | Medium | Environment-specific values in source code | 1 sprint | Extract to env vars/config service |
| No CI/CD pipeline | High | Manual deployment prone to errors | 1 sprint | Set up GitHub Actions |
Got: Technical debt catalogued with severity, impact, effort estimates. If fail: Debt inventory overwhelming? Prioritize top 5 items by impact/effort ratio.
Step 7: Review Architecture Decision Records (ADRs)
ADRs exist? Evaluate:
- Decisions have clear context (what problem was being solved)
- Alternatives considered and documented
- Trade-offs explicit
- Decisions still current (not superseded without documentation)
- New significant decisions have ADRs
ADRs do not exist? Recommend establishing them for key decisions.
Step 8: Write the Architecture Review
## Architecture Review Report
### Executive Summary
[2-3 sentences: overall health, key concerns, recommended actions]
### Strengths
1. [Specific architectural strength with evidence]
2. ...
### Concerns (by severity)
#### Critical
1. **[Title]**: [Description, impact, recommendation]
#### Major
1. **[Title]**: [Description, impact, recommendation]
#### Minor
1. **[Title]**: [Description, recommendation]
### Technical Debt Summary
[Top 5 debt items with prioritized recommendations]
### Recommended Next Steps
1. [Actionable recommendation with clear scope]
2. ...
Got: Review report actionable with prioritized recommendations. If fail: Review time-boxed? Clearly state what was covered and what remains unassessed.
Checks
- System context documented (purpose, scale, dependencies, team)
- Coupling and cohesion assessed with specific code examples
- SOLID principles evaluated where applicable
- API design reviewed (if applicable)
- Scalability and reliability assessed against requirements
- Technical debt catalogued and prioritized
- ADRs reviewed or their absence noted
- Recommendations specific, prioritized, actionable
Pitfalls
- Review code instead of architecture: This skill is about system-level design, not line-level code quality. Use
code-reviewerfor PR-level feedback. - Prescribe specific technology: Architecture reviews should identify problems, not mandate specific tools unless clear technical reason.
- Ignore team context: "Best" architecture for 3-person team differs from 30-person team. Consider organizational constraints.
- Perfectionism: Every system has tech debt. Focus on debt actively causing pain or blocking future work.
- Assume scale: Don't recommend distributed systems for app serving 100 users. Match architecture to actual requirements.
See Also
security-audit-codebase— security-focused code and configuration reviewconfigure-git-repository— repository structure and conventionsdesign-serialization-schema— data schema design and evolutionreview-data-analysis— review of analytical correctness (complementary perspective)
Dépôt GitHub
Compétences associées
executing-plans
DesignUtilisez la compétence executing-plans lorsque vous disposez d'un plan de mise en œuvre complet à exécuter par lots contrôlés avec des points de contrôle de revue. Elle charge et examine le plan de manière critique, puis exécute les tâches par petits lots (3 tâches par défaut) tout en rapportant la progression entre chaque lot pour une revue par l'architecte. Cela garantit une mise en œuvre systématique avec des points de contrôle de qualité intégrés.
requesting-code-review
DesignCette compétence délègue un sous-agent réviseur de code pour analyser les modifications apportées au code par rapport aux exigences avant de poursuivre. Elle doit être utilisée après avoir terminé des tâches, implémenté des fonctionnalités majeures, ou avant une fusion vers la branche principale. La revue aide à détecter précocement les problèmes en comparant l'implémentation actuelle avec le plan initial.
connect-mcp-server
DesignCette compétence fournit un guide complet permettant aux développeurs de connecter des serveurs MCP à Claude Code via les transports HTTP, stdio ou SSE. Elle couvre l'installation, la configuration, l'authentification et la sécurité pour intégrer des services externes tels que GitHub, Notion et des API personnalisées. Utilisez-la lors de la configuration d'intégrations MCP, de la configuration d'outils externes ou du travail avec le Protocole de Contexte de Modèle de Claude.
web-cli-teleport
DesignCette compétence aide les développeurs à choisir entre les interfaces Web et CLI de Claude Code en fonction de l'analyse des tâches, puis permet une téléportation transparente des sessions entre ces environnements. Elle optimise le flux de travail en gérant l'état et le contexte de la session lors du passage entre le web, la CLI ou le mobile. Utilisez-la pour des projets complexes nécessitant différents outils à diverses étapes.
