awesome-ai-agent-skills: Skill for Claude Code

System Architecture & Reliability Skill

Covers modern AI-Native software architecture, distributed consistency, and living documentation standards.

1. Modular Monolith Architecture

• Motivation: Keeps all domain boundaries logically separated but physically within a single repository, maximizing LLM context window clarity and enabling safe cross-domain refactoring.
• Rules:
  • Enforce explicit module boundaries. Modules must communicate via well-defined domain APIs or internal event buses.
  • No direct circular dependencies between modules.

2. Transactional Outbox Pattern & CDC

• Problem: Solves the Dual-Write problem when persisting state to a DB and emitting event messages to Kafka/RabbitMQ/Service Bus.
• Implementation:
  1. Write business data and insert an event payload into a local outbox table within the same local DB transaction.
  2. Use a dedicated Message Relay or Change Data Capture (CDC) tool (e.g., Debezium, DynamoDB Streams) to tail the database transaction log and publish messages asynchronously.
  3. Idempotency: All event consumers (including AI agents) must maintain an idempotency_key table to deduplicate retried messages.

3. Living Architecture Decision Records (ADRs)

• Maintain an docs/adr/ directory with records of major architectural choices.
• Living ADR Rule: Update existing ADRs with date stamps and production feedback rather than creating conflicting duplicate documents. Feed ADRs as context injection to AI Agents before structural changes.