Microservices Architect guides Claude through distributed system design in six steps: domain analysis, communication design, data strategy, resilience, observability and deployment. Each step has a validation checkpoint, for example that every service owns its data, every external call has a timeout and retry budget, and one request can be traced end to end by its correlation ID.
Five reference files cover decomposition, communication, resilience patterns, data management and observability.
What You Get
- A service boundary diagram with bounded contexts
- Communication patterns and a REST, gRPC and event decision matrix
- A data ownership and consistency model with sagas, event sourcing and CQRS
- Resilience settings for each integration: circuit breaker, retry, bulkhead and timeout
- Observability design with tracing, correlation IDs, SLOs and alert levels
- Deployment requirements with probes and a canary or blue-green plan
Common Use Cases
- Splitting a monolith into services along domain boundaries
- Choosing between REST, gRPC and messaging for service calls
- Designing a saga for an order, payment and shipment flow
- Adding circuit breakers and graceful degradation to service calls
- Setting up distributed tracing and SLO-based alerting
What Makes It Unique
The checkpoints make the design testable: no shared database schema, async messaging for long-running or cross-aggregate work, an explicit timeout and fallback on every external call, and end-to-end tracing. The constraints name the distributed monolith and chatty interfaces as things to avoid, and the references add Conway’s law, SLO and error budget guidance.
Under The Hood
The skill is a SKILL.md with five reference files and no scripts, and the references load by topic. It is a design skill: the agent produces diagrams, documents and sample code and does not need to run commands in the workflow. The examples call an inventory service over HTTP with a timeout, set response headers, run saga steps against services and database tables, and set up tracing. It names Kafka, RabbitMQ, Istio, Jaeger and Zipkin as options but installs nothing and fetches no web pages.