Specialized personas Claude Code delegates to for specific work — architecture reviews, security audits, debugging, data analysis. Each ships as a Markdown file in .claude/agents/ with a system prompt and a scoped tool set.
8 sub-agents
Designs RESTful APIs, microservice boundaries, and database schemas. Use for new backend services or when refactoring cross-service contracts.
Profiles hot paths, spots N+1 queries, suggests caching layers, and fixes render bottlenecks. Use when a page or endpoint is slow and root cause is unknown.
Ships React/Next.js UI components with accessibility, state management, and performance baked in. Use when building or refactoring UI code.
Reviews code for OWASP-style vulnerabilities: injection, auth flaws, secret leaks, unsafe deserialization. Use before merging any code that touches auth, inputs, or external systems.
Writes unit, integration, and e2e tests. Picks the right framework, structures the test suite, and generates fixtures. Use when a feature ships without tests.
Designs multi-region cloud topologies on AWS/GCP/Azure — VPCs, load balancers, IAM, cost tradeoffs. Use when moving off a single-region monolith or planning a migration.
Designs Kubernetes clusters, ingress topology, RBAC, and workload placement. Use for greenfield k8s setups or when scaling past 10 services on a single cluster.
Drives strict red-green-refactor: writes a failing test, runs it, implements just enough to pass, refactors. Use when you want to enforce TDD on a new feature.