Claude Agent SDK (TypeScript)

Production patterns for integrating the Anthropic Claude Agent SDK in TypeScript/Node.js projects. Covers the full operational stack from provider routing and credential management to spending cap enforcement, structured audit logging, and MCP tool registration. Grounded in patterns from Shannon — a 13-agent autonomous security testing pipeline achieving 96.15% benchmark success using Claude Agent SDK v0.2.38.

When to Use This Skill

Use this skill for:

• Building autonomous Claude agents in TypeScript/Node.js
• Implementing multi-provider LLM routing (Anthropic, AWS Bedrock, Google Vertex AI, OpenRouter)
• Adding spending cap enforcement to prevent runaway agent costs
• Structuring per-invocation audit logging for agent observability
• Classifying errors as retryable vs. fatal in agent workflows
• Registering MCP tools within Claude Agent SDK context
• Designing agent identity and specialization via system prompts

Trigger phrases: "Claude Agent SDK", "Claude SDK TypeScript", "autonomous agent Node.js", "multi-provider Claude", "agent spending cap", "Claude SDK retry", "Claude SDK audit", "agent SDK production"

What This Skill Does