claude-sdk-integration-patterns_skill

Overview

This skill provides production-ready integration patterns for the Claude API and the TypeScript SDK, focused on streaming, tool execution, error handling, token optimization, and deployable implementations. It packages practical code patterns and architecture advice to build reliable, cost-efficient AI features in Node.js/TypeScript applications.

How this skill works

The skill documents concrete patterns for the Messages API, streaming response handling, tool (function) definitions and runners, batch requests, retry/backoff strategies, token counting, and context pruning. Each pattern includes when-to-use guidance and short code sketches that show initialization, streaming loops, tool validation, and error recovery for production use.

When to use it

• Adding Claude conversational features to Node.js/TypeScript apps
• Delivering real-time UX with streaming responses
• Creating AI agents that call external tools or APIs safely
• Handling high-volume or bulk tasks with message batching
• Deploying production-ready systems with retries, timeouts, and monitoring

Best practices

• Keep API keys in environment variables and never hardcode secrets
• Choose the model that matches task needs (Opus for complex reasoning, Sonnet for balance, Haiku for speed)
• Stream user-facing long outputs; avoid streaming for deterministic batch jobs
• Implement exponential backoff for 429s and sensible timeouts for network reliability
• Estimate tokens, set max_tokens, and prune or summarize old context to control costs

Example use cases

• Chatbots with multi-turn context management and streaming UI updates
• Agent workflows that validate inputs with Zod tools and run external APIs
• Bulk document summarization using message batching to reduce per-request cost
• Real-time code assistants that stream partial outputs and support user abort
• Customer support automation with robust retry and rate-limit handling