ai-integration

This skill provides a robust architectural pattern for integrating Anthropic's Claude API into modern software applications. It is designed for developers building AI-powered features within a TypeScript ecosystem, offering standardized methods for handling request lifecycle management, response streaming, and complex function calling. By leveraging this skill, engineers can rapidly scaffold conversational interfaces, automated content generation engines, and adaptive workflows that require deterministic structured outputs. The module emphasizes maintainability by providing clear examples for API client initialization, message handling, and integrating Zod schemas to ensure type-safe data parsing from AI responses.

• Standardized Anthropic SDK client setup with environment variable configuration for secure API access.

• Flexible chat completion patterns supporting both basic request-response cycles and efficient server-side event streaming via ReadableStream.

• Advanced structured output generation using Zod schemas for validation, perfect for generating JSON-compliant domain objects like workout plans or data reports.

• Comprehensive Tool Use (Function Calling) architecture that allows Claude to execute defined actions, such as calendar updates or intensity adjustments, with clear input schema validation.

• Predefined model selection strategies mapping task requirements (fast, balanced, or powerful) to specific Claude model versions (e.g., Haiku or Sonnet).

• Best practice: Always include a system prompt to define the AI's persona, operational constraints, and required output format to minimize hallucination.

• Practical Tip: When using tool use, provide descriptive schemas for properties and requirements to ensure the model reliably predicts correct parameter values.

• Constraint: Ensure that the environment variable ANTHROPIC_API_KEY is correctly injected into your runtime environment to prevent client initialization errors.

• Usage: Ideal for applications requiring high-fidelity interaction, such as AI coaches, personal assistants, or automated document processing pipelines.

• Data Integrity: Always wrap the output of generate methods in schema validation logic to handle cases where the model might deviate from strictly enforced JSON formatting.