effect-ai_skill

Overview

This skill provides type-safe integration for using large language models and AI providers through the Effect AI (@effect/ai) toolkit. It unifies providers like OpenAI and Anthropic, adds schema-validated structured output, and supports tool calling and multi-step execution plans. Use it to build reliable, testable AI workflows with typed inputs and outputs.

How this skill works

The skill exposes a common AiChat provider abstraction and provider-specific layers (for OpenAI, Anthropic) that you configure with API keys and model options. You define typed tools using schemas, call them from prompts, validate responses with generateObject and Schema, stream output, and compose multi-step AiPlan workflows to orchestrate complex tasks. It also offers testing layers and error handling primitives for retries and graceful failures.

When to use it

• Integrating OpenAI, Anthropic, or other LLM providers with a single API
• Enforcing typed inputs/outputs for AI responses and tool calls
• Building multi-step AI workflows or execution plans
• Streaming long or incremental model responses for better UX
• Writing unit tests or mocks for AI interactions

Best practices

• Define tools and AI inputs with Schema for compile-time type safety
• Use generateObject with Schema to validate structured outputs
• Implement retry and backoff for rate limit and transient errors
• Mock AiChat in tests to avoid calling real APIs
• Stream large responses to improve responsiveness and memory usage

Example use cases

• Create a weather lookup tool with a typed input schema and let the model call it
• Compose a research AiPlan that searches, analyzes results, and returns a schema-validated summary
• Build a chatbot that uses multiple tools (search, calculator) and returns structured JSON
• Stream a generated story or long report to the client while the model produces chunks
• Mock AI outputs in CI to test application logic without external API calls