What problem does it solve? Building features that call external AI APIs is slow and expensive to test without a deterministic mock layer. This Skill provides a structured work procedure for implementing AI-powered features (multimodal analysis, quest generation, clustering) with a mock/real toggle so development and testing run without live API keys. ## Core Features & Use Cases - Mock-First AI Development: Create deterministic mock response files in src/lib/ai/mock/ before writing any real API integration, toggled via the USE_MOCK_AI environment variable. - Schema-Validated AI Contracts: Define input/output schemas with Zod, design prompt templates, and validate both mock and real responses through the same parsing path. - End-to-End TDD Workflow: Covers AI client implementation, API route with input sanitization, UI loading/error/retry states, bilingual locale testing, and browser verification via agent-browser. - Use Case: Implementing a GPT-4o image analysis endpoint for a children's talent-discovery app — you create three mock talent-detection responses, write failing tests, build the OpenAI client with mock toggle, and verify the full flow in both English and Indonesian locales. ## Quick Start Use the ai-feature-worker skill to implement the multimodal artifact analysis endpoint with mock responses and tests.
