What problem does it solve? LLM outputs are often unreliable, inconsistent, or hallucinated when prompts are written through trial and error. This Skill replaces guesswork with proven prompting patterns so you get consistent, structured, and cost-efficient responses from models like GPT-4, Claude, and Gemini. ## Core Features & Use Cases - Technique Decision Framework: Choose between zero-shot, few-shot, chain-of-thought, JSON mode, tool calling, prompt chaining, and RAG based on task requirements, token cost, and reliability. - Structured Output Generation: Produce validated JSON using OpenAI JSON mode, Anthropic tool use, and Zod schemas with the Vercel AI SDK. - Production Practices: Prompt versioning, token cost monitoring, retry logic, injection sanitization, and A/B testing workflows. - Use Case: You are building a RAG question-answering feature and the model keeps returning malformed JSON. Use this Skill to switch to tool calling with a schema, add validation, and reduce hallucination with cited sources. ## Quick Start Ask the AI to design a chain-of-thought prompt with structured JSON output for your classification task using the prompt engineering skill.
