What problem does it solve? AI features embedded in applications often produce inconsistent outputs, hallucinated fields, or unreliable JSON because their prompts lack structure, examples, and explicit constraints. This Skill provides a systematic protocol for writing, testing, and versioning prompts so LLM features behave predictably in production. ## Core Features & Use Cases - Structured System Prompts: Builds prompts from four components (role, goal, constraints, format) with TypeScript templates for extraction, classification, summarization, RAG, and code generation tasks. - Few-Shot & Chain-of-Thought Patterns: Supplies ready-to-adapt examples for classification and multi-step reasoning, plus RAG context injection with boundary instructions that prevent hallucination outside provided context. - Prompt Versioning & Evaluation: Establishes a versioned prompt registry with changelogs and an eval harness that tests prompts against fixed input sets to measure accuracy over time. - Use Case: Your invoice extraction feature keeps inventing missing dates. Use this Skill to add explicit null-handling constraints, few-shot edge-case examples, and a Zod-schema-aligned output format, then verify the fix with an eval test suite. ## Quick Start Ask the AI to audit your existing system prompt for an extraction or classification feature and rewrite it using the four-component structure with few-shot examples.
