What problem does it solve? Writing effective LLM prompts is often ad-hoc and untested, leading to inconsistent outputs, no version history, and no way to measure which prompt variant performs better. This Skill provides structured patterns for building, testing, and managing prompts systematically. ## Core Features & Use Cases - Structured Prompt Builders: Compose system prompts from ordered sections (role, task, constraints, output format) and build few-shot or chain-of-thought prompts programmatically. - Prompt Optimization & Evaluation: Run A/B tests between prompt variants with epsilon-greedy selection, and score prompts on clarity, specificity, completeness, and length with actionable suggestions. - Prompt Versioning & Templates: Register versioned prompts in a file-based registry and render reusable templates with variable substitution. - Use Case: You are building a sentiment classification feature. Use this Skill to create a few-shot prompt with labeled examples, register it as v1, A/B test it against a chain-of-thought variant, and promote the winner based on measured success rates. ## Quick Start Design a structured system prompt with few-shot examples for classifying customer support tickets by urgency, then set up A/B testing to compare it against a chain-of-thought variant.
