optimizing-ai-model-and-prompt-costs

What problem does it solve? AI work often wastes money through wrong model selection, unmeasured retries, and poorly structured prompts. This Skill finds the lowest total cost that still meets quality requirements by classifying task risk, gating cost surfaces, and separating stable cacheable prompt content from volatile data. ## Core Features & Use Cases - Model and Reasoning Routing: Classify tasks as SIMPLE_BULK, ROUTINE_BALANCED, or HIGH_RISK_REASONING and recommend logical model tiers (Luna/Terra/Sol) plus reasoning effort levels. - Cost Surface Gate: Distinguish subscription-included usage from separately metered API, credits, or SaaS paths before any cost calculation, blocking unapproved paid routes. - Prompt Cache Design: Split prompts into stable prefixes and dynamic suffixes while excluding secrets, tokens, and volatile values from cacheable regions. - Use Case: Before starting a large batch of log summarization tasks, invoke the [모델 추천] checkpoint to get a model tier and reasoning recommendation, then estimate total cost including retries and higher-model rework. ## Quick Start Ask the assistant to recommend a model and reasoning level for your current task using the [모델 추천] checkpoint before proceeding.