Optimizes LLM token costs and latency by implementing provider-native prompt caching for Claude and OpenAI.
The Prompt Caching skill provides developers with a production-ready framework to implement native caching strategies for Claude and OpenAI models within Claude Code. By strategically placing cache breakpoints and organizing message structures to prioritize stable prefixes—such as system prompts, tools, and few-shot examples—this skill enables up to 90% cost savings on input tokens. It includes templates for Python implementations, guidance on TTL management (5m vs 1h), and monitoring tools to calculate cache hit rates, making it essential for high-volume AI applications and cost-sensitive development.
Características Principales
01Configurable TTL settings for ephemeral (5m) and extended (1h) cache persistence.
02Standardized cost-tracking templates for calculating real-world savings.
03Automatic OpenAI prefix caching optimization and monitoring.
04Best-practice prompt ordering to ensure high cache hit rates.
05Native Claude cache breakpoint implementation with support for up to 4 breakpoints.
0669 GitHub stars
Casos de Uso
01Reducing costs in chatbots with long system instructions or fixed context.
02Speeding up automated code generation by caching large repository structures.
03Optimizing RAG pipelines where the retrieved context remains stable across multiple turns.
