If you're building a harness, here is a simple tool to catch cache invalidation in your calls to LLMs

https://preview.redd.it/o6c5enqx3zdh1.png?width=1371&format=png&auto=webp&s=20f75d4cefa5f51f40c000db8f9bd7114758b354 Hello, I know we're a lot of harness builders out there, because it's fun and because it makes us learn a lot. I've been focusing on a local-first harness and prefill costs become obv