How does prompt caching fundamentally work?	The initial request processes and saves work to a temporary cache; subsequent identical requests load this saved work instead of reprocessing.
How long is the saved work in the prompt cache valid?	It is only stored for one hour.
How is prompt caching enabled in Claude?	It must be manually enabled by adding a cache breakpoint to a block within a message.
What must be true about a follow-up request to utilize cached work?	The content must be identical up to and including the cache breakpoint.
What content is cached when a breakpoint is added to a block?	All content in the request up to and including that breakpoint.
List three types of elements where a cache breakpoint can be placed.	Message blocks, tool schemas, and system prompts (or image blocks, tool use, tool results).
Why is the "long form" structure required when enabling cache control?	The short form (direct string assignment) does not provide a field to add the cache control.
In what order are tools, system prompts, and messages joined when fed into Claude?	Tools, then the system prompt, then the list of messages.
What is the minimum content length required for content to be written to the cache?	At least 1024 tokens.