Anthropic Prompt Caching and Context Runtime

When to use

Invoke when ai-architecture.md defines a Claude capability with a cost or latency budget and the caching and context behavior — cache_control breakpoint placement, cache-hit economics, extended-thinking budget and thinking-block retention, or long-context packing — must be made deliberate rather than incidental.

Do not use to decide whether a feature should use AI, to pick a model tier, to design output schemas, to execute tools, to design retrieval topology, or to design agent control flow.

Inputs

Required: