Skills
Separate config from state at the prompt layer and add explicit cache breakpoints for 2–5× hit-rate gains
Beyond automatic prefix caching, explicitly marking segments with cache_control breakpoints (system prompt cached separately from tool definitions, long reference docs cached, dynamic user context left uncached) delivers 2–5× hit-rate improvement over automatic caching alone for ~1–2 hours of work. The discipline is structural: static configuration goes in cacheable zones, mutable state stays out, and a CI assertion guards the boundary.
Source
↳ Follow the thread