prompt-cachinglisted

# Prompt Caching & Cost Efficiency Prompt caching is the single biggest lever for cutting token cost. Cached tokens are re-read at **~0.1x** the normal input price; the first write costs **~1.25x**. A warm session can pay a tenth of the price for everything it has already seen - far more savings than any model downgrade. The whole game is keeping the cache warm. ## The one rule everything follows from **Caching is a prefix match. Any change anywhere in the early context invalidates everything after it.** The cache reuses content from the start of the conversation up to the point where something changes. Edit something near the beginning and every token after it must be re-read at full price. Append something new at the end and the entire prior prefix is still served from cache. ## Cache-friendly workflow in Claude Code - **Keep `CLAUDE.md` and early context stable.** It sits at the front of the prefix - editing it mid-session re-reads everything below it at full price. Get it right early, then leave it alone. - **Don't re-read or re-edit files you touched at the start of the session.** Reading file X early, then editing X again 30 turns later, churns the prefix. Do the work on X while it's still recent, or batch all edits to X together. - **Batch related edits.** Several edits to one file in one turn cache better than the same edits spread across many turns interleaved with other work. - **Don't thrash the tool or file set early.** A stable, predictable sequence of oper