Prompt Cache Linter

Prompt caches match exact byte prefixes, so one timestamp at the top of your system prompt quietly re-bills your whole prefix on every request. Paste one payload for a volatility scan, or paste two consecutive real requests and this finds the exact character where your cache breaks, the stable content stranded below it, and what that is worth per month. Companion to Prompt Caching Is Free Money.

# Your prompts never leave this tab. No upload, no server. The linter is plain JS on this page. View source.

Assumptions: tokens estimated at ~4 characters each. Savings math uses the selected model's input price from the hand-verified reference table with Anthropic-style cache economics (reads at 10% of input price); OpenAI caches automatically at a smaller discount, so treat the dollar figure as the upper bound and the location of the break as the universal part. Byte-exact matching means your SDK's serialization matters too: keys in a different order lint clean here but still miss in production, which is why the two-payload mode is the honest test. The engine is open source, tested, with a CLI: github.com/AlexRyan92/prompt-cache-linter.