Under the hoodTier: how much it should matter to you
1Useful: my rating, 1 to 5
1Signal: worth watching, 1 to 5
TelemetryArea: what it touches
Internal ChangesKind: in v2.1.315,
Internal ChangesSection of the release
What
The prompt cache lets the model reuse work it has already done on the start of a conversation. Claude Code tracks when that cache stops matching, using fingerprints of the instructions and the tool list. It now stores these fingerprints separately for each source rather than as a single pair, and uses the previous record when making new ones.
Why
This only affects diagnostics and changes nothing you see.
