Attached files can be referenced by an id called file_uuid. When Claude Code fetches these files, it now keeps a cache per context and reuses a stored copy when possible. A stored copy is used only when both of these hold:
- The sign-in token matches the one it was fetched with. The comparison uses a hash, a scrambled fingerprint of the token.
- The file on disk still matches its SHA-256 checksum, a fingerprint of the file's contents.
If the same file is already being fetched ahead of time, the new request waits for that download instead of starting a second one. When the cache is full, the oldest entry is removed. New counters record how many files were reused, joined or prefetched.
The same attachment is downloaded fewer times. The two checks stop Claude Code from using a copy that has changed on disk or was fetched under a different sign-in.