What
claude --cloud starts a cloud session: a Claude Code session that runs on Anthropic's machines rather than your own. File sync copies your local files to that session and keeps both sides in step. This release adds a large set of new file-sync status and error messages, including the following.
- File sync now has a switch on Anthropic's side that can turn it off. When the service refuses sync, sync pauses for the session instead of ending it. Both sides keep the files they already have, and each message you send checks once whether sync is back on. When it comes back, the session catches up.
- While sync is paused, if the session does not yet have any of your files, your messages stay on this machine. The error text suggests starting a cloud session without file sync as a workaround.
- When a cloud session can't start from GitHub, the hint about uploading without file sync now depends on why your local state couldn't be read. The new causes are
withheld_unlistable,withheld_list_unapplied,withheld_list_unsettledandwithheld_versions_too_many. They get a third wording that ends 'such files then go as committed; untracked files stay here'. Thehost_configcause still says only commits are uploaded.
This site has not read the tengu_dir_sync_git_kill_switch gate that goes with the switch.
Why
A session that is already running should keep working if sync is turned off from Anthropic's side, and you get a plain explanation of what happened. The new hint tells you more accurately what the fallback without file sync (CLAUDE_CODE_DIR_SYNC_ENGINE=none) will upload and what stays on your machine.
It is not clear what else decides whether cloud file sync is available at all in this build.