We're rolling out Claude, Cowork, and Claude Code, but we've had several incidents where Cowork data could no longer be opened by Claude. In some cases, users lost a week or two of work, and the long file paths used by Cowork can exceed Windows path-length limits, causing problems with tools such as OneDrive and robocopy. Has anyone found a reliable way to back up, restore, and organize this data?
4 Answers
Keep important working data in OneDrive or another managed file location that is included in your backup schedule. We back up users’ OneDrive data several times a day and advise them not to rely on Claude projects as the only copy of anything important.
Avoid treating Claude project folders as the permanent storage location. Create a normal folder such as C:cowork, add the relevant folders to the project, and instruct Claude to save deliverables there. Use projects for ongoing context, while keeping the actual files in a location covered by your normal backup process.
Standard robocopy jobs may fail because Cowork can create paths longer than Windows normally supports. One workaround is using bsdtar to archive the data; it handles long filenames better. The resulting archive may not open through File Explorer, but it can be extracted with PowerShell. A temporary staging folder can also make the process easier to manage.
For files that are already inaccessible or behaving strangely, contact Anthropic support directly. They may be able to determine whether the problem is a corrupted project, a product issue, or an unsupported file structure.
That’s the important distinction: a normal copy looks fine on paper, but long paths and sync clients can cause failures. Test the archive and restore process before relying on it for production backups.