China Mobile Cloud Drive File Management

Problem

When an LLM works with China Mobile Cloud Drive files, the tricky part is mixing cloud paths, file IDs, login state, and sandbox boundaries. That can cause searches, uploads, downloads, or copies to fail or touch the wrong scope. cm-cloud-manage wraps common drive operations in a command-line interface and separates full-drive read search from sandbox-scoped writes.

How It Works

The entry point is python3 {baseDir}/scripts/main.py; check each subcommand with -h. The usual workflow is to search or list for IDs, then call workspace to get the sandbox path and fileId.
- Search: search_keyword searches all files by filename keyword; search_filter filters by -t, --ext, --start-at / --end-at, and --size-min / --size-max, but does not accept keywords. Use --limit and the previous page meta cursor for pagination.
- List: list takes up to 20 comma-separated dirId values and lists direct children by default; add -r to include subdirectories.
- Sandbox writes: workspace returns the sandbox path and fileId; mkdir creates a folder whose name cannot contain /; upload sends a local localPath to a target dirId. For mkdir, upload, and download source files, the target/source must be inside the sandbox.
- Download: If the source is outside the sandbox, first copy it into the sandbox, then download. copy can use full-drive srcId values, while --to must be a sandbox directory; the returned newFileId is the copy ID.
- Auth: When record: auth appears, send the loginUrl for login and retry the same command after the user confirms.

Boundaries

Use this for scripted search, directory lookup, upload, copy-to-sandbox, and download-from-sandbox. Prefer interface-returned IDs over cloud paths except for the root /. Confirm write operations, keep time filters paired, pass cursor exactly, remember that search_filter does not accept keywords, and do not download non-sandbox files directly.