---

name: hub-and-cli description: "Use ModelScope Hub, cache, download/upload, and CLI workflows safely from the modelscope package." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

ModelScope Hub and CLI

Use this sub-skill when the task is about ModelScope Hub access, the modelscope/ms command line, downloading repositories or files, upload command selection, authentication boundaries, endpoint/cache settings, offline cache reuse, or cache troubleshooting.

Fast routing

• Need CLI command selection, help boundaries, download/cache syntax, or legacy command aliases: read CLI reference.
• Need Python APIs such as snapshot_download, dataset_snapshot_download, model_file_download, dataset_file_download, HubApi, credentials, endpoint variables, or modelscope_hub compatibility behavior: read Hub API reference.
• Need to diagnose auth, repo id/revision, endpoint, network, cache, offline, missing modelscope_hub, or local_dir/cache_dir confusion: read troubleshooting.
• Need a safe dry-run plan before downloading: run python scripts/plan_download.py --help, then use the generated CLI command or Python snippet after approval. The script performs no network calls.

Boundaries and cross-skill routes

• This sub-skill stops at obtaining or managing local Hub files. For pipeline inference after a model download, route to ../pipelines-and-models/SKILL.md.
• For MsDataset.load, dataset streaming, dataset recipe validation, or config/file I/O after a dataset snapshot is available, route to ../datasets-config/SKILL.md.
• For server/studio deployment details, vLLM integration, export, and large checkpoint tools, route to ../serving-export-and-tools/SKILL.md.
• Treat uploads, cache clearing, login persistence, and endpoint changes as operations with external side effects. Ask for explicit user approval before mutating remote Hub repositories, deleting caches, or saving credentials.

Working checklist

1. Identify the interface: modelscope/ms CLI, Python download helper, or HubApi.
2. Confirm repo id shape (owner/name), repo type (model or dataset for most downloads), revision, and whether the target is public or private.
3. Decide storage semantics: use local_dir for a user-visible copy, cache_dir/MODELSCOPE_CACHE for reusable cache, and local_files_only=True only when the cache is already populated.
4. Avoid exposing tokens. Prefer environment variables and Python token= parameters over literal tokens in shell history.
5. For include/exclude filters, quote glob patterns in shell commands and remember that explicit file paths take precedence over patterns.