external-mindstudio-msmodelslim-model-adapt

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 1.00). The skill explicitly downloads and loads arbitrary model repositories and files (e.g., via "modelscope download" and the --model-path / AutoConfig.from_pretrained / AutoProcessor.from_pretrained flows that read config.json and modeling_*.py, often with trust_remote_code=True), so the agent is required to ingest and potentially execute untrusted third-party model code and metadata that can materially influence subsequent actions.