model-verify-flagos

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly downloads user-specified models from public repositories (ModelScope via snapshot_download or HuggingFace IDs in Step 2) and then loads/runs them inside the container (commands include --trust-remote-code), so untrusted third-party model code and outputs directly influence execution, error classification, and recommended actions.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill downloads a remote model at runtime via modelscope.snapshot_download('<MODEL_ID>') (e.g., https://huggingface.co/<MODEL_ID>) and then runs inference with the --trust-remote-code flag, which allows fetched remote model code to be executed inside the container, so the remote model repo URL is a runtime dependency that can execute code.