Skills
skills/huaweicloud/huaweicloud-skills/huawei-cloud-msmodelslim-model-adapt/Gen Agent Trust Hub

huawei-cloud-msmodelslim-model-adapt

Pass

Audited by Gen Agent Trust Hub on Jun 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/step2_run_quantization.py to invoke the msmodelslim quantization module. This is implemented securely by passing arguments as a list and using the current Python environment's executable.- [REMOTE_CODE_EXECUTION]: Several scripts and templates (e.g., scripts/step1_generate_test_model.py, assets/vlm_model_adapter_template.py) enable the trust_remote_code=True parameter within the Hugging Face Transformers library. This functionality is a standard requirement for loading models with custom architectures, allowing the execution of logic defined in the model's own repository.- [EXTERNAL_DOWNLOADS]: The documentation references downloading model non-weight files from ModelScope, a well-known model repository. These downloads are part of the intended model adaptation process and target established infrastructure.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 17, 2026, 09:10 AM
Security Audit — agent-trust-hub — huawei-cloud-msmodelslim-model-adapt