moe-training

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFE
Full Analysis
  • EXTERNAL_DOWNLOADS (SAFE): The skill references standard machine learning libraries and a trusted repository.
  • Evidence: pip install deepspeed>=0.6.0 and git clone https://github.com/microsoft/Megatron-DeepSpeed are included in the training guide.
  • Context: The Microsoft GitHub organization is a trusted source, and DeepSpeed is a standard industry library. No suspicious remote execution patterns (like curl | bash) were found.
  • COMMAND_EXECUTION (SAFE): The skill contains example bash scripts and Python code for model inference and training.
  • Evidence: vllm.LLM(...) calls and deepspeed command-line examples.
  • Context: These are static documentation examples for performance optimization and do not contain hardcoded malicious commands or path traversal risks.
  • DATA_EXFILTRATION (SAFE): No patterns of sensitive data access or external network exfiltration were detected. URL references point to reputable research (arXiv) and official documentation.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:23 PM