moe-training

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches configuration and code from Microsoft's official Megatron-DeepSpeed GitHub repository.
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of large-scale training datasets through the --data-path argument in training scripts.
  • Ingestion points: External data sources provided at runtime via command-line arguments in SKILL.md and references/training.md.
  • Boundary markers: The code snippets utilize standard variable interpolation for paths.
  • Capability inventory: The skill uses the deepspeed CLI for training execution and performs file system reads for data and vocabulary files.
  • Sanitization: Not present, as is standard for raw model training data pipelines.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 07:54 PM
Security Audit — agent-trust-hub — moe-training