hugging-face-model-trainer
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS] (LOW): Documentation and example scripts (e.g., train_grpo_example.py) reference the execution of training scripts hosted on the 'huggingface' GitHub organization. This is considered low risk as the source is an explicitly trusted organization.\n- [REMOTE_CODE_EXECUTION] (LOW): The skill utilizes the hf_jobs interface to execute remote Python scripts. This behavior is downgraded to low risk because the target scripts are hosted by a trusted organization (Hugging Face).\n- [COMMAND_EXECUTION] (LOW): The GGUF conversion guide suggest using subprocess calls to execute system-level commands like 'apt-get' and 'cmake'. These are typical for environment setup in ephemeral training jobs.\n- [PROMPT_INJECTION] (LOW): Ingestion surface for indirect prompt injection via 'load_dataset' in training scripts (e.g., train_sft_example.py). Boundary markers: Absent. Capability inventory: 'trainer.train()', 'trainer.push_to_hub()'. Sanitization: Absent.
Audit Metadata