hugging-face-model-trainer
Pass
Audited by Gen Agent Trust Hub on Mar 10, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches dataset inspection scripts and training templates from external repositories on Hugging Face and GitHub, which are recognized technology platforms.
- [REMOTE_CODE_EXECUTION]: The skill utilizes the
hf_jobsMCP tool to execute Python training scripts on remote cloud infrastructure, which is the primary purpose of the skill. - [COMMAND_EXECUTION]: The skill instructions include bash commands for installing Python dependencies and running the Hugging Face Jobs CLI directly.
- [PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it ingests untrusted datasets from external sources and processes them via remote code execution.
- Ingestion points: Dataset loading via
load_dataset()in training script templates and thedataset_inspector.pyutility. - Boundary markers: Absent; the skill does not implement delimiters or safety instructions to distinguish data from command context.
- Capability inventory: Remote code execution via
hf_jobsas documented inSKILL.md. - Sanitization: Absent; the skill passes raw dataset content to model trainers.
Audit Metadata