ml-model-integration
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill interacts with HuggingFace Hub, a well-known service, to discover and download machine learning models and metadata using official client libraries.
- [INDIRECT_PROMPT_INJECTION]: The workflow involves processing external data for model evaluation, which presents a surface for indirect instructions.
- Ingestion points: Text data processed via the
evaluate_modelfunction inSKILL.md. - Boundary markers: Not present in the implementation logic to distinguish instructions from data.
- Capability inventory: Includes model inference and file system writes for saving fine-tuned models in
SKILL.md. - Sanitization: Input text is passed directly to the inference pipeline without filtering or escaping.
Audit Metadata