adapting-transfer-learning-models

Pass

Audited by Gen Agent Trust Hub on Mar 12, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill is designed to fetch pre-trained models and datasets from well-known repositories such as the Hugging Face Model Hub and Dataset Hub.
  • [COMMAND_EXECUTION]: The skill uses the Bash(cmd:*) tool to automate machine learning pipelines, including model training, evaluation, and data validation scripts.
  • [REMOTE_CODE_EXECUTION]: The skill's workflow involves generating Python code using ML frameworks like TensorFlow or PyTorch and executing it to perform model fine-tuning and adaptation.
  • [PROMPT_INJECTION]: The skill ingests untrusted data from user requests and external datasets (e.g., image directories or text files) without explicit boundary markers, creating a surface for indirect prompt injection.
  • Ingestion points: User requests for fine-tuning and external datasets specified via input parameters in SKILL.md and scripts/adapt_model.py.
  • Boundary markers: No delimiters or explicit instructions to ignore embedded commands were identified in the skill instructions.
  • Capability inventory: Access to Bash(cmd:*), Write, Edit, Read, Glob, and Grep tools.
  • Sanitization: No validation or sanitization of the content within processed datasets is performed before use in the model adaptation process.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 12, 2026, 02:53 AM