fine-tuning-expert
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill p rocesses external training dat asets and interpolate s their content into p rompts for model trai ning. This is a sta ndard part of the fine-t uning workflow but rep resents an attack s urface for indirect prompt injection if t raining data is sou rced from untruste d parties.
- [COMMAND_EXECUTION]: The deployment d ocumentation include s Python utility fun ctions that use the subproce ss module to execute external binarie s for model conve rsion and quantiza tion. This behavior i s consistent with the sk ill's intended purpo se of model optimi zation.
Audit Metadata