llm-fine-tuning-guide
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFE
Full Analysis
- [Indirect Prompt Injection] (SAFE): The skill contains a data formatting utility in
scripts/data_preparation.pythat interpolates external data into prompt templates for instruction tuning. While this is an inherent surface for indirect prompt injection, it is the standard and necessary workflow for the skill's primary purpose. Boundary markers are present to help the model distinguish instructions from data. - Ingestion point:
scripts/data_preparation.py(line 92,format_for_instruction_tuningfunction) - Boundary markers: Present (uses
### Instruction:,### Input:, and### Response:delimiters) - Capability inventory: The skill uses standard model training functions (
Trainer.train) and data manipulation; no high-risk system-level capabilities are exposed. - Sanitization: Absent (standard for LLM training scripts).
- [Unverifiable Dependencies] (SAFE): All dependencies listed in the documentation and used in the code are reputable, widely-used machine learning libraries from trusted maintainers.
- [Remote Code Execution] (SAFE): No instances of arbitrary command execution, shell piping, or dynamic execution of untrusted remote content were found.
Audit Metadata