llm-fine-tuning-guide
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill utilizes official repositories and model hubs from trusted organizations, including Meta (Llama), Google (Gemma), and Mistral, alongside the OpenAI API for fine-tuning tasks.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided datasets for model training, which inherently involves handling untrusted data.
- Ingestion points:
scripts/data_preparation.pyprovides functions to ingest, validate, and format dataset samples. - Boundary markers: The skill uses explicit markdown templates to clearly separate instructions, contextual input, and model responses.
- Capability inventory: Capabilities are restricted to local file manipulation for dataset preparation and model training/evaluation using the
transformersandpeftlibraries. - Sanitization: Structural validation is provided through a
DatasetValidatorclass that checks for duplicates, empty samples, and length outliers.
Audit Metadata