implementing-llms-litgpt
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides standard documentation for machine learning operations using the LitGPT framework. All external resources, including the
litgptlibrary itself and model weights from providers like Microsoft, Google, and Meta, are well-known and trusted within the research community. - [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of user-defined training data, which inherently creates a surface for indirect prompt injection attacks. This is a fundamental aspect of the skill's intended machine learning purpose.
- Ingestion points: Data is loaded from local files such as
data/my_dataset.jsonanddata/my_corpus.txt(SKILL.md). - Boundary markers: Training examples do not explicitly include safety delimiters or instructions to disregard embedded commands.
- Capability inventory: The skill invokes
litgpttraining utilities and execution of local preparation scripts via the shell (SKILL.md). - Sanitization: The workflow relies on standard tokenization and dataset preparation steps within the LitGPT and PyTorch ecosystem.
Audit Metadata