llama-factory
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill provides numerous examples of shell commands for installing software, configuring environments, and running model training/inference (e.g.,
llamafactory-cli,torchrun,deepspeed,accelerate). These are standard instructions for the intended use-case of the LLaMA-Factory tool. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided datasets and model configurations, which represents an indirect injection surface.
- Ingestion points: External datasets in Alpaca or ShareGPT formats and pre-trained models from Hubs (referenced in
references/getting_started.md). - Boundary markers: Not explicitly specified in the documentation; the system relies on internal library parsing logic.
- Capability inventory: The skill facilitates file system writes for model checkpoints, execution of CLI tools like
llamafactory-cli, and network operations for downloading model weights and datasets. - Sanitization: No specific sanitization or filtering of dataset content is described in the provided reference materials.
- [EXTERNAL_DOWNLOADS]: The documentation describes downloading dependencies, models, and development tools from established and reputable platforms including HuggingFace, ModelScope, NVIDIA's developer portal, and the official LLaMA-Factory GitHub repository.
Audit Metadata