huggingface-local-models
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides instructions to compile the llama.cpp project from source using make and run the resulting binaries (llama-cli, llama-server). It also includes commands to execute Python scripts (convert_hf_to_gguf.py) for model conversion.
- [EXTERNAL_DOWNLOADS]: Fetches the llama.cpp source code from its official GitHub repository and downloads model checkpoints from the Hugging Face Hub using the hf CLI tool. It also queries the Hugging Face Tree API for file metadata.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (model repository names, file paths, and recommended quantization labels) from the Hugging Face Hub to construct and execute shell commands.
- Ingestion points: Hugging Face Hub search results and Tree API responses as described in references/hub-discovery.md.
- Boundary markers: Absent.
- Capability inventory: Execution of local binaries (llama-cli, llama-server) and Python scripts as detailed in SKILL.md and references/quantization.md.
- Sanitization: Absent.
Audit Metadata