The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates downloading the llama.cpp source code from its official GitHub repository and model files from the Hugging Face Hub.
[COMMAND_EXECUTION]: Provides instructions for compiling software using make and running local binaries like llama-cli and llama-server to perform inference and model quantization.
[PROMPT_INJECTION]: Contains a surface for indirect prompt injection (Category 8) as it instructs the agent to parse and use metadata (repository names, file paths, and quantization labels) from the Hugging Face Hub to construct shell commands.
Ingestion points: Metadata and file tree information fetched from the Hugging Face API (https://huggingface.co/api/models/<repo>/tree/main).
Boundary markers: No explicit instructions or delimiters are used to ensure the agent ignores potentially adversarial instructions embedded in repository names or descriptions.
Capability inventory: The skill uses subprocess calls to execute llama-cli, llama-server, and Python scripts for model conversion.
Sanitization: No input validation or sanitization of the strings retrieved from the Hugging Face Hub is performed before they are interpolated into executable commands.

huggingface-local-models