The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implementation patterns include the use of subprocess.run to execute quantization binaries and conversion scripts. This functionality is essential for model optimization and is accompanied by examples of secure execution practices, such as environment isolation and path validation.
[INDIRECT_PROMPT_INJECTION]: The skill provides templates that ingest external model data and prompts, which represents an attack surface for indirect prompt injection.
Ingestion points: External model files (.gguf, .bin) and test prompt strings are processed by the quantizer and benchmark tools as shown in SKILL.md and references/advanced-patterns.md.
Boundary markers: Not explicitly defined in the provided code snippets for LLM-based processing.
Capability inventory: Utilizes subprocess.run to execute system binaries (quantize) and scripts (convert_hf_to_gguf.py) and uses the llama-cpp-python library for inference.
Sanitization: The skill mitigates risks by including mandatory integrity verification via a ModelVerifier class (SHA256) and rigorous path validation via validate_model_path to prevent directory traversal attacks.

model-quantization