awq-quantization

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill provides instructions for processing external text data through Large Language Models during inference and quantization.
  • Ingestion points: Text prompts passed to model.generate in SKILL.md and calibration samples passed to model.quantize in references/advanced-usage.md.
  • Boundary markers: Not present in the provided code snippets.
  • Capability inventory: The skill focuses on model optimization and inference; it does not demonstrate access to shell execution, file-system writes, or network operations based on model output.
  • Sanitization: Not applicable for the provided library usage examples.
  • [EXTERNAL_DOWNLOADS]: The skill references standard machine learning libraries and model repositories.
  • Downloads autoawq, transformers, and torch from the Python Package Index (PyPI).
  • Fetches model weights and configurations from well-known repositories on HuggingFace (e.g., TheBloke, mistralai).
  • These actions are fundamental to the skill's purpose and utilize trusted, well-known services.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 07:07 PM
Security Audit — agent-trust-hub — awq-quantization