awq-quantization
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- EXTERNAL_DOWNLOADS (LOW): The skill requires the installation of autoawq and vllm. These packages are not provided by the pre-approved trusted organizations listed in the security policy. This finding is downgraded to LOW as these dependencies are essential for the primary purpose of the skill.
- PROMPT_INJECTION (LOW): The skill's quantization process ingests external text data, presenting a surface for indirect prompt injection. 1. Ingestion points: calib_data parameter in model.quantize (SKILL.md). 2. Boundary markers: Absent. 3. Capability inventory: File system writing via model.save_quantized and network model loading. 4. Sanitization: Not implemented in provided code examples.
Audit Metadata