local-models

Pass

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches vision model weights from HuggingFace's official ggml-org repository. This is an expected behavior for the vision functionality and uses a trusted source.
  • [COMMAND_EXECUTION]: Executes local llama.cpp binaries (llama-completion, llama-cli, llama-server) to perform inference tasks. Command arguments are properly quoted to prevent shell injection.
  • [DATA_EXFILTRATION]: No network exfiltration detected. The skill is specifically designed for offline and private processing, accessing local model weights on disk.
  • [PROMPT_INJECTION]: The system prompts for presets (anonymization, summarization, etc.) are standard and do not contain attempts to bypass safety filters or reveal system instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 18, 2026, 06:24 PM
Security Audit — agent-trust-hub — local-models