segment-anything-model

Pass

Audited by Gen Agent Trust Hub on Sep 5, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill installs the segment-anything library and related video segmentation tools directly from the Facebook Research GitHub organization.
  • [EXTERNAL_DOWNLOADS]: Fetches model weights (checkpoints) from Meta AI's official public file hosting service (fbaipublicfiles.com).
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted image and video files, as well as text-based prompts for grounded segmentation tasks. While this creates a data ingestion surface, the model inference operations are confined to computer vision tasks using standard machine learning libraries and do not involve executing input as system commands or scripts.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 5, 2026, 12:08 PM