yolo-inference
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external media (images, videos, RTSP/RTMP streams) and performs semantic searches over local folders via VisualAISearch. Maliciously crafted filenames or metadata in these sources could potentially influence agent behavior. 1. Ingestion points: Media sources in SKILL.md (URLs, file paths, directories) and folder search in solutions.md (VisualAISearch). 2. Boundary markers: None explicitly specified for handling untrusted file metadata. 3. Capability inventory: File writing (save=True, ObjectCropper), network requests (requests.post, SecurityAlarm), and system interactions via the yolo CLI. 4. Sanitization: Relies on internal library processing; no explicit sanitization of external metadata is documented.
- [SAFE]: The skill correctly uses placeholders (YOUR_API_KEY, YOUR_DEPLOYMENT_URL) for deployment configuration, encouraging secure credential management. It utilizes official vendor resources and well-known libraries for its primary functionality.
Audit Metadata