VLM
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process external and potentially untrusted data such as images, videos, and files. This introduces an inherent surface for indirect prompt injection where instructions could be embedded in visual content or document metadata.
- Ingestion points: The skill accepts external image/video/file URLs and local file paths (VLM/scripts/vlm.ts, SKILL.md).
- Boundary markers: The documentation provides prompt engineering advice, though the code examples do not explicitly use strict delimiters for user-provided text.
- Capability inventory: The skill uses the file system to read local images (fs.readFileSync) and performs network operations via the SDK to communicate with AI models.
- Sanitization: The skill includes a dedicated 'Security Considerations' section in SKILL.md that explicitly instructs developers to validate URLs and sanitize user-provided data, which is a positive security practice.
Audit Metadata