agentic-ecology-camera-traps
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructions and technical reference (references/technical_reference.md) describe a feature for 'Custom Image Similarity Search' that downloads images from arbitrary user-provided HTTP/S query URIs.
- Evidence: 'The backend server... must implement on-the-fly embedding extraction for custom HTTP/S query URIs by downloading the image...' (SKILL.md).
- Evidence: 'Download the image (for HTTP/S URLs) using urllib.request or load it from disk...' (references/technical_reference.md).
- Security Concern: Providing a capability to fetch remote content based on unvalidated user input creates a Server-Side Request Forgery (SSRF) risk, potentially allowing an attacker to probe internal networks or access sensitive metadata services.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources (remote URLs and local image directories) and possesses capabilities that could be abused if the ingested data is malicious.
- Ingestion points: Target camera trap images directory, custom HTTP/S query URIs (SKILL.md, references/technical_reference.md).
- Boundary markers: None present. The instructions do not specify the use of delimiters or warnings for processed content.
- Capability inventory: Network operations via urllib.request, file system writes via perch-hoplite database operations, and model inference via speciesnet (references/technical_reference.md).
- Sanitization: None present. The technical reference does not include validation or sanitization steps for the URLs or the downloaded image data.
Audit Metadata