agentic-ecology-camera-traps
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references and downloads pre-trained models from Kaggle (kaggle:google/speciesnet/pyTorch/v4.0.3a/1). These are official resources from a well-known service used for the skill's primary function.
- [DATA_EXFILTRATION]: The technical reference describes a feature for custom image similarity search that fetches images from arbitrary HTTP/S URLs. This involves network read operations to external, non-whitelisted domains.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources (user-provided URLs and images) for processing and indexing.
- Ingestion points: REFERENCE.md specifies that the system should download images from external HTTP/S query URIs.
- Boundary markers: No explicit boundary markers or safety instructions are defined to separate untrusted image content from the agent's processing logic.
- Capability inventory: The skill possesses file write capabilities (creating a Hoplite database and copying model files) and network access (downloading images).
- Sanitization: While the instructions mention converting images to RGB format and performing bounding box detection, there is no explicit sanitization of URLs or image metadata to prevent adversarial exploitation.
Audit Metadata