gitframes-effects
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads machine learning models (e.g., RTMDet, RTMO, Selfie Segmenter) for vision-based effects to a local cache directory (~/.cache/gitframes/models). The documentation includes instructions for pre-downloading and verifying the integrity of these models using SHA-256 hashes.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted media assets and effect configuration files (such as .cube LUT files provided via lutUrl).\n
- Ingestion points: Source paths for videos/images in Layer and Media classes, and external URLs for Color Lookup Tables (LUTs) in the ApplyLUT effect.\n
- Boundary markers: The instructions do not specify explicit delimiters or safety instructions to prevent the agent from following directions embedded in processed media or metadata.\n
- Capability inventory: The skill utilizes WebGPU shader execution, vision model inference, network requests for external assets, and local file system access for model caching.\n
- Sanitization: Integrity verification (SHA-256) is mentioned for model downloads, but no specific sanitization is noted for external media content or LUT files.
Audit Metadata