anycap-gemini-omni-video-edit
Pass
Audited by Gen Agent Trust Hub on Jul 29, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill communicates with AnyCap's official API via the
anycapCLI and references established Google Gemini documentation for model behavior and schemas. These interactions are limited to vendor-provided services and trusted cloud platforms. - [COMMAND_EXECUTION]: The instructions utilize the
anycapCLI for executing video edits and theffmpeg/ffprobeutilities for local quality assurance. These operations are scoped strictly to the skill's primary purpose of video manipulation and metadata verification. - [PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection by ingesting external video and image data.
- Ingestion points: Processes source videos and reference images from local paths or remote URLs.
- Boundary markers: Employs explicit preservation instructions (e.g., "Keep everything else the same") to focus the model and reduce susceptibility to content-based overrides.
- Capability inventory: Limited to media editing commands and quality assurance scripts with no high-risk shell capabilities exposed to untrusted inputs.
- Sanitization: Relies on the host AI platform's inherent safety filters and specific instructional templates to govern how processed media influences the agent's behavior.
Audit Metadata