ad-video-gen
Pass
Audited by Gen Agent Trust Hub on Mar 25, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches generated video assets from Volcengine's official API endpoint (ark.cn-beijing.volces.com). This is a well-known cloud service and the behavior is expected for the skill's primary function.\n- [COMMAND_EXECUTION]: Utilizes
ffmpegto overlay subtitles onto the generated video. The execution uses a structured list of arguments without a shell environment, which prevents typical command injection vulnerabilities.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by interpolating external JSON-based ad copy into a prompt for the remote video model.\n - Ingestion points: Reads advertisement variants from the local file path specified by the
--copy-jsonargument.\n - Boundary markers: No delimiters or safety warnings are used to separate user-provided copy from the internal motion prompts.\n
- Capability inventory: The script performs authenticated API requests to Volcengine and executes local
ffmpegcommands.\n - Sanitization: The input text is not sanitized or validated before being sent to the remote model.
Audit Metadata