ffmpeg
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external media files (MP4, GIF, WAV, etc.) and extract metadata using
ffprobe. This constitutes an ingestion surface where malicious content, such as prompt injection strings embedded in media metadata tags (e.g., comments or titles), could be introduced into the agent's context when it reads the tool output. - Ingestion points: User-provided media files (
input.mp4,input.gif, etc.) and concatenation list files (list.txt) inSKILL.md. - Boundary markers: None; the skill does not provide instructions to delimit or treat media metadata as untrusted content.
- Capability inventory: Execution of powerful shell utilities (
ffmpeg,ffprobe) and file system operations (reading/writing media assets) inSKILL.mdandreference.md. - Sanitization: No specific sanitization or validation of filenames or internal media metadata is performed before the agent processes or displays the information.
- [COMMAND_EXECUTION]: The skill relies on shell command execution to perform media transformations. While this is the intended purpose, the provided templates include flags like
-safe 0in concatenation operations, which relaxes FFmpeg's internal security boundaries by allowing the tool to access files via absolute paths or potentially unsafe filenames.
Audit Metadata