scan-assets
Warn
Audited by Gen Agent Trust Hub on Jul 30, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill generates shell commands for ffmpeg and python3 using file names and stems retrieved from Google Drive. These inputs are not sanitized before being interpolated into shell strings, which allows a maliciously named file to execute arbitrary commands on the system.\n- [EXTERNAL_DOWNLOADS]: The skill depends on the openai-whisper Python package and the ffmpeg system utility being pre-installed in the execution environment.\n- [DATA_EXFILTRATION]: The skill automatically changes file permissions to anyone-with-link-reader to facilitate human previews. This can lead to sensitive company assets being accessible to unauthorized parties if the links are shared or discovered.\n- [PROMPT_INJECTION]: The skill processes untrusted filenames and transcripts from Google Drive which are used as inputs for system tools.\n
- Ingestion points: Google Drive file names and audio transcripts extracted via Whisper.\n
- Boundary markers: A blockquote disclaimer is present in SKILL.md instructing the agent to treat content as untrusted.\n
- Capability inventory: The skill uses google-drive MCP tools for file manipulation and shell execution for video processing.\n
- Sanitization: The skill applies kebab-case formatting to output filenames but lacks sanitization for the input filenames used in the shell commands.
Audit Metadata