video-edit
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill uses ffprobe to extract metadata from video files into a JSON format that the agent then processes. This ingestion of untrusted data from media files (such as titles, comments, or other metadata fields) could be exploited to inject instructions that influence the agent's behavior.
- Ingestion points: Metadata extraction using ffprobe commands in SKILL.md and references/operations.md.
- Boundary markers: The instructions lack delimiters or explicit warnings to isolate untrusted file metadata from the agent's core instructions.
- Capability inventory: The skill enables powerful file manipulation and shell command execution capabilities through the ffmpeg suite.
- Sanitization: There is no instruction provided for the agent to sanitize or validate the metadata ingested from external files.
Audit Metadata