remotion-best-practices
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external media and scripts (YouTube frames, screenshots, voiceover text) to generate video compositions. This represents a vulnerability surface where instructions hidden in the reference data could influence agent behavior. * Ingestion points: YouTube video extraction and screenshot processing via ffmpeg and Read tools. * Boundary markers: No explicit boundaries or instructions to disregard embedded data are provided. * Capability inventory: Includes shell command execution (npx, ffmpeg, node) and file system writes. * Sanitization: No validation or sanitization of external content is specified.
- [DATA_EXFILTRATION]: The skill instructions contain hardcoded local file paths revealing the local system username of the author. * Evidence: The path /Users/antonabyzov/Movies/.../Source.mov is used in a frame extraction example.
Audit Metadata