wjs-creating-video-book

Pass

Audited by Gen Agent Trust Hub on Aug 21, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The skill accesses sensitive OAuth tokens and credentials at ~/.config/youtube/token.json to enable video uploads and comment management on YouTube.
  • [COMMAND_EXECUTION]: The skill executes various system commands including ffmpeg, ffprobe, pandoc, and brew to handle media processing and document conversion.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted book content which could contain instructions to manipulate agent behavior during text analysis and image prompt generation.
  • Ingestion points: Processes chapter.md derived from external PDF/EPUB files in scripts/post-transcript-comments.py and scripts/render-chapter.py.
  • Boundary markers: Absent; there are no clear delimiters separating untrusted text from instructions.
  • Capability inventory: Includes subprocess calls for media tools (ffmpeg, ffprobe) and network requests for YouTube API interactions.
  • Sanitization: Absent; the skill does not sanitize or escape chapter content before processing it for video metadata or image generation.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 21, 2026, 08:57 AM
Security Audit — agent-trust-hub — wjs-creating-video-book