media
Fail
Audited by Gen Agent Trust Hub on Jun 19, 2026
Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The script 'scripts/auth.py' reads sensitive Google authentication cookies (SID and HSID) from the file path '~/.notebooklm/storage_state.json' to authenticate with Google services.\n- [COMMAND_EXECUTION]: The script 'scripts/extract.py' utilizes 'subprocess.run' to execute the 'yt-dlp' module for video metadata extraction and media downloading.\n- [EXTERNAL_DOWNLOADS]: Fetches media files and research artifacts from well-known Google and YouTube domains.\n- [PROMPT_INJECTION]: The skill processes untrusted data from external sources, presenting an attack surface for indirect prompt injection.\n
- Ingestion points: 'scripts/extract.py' (YouTube transcripts and comments) and 'scripts/generate.py' (PDF and URL sources).\n
- Boundary markers: None identified in the scripts when processing or outputting extracted text content.\n
- Capability inventory: 'subprocess.run' execution, file system writes, and network operations via specialized libraries.\n
- Sanitization: Video IDs are sanitized via regular expressions, but the textual content of transcripts and comments is not filtered or sanitized before being returned to the agent context.
Recommendations
- AI detected serious security threats
Audit Metadata