youtube-summarizer

Fail

Audited by Gen Agent Trust Hub on Apr 4, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads code from a personal GitHub repository that is not a well-known service.
  • Evidence: Clones 'https://github.com/kimtaeyoon83/mcp-server-youtube-transcript.git' during installation as specified in SKILL.md.
  • [REMOTE_CODE_EXECUTION]: Build scripts are executed on untrusted external code.
  • Evidence: The installation process runs 'npm install && npm run build' within the third-party repository.
  • [COMMAND_EXECUTION]: Dynamic execution of downloaded JavaScript code.
  • Evidence: Uses 'node -e' to import and run modules from the './dist/' folder of the cloned dependency.
  • [COMMAND_EXECUTION]: Access to and manipulation of sensitive system directories.
  • Evidence: The skill creates and writes to directories under '/root/clawd/', which typically requires root-level permissions.
  • [DATA_EXFILTRATION]: Transmission of local files to an external messaging service.
  • Evidence: Utilizes 'curl' to send transcript files from '/tmp/' to the Telegram Bot API using stored secrets.
  • [PROMPT_INJECTION]: Processing of untrusted video transcript data creates an attack surface for indirect prompt injection.
  • Evidence: Transcripts are converted to strings and inserted into summarization templates without sanitization or protective boundary markers.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Apr 4, 2026, 09:05 AM