media-transcript-search

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHPROMPT_INJECTIONNO_CODE
Full Analysis
  • [PROMPT_INJECTION] (HIGH): Vulnerable to Indirect Prompt Injection via external data ingestion.
  • Ingestion points: Fetches video transcripts, playlist metadata, and video titles from YouTube (attacker-controlled external source).
  • Boundary markers: Absent. The description does not mention any delimiters or instructions to the agent to treat transcript content as data rather than instructions.
  • Capability inventory: The skill is designed to write files (Export to .txt/.srt) and process text for 'finding specific topics,' which likely influences downstream agent decisions.
  • Sanitization: None documented. Untrusted content is likely passed directly into the agent's reasoning loop.
  • [NO_CODE] (MEDIUM): The primary execution logic resides in get_transcript.py, which is referenced but not provided in the analyzed package. This prevents a full security audit of the command execution and network behavior.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 01:09 PM