transcribe-mp3

Fail

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill constructs shell commands for ffmpeg and whisper-cpp by interpolating a video_id variable derived from filenames. A maliciously named file could lead to arbitrary code execution if the variable contains shell metacharacters. Evidence: ffmpeg -i "output/audio-{video_id}.mp3" and whisper-cpp -f /tmp/whisper_{video_id}.wav.
  • [PROMPT_INJECTION]: The skill facilitates indirect prompt injection by transcribing external audio data into text without proper encapsulation or sanitization. Ingestion points: Reads transcription output from /tmp/whisper_{video_id}.wav.txt. Boundary markers: Absent. Capability inventory: Shell command execution and file system writes. Sanitization: Absent.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jun 18, 2026, 10:42 PM
Security Audit — agent-trust-hub — transcribe-mp3