The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes the whisperx library, which downloads pre-trained machine learning models for transcription and speaker diarization from HuggingFace. This is an expected behavior for the skill's primary purpose.
[COMMAND_EXECUTION]: The scripts/transcribe.py script executes the ffmpeg utility via subprocess.run to extract audio tracks from video files. The implementation uses a list of arguments, which is a secure method that prevents shell-based command injection.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted audio or video data and outputs the transcription as text, creating a vulnerability surface for indirect prompt injection.
Ingestion points: The input argument in scripts/transcribe.py accepts external media files for processing.
Boundary markers: The output transcript does not use specific delimiters or warnings to distinguish between the transcribed content and instructions.
Capability inventory: The skill possesses file-writing capabilities and command execution via ffmpeg in scripts/transcribe.py.
Sanitization: Transcribed text is saved directly to output files without validation or sanitization, meaning any spoken instructions in the source media could be interpreted by an agent reading the resulting file.

video-to-text