The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses the yt-dlp utility in scripts/download-audio.mjs to download media from user-provided URLs. This is a standard and expected function for media processing skills.\n- [COMMAND_EXECUTION]: The skill executes external binaries (yt-dlp and ffmpeg) via node:child_process and Python's subprocess module. Both implementations in scripts/download-audio.mjs and scripts/transcribe.py use argument arrays rather than shell strings, which is the recommended practice to prevent command injection vulnerabilities.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted media content to generate transcripts. If the transcribed text is subsequently used by an agent, there is a potential surface for indirect prompt injection if the media contains spoken instructions intended to manipulate the agent.\n
Ingestion points: Media files or URLs processed by yt-dlp and Whisper in the provided scripts.\n
Boundary markers: The output is structured in JSON format, which provides a natural delimiter, though the prompt does not explicitly instruct the agent to ignore instructions within the transcript text.\n
Capability inventory: The skill possesses file writing and subprocess execution capabilities needed for media transformation.\n
Sanitization: The skill relies on the Whisper model for transcription and does not perform secondary text sanitization on the resulting transcript.

transcript-extractor