The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes transcription data and YouTube video content, which constitutes an attack surface for indirect prompt injection where instructions embedded in the audio or video could be interpreted by the agent.
Ingestion points: Processes content from audio files (mp3, m4a, wav) via the transcribe script and YouTube transcripts via the mcp__youtube-transcript__get_transcript tool.
Boundary markers: The instructions do not specify the use of delimiters or boundary markers to separate external data from system instructions.
Capability inventory: The skill utilizes Bash (executing ffmpeg and custom scripts), Read, and Write tools to manage the audio workflow.
Sanitization: No explicit sanitization or filtering of transcribed text is mentioned before processing.

audio-pipeline