The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Python subprocess module to call ffmpeg and ffprobe. These tools are used for legitimate audio processing tasks, including file validation, resampling to 16kHz mono, and chunking long audio files.
[EXTERNAL_DOWNLOADS]: The skill references standard, reputable libraries for speech processing and networking, specifically faster-whisper, pyannote.audio, torch, and httpx. These are industry-standard dependencies for the described functionality.
[PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface as it transcribes untrusted audio data for consumption by downstream LLMs.
Ingestion points: Raw audio files in various formats (wav, mp3, etc.) processed in SKILL.md.
Boundary markers: The instructions suggest using structured JSON schemas and SRT formats for handoff, which provide basic data boundaries.
Capability inventory: The environment includes system command execution (ffmpeg) and network capabilities via httpx to deliver transcripts to CMS endpoints.
Sanitization: Includes normalization steps such as whitespace cleanup and filler word removal, although these are intended for quality rather than security filtering.

Voice AI Integration Engineer