The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several subprocesses to perform its tasks:
Invokes yt-dlp to download specific sections of YouTube videos.
Uses ffmpeg for converting media files to MP4 and burning SRT subtitles into the video stream.
Employs shell commands (grep, cut, tr) to extract the Deepgram API key from a local .env file.
Executes a bundled Python script json-to-srt.py to transform API responses into subtitle files.
[EXTERNAL_DOWNLOADS]: Communicates with external services to fulfill its primary functions:
Downloads video and audio streams from YouTube.
Sends audio data to api.deepgram.com (a well-known service) for transcription.
[PROMPT_INJECTION]: An indirect prompt injection surface is present during the subtitle cleanup workflow (Step 5):
Ingestion points: The agent reads the content of clip.srt, which is generated from external YouTube audio transcribed by a third-party API.
Boundary markers: There are no explicit delimiters or safety instructions provided to the agent to treat the subtitle content as untrusted data or to ignore potential instructions within it.
Capability inventory: The skill has the ability to execute system commands (ffmpeg, yt-dlp, python3) and perform file system writes.
Sanitization: The transcription data is not sanitized or validated for malicious prompts before being processed by the agent.

caption-clip