The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands to perform its tasks. These include curl for fetching HTML, yt-dlp for subtitle and media extraction, ffmpeg for audio conversion, and uv run for executing a local transcription script. These commands use variable interpolation for URLs and file paths, which requires proper sanitization to prevent command injection.
[EXTERNAL_DOWNLOADS]: The skill is designed to download content from arbitrary external URLs provided by the user. While this is the core functionality, fetching untrusted data from the internet is a prerequisite for other attack vectors.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection (Category 8). It ingests untrusted data from external URLs and processes it into text transcripts which are then returned to the agent's context.
Ingestion points: HTML source code from curl and audio/video streams via yt-dlp (file: SKILL.md).
Boundary markers: No specific delimiters or safety instructions are used when passing the transcribed text back to the agent.
Capability inventory: The agent has access to shell execution (curl, yt-dlp, ffmpeg, uv, pgrep, lsof) and file system operations within the work/ and outputs/ directories.
Sanitization: The skill does not perform sanitization on the external content or the resulting transcriptions before they are processed by the agent.

url-transcript