The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill suggests installing an alternative transcription tool directly from an unverified GitHub repository using pip install git+https://github.com/m1guelpf/yt-whisper.git. This allows for the execution of code from a personal repository that has not undergone the vetting associated with official package registries or trusted organizations.
[COMMAND_EXECUTION]: The skill's instructions include running various CLI tools with system access and specifically recommend sudo apt install ffmpeg. The use of sudo to acquire administrative privileges for installation increases the potential impact of any command execution.
[EXTERNAL_DOWNLOADS]: The skill uses yt-dlp to download audio from YouTube URLs. While YouTube is a well-known service, downloading and processing external binary data is a common vector for remote exploits.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it converts untrusted audio from external sources into text transcripts. Malicious instructions spoken within a video could potentially hijack the agent's behavior if it processes the resulting transcript without isolation.
Ingestion points: External YouTube audio streams processed via yt-dlp and transcribed by whisper (SKILL.md).
Boundary markers: Absent. No instructions are provided to the agent to treat the transcript as untrusted or to wrap it in delimiters.
Capability inventory: The skill has the capability to execute system commands and network operations through the yt-dlp and whisper tools.
Sanitization: Absent. Transcripts are generated and presented without any filtering or validation for instruction-like content.

youtube-transcription