The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses yt-dlp to download subtitles. It implements strict validation of the YouTube URL using a regular expression that limits video IDs to alphanumeric characters and underscores, effectively preventing shell injection via the URL parameter. The script scripts/transcribe_youtube.py builds commands as argument lists for subprocess.run, avoiding the risks associated with shell interpolation.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-generated subtitle text from YouTube, which acts as a vector for indirect prompt injection if the content contains malicious instructions designed to influence the agent's behavior.
Ingestion points: The run function in scripts/transcribe_youtube.py reads external WebVTT files downloaded from YouTube.
Boundary markers: The generated transcript files do not include delimiters or instructions to help downstream agents distinguish between transcript content and potential commands.
Capability inventory: The skill possesses the ability to execute shell commands (yt-dlp) and write files to the local file system.
Sanitization: The script removes HTML tags and formatting metadata but does not perform semantic filtering or sanitization of the transcript text for prompt injection keywords.
[CREDENTIALS_UNSAFE]: The skill supports accessing browser cookies via yt-dlp to bypass YouTube authentication barriers for restricted videos. The SKILL.md instructions mitigate this risk by requiring the agent to obtain user approval for specific browser profiles and strictly forbidding the exfiltration or persistent storage of cookie data.

youtube-transcribe-skill