The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to execute yt-dlp --cookies-from-browser chrome. This command attempts to extract session cookies from the user's local web browser to facilitate access to restricted videos. This behavior exposes highly sensitive authentication tokens and private session data to the agent's execution environment.
[COMMAND_EXECUTION]: The skill relies on executing various shell commands (yt-dlp, ffmpeg, faster-whisper) that handle external inputs like URLs. This poses a risk if the underlying tools have vulnerabilities or if the inputs are not properly sanitized before being passed to the shell.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text data extracted from external videos (subtitles and transcripts) through an LLM for correction and summarization. This provides an attack surface where malicious instructions hidden in the video's audio or text could potentially manipulate the agent's behavior.
Ingestion points: Audio and text content extracted from external video URLs (SKILL.md).
Boundary markers: The skill lacks explicit delimiters or instructions to the agent to ignore embedded commands within the extracted transcripts.
Capability inventory: The skill possesses broad capabilities including writing files to the user's Desktop and executing multiple command-line utilities.
Sanitization: There is no mention of sanitizing or escaping the content extracted from videos before it is processed by the LLM for correction.

video-extract