The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's script scripts/download.py includes a mechanism to automatically install or update the yt-dlp package using pip if it is not found on the system. This involves downloading code from the Python Package Index (PyPI).
[COMMAND_EXECUTION]: The script uses the subprocess.run function to execute yt-dlp commands and the pip package manager. The calls are performed using list-based arguments rather than shell strings, which mitigates the risk of command injection.
[CREDENTIALS_UNSAFE]: To handle authentication-restricted content and bot blocks, the script searches for and reads browser cookies from specific locations, including the user's Downloads folder and local browser profile directories (Chrome, Firefox, Edge, etc.). This access to sensitive session data is a core part of the skill's stated recovery strategy.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of video URLs provided by the user. While the script uses safe execution patterns, the output from the external yt-dlp tool is returned to the agent's context, which could theoretically contain malicious text intended to influence the agent's behavior.
Ingestion points: User-provided URLs are passed to the download.py script.
Boundary markers: None explicitly implemented within the script output, though the platform typically handles tool output boundaries.
Capability inventory: Subprocess execution for package installation and video downloading via yt-dlp.
Sanitization: Use of argument lists in subprocess.run prevents shell injection attacks via malicious URLs.

video-download