The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation provides commands to download and execute shell and PowerShell scripts directly from a remote GitHub repository belonging to the author, which is not a whitelisted trusted source.
[PERSISTENCE]: The installation process includes instructions for registering the bridge daemon as a permanent system-level service using systemd, launchd, or Windows Scheduled Tasks, allowing for continued execution across sessions.
[COMMAND_EXECUTION]: The bridge daemon utilizes subprocess spawning to run external binaries such as 'yt-dlp' and 'ffmpeg' for media processing and invokes system-level file management utilities.
[OBFUSCATION]: The skill includes minified third-party code in 'scripts/vendor/turndown.bundle.mjs', which complicates security auditing and could be used to hide malicious functionality.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from the web and documents into the agent context, creating a surface for potential prompt injection attacks through embedded instructions.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to external domains, including arXiv and various search engines, to fetch document data and search results.

larksnap-fetch