The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses child_process.spawn and execFile to execute local tools including yt-dlp for video downloading and system file managers (Finder, Explorer) to reveal downloaded files. These calls incorporate argument whitelisting and input sanitization to prevent command injection.
[COMMAND_EXECUTION]: The skill performs extensive file system operations, including creating directories and writing files to user-specified or default local paths. The zip extraction utility includes logic to prevent directory traversal attacks.
[REMOTE_CODE_EXECUTION]: The setup.md file provides installation instructions that use curl | bash and PowerShell iex to download and execute scripts from the author's GitHub repository. While this is a common distribution method for CLI tools, it involves executing remote code.
[EXTERNAL_DOWNLOADS]: The skill fetches content from external sources including arxiv.org, Feishu/Lark domains, and various web pages (via a browser extension or direct HTTP requests). It also downloads video content using yt-dlp from various video hosting platforms.
[DATA_EXFILTRATION]: The skill accesses sensitive browser session data, such as cookies and headers, via the browser extension bridge to enable authenticated downloads and document editing. This data is handled locally between the CLI, the daemon, and the extension.
[PROMPT_INJECTION]: The skill processes untrusted data from the web (converting HTML to Markdown). It lacks explicit boundary markers or sanitization for potential prompt injection strings within the downloaded content, creating a surface for indirect prompt injection when the agent processes the resulting Markdown files.

larksnap-fetch