The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation and check scripts suggest installing the Bun runtime using a piped shell command from bun.sh. This is a well-known and trusted service within the developer community.
[EXTERNAL_DOWNLOADS]: The md-to-html.ts script programmatically downloads images from remote URLs provided within user-supplied Markdown files using the Node.js https module to prepare content for article publishing.
[COMMAND_EXECUTION]: Multiple scripts utilize spawn and spawnSync to interact with system-level utilities for automation purposes:
Process Management: Uses pkill and pgrep to manage existing Chrome/Chromium instances to ensure the automation port is available.
Keyboard Automation: Executes osascript (macOS), powershell.exe (Windows), and xdotool or ydotool (Linux) to simulate Cmd+V or Ctrl+V keystrokes for pasting content into the browser.
Compiler Access: Uses the swift compiler on macOS to dynamically execute small scripts for clipboard image handling via AppKit.
Clipboard Access: Calls wl-copy and xclip on Linux to manage clipboard data.
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes untrusted Markdown data from the user which is then converted to HTML and pasted into a browser context.
Ingestion points: Markdown files passed as arguments to x-article.ts and md-to-html.ts.
Boundary markers: No explicit delimiters or instructions are used to separate untrusted content from the agent's logic during interpolation.
Capability inventory: The skill can read local files, download images from the web, and interact with the system clipboard and keyboard.
Sanitization: Employs the marked library for parsing and includes escapeHtml functions, though these are primarily for formatting rather than security-focused sanitization.

canghe-post-to-x