The Agent Skills Directory

[COMMAND_EXECUTION]: Extensive use of osascript (AppleScript) across multiple scripts to automate UI interactions, keystrokes, and system events in native macOS applications.
[COMMAND_EXECUTION]: Potential AppleScript injection vulnerability in scripts like test-discord-bot.sh, test-slack-bot.sh, etc., where user-supplied messages are directly interpolated into osascript command strings without escaping.
[DATA_EXFILTRATION]: The skill provides tools to read and write to the system clipboard (pbpaste, agent-browser clipboard read, set the clipboard to) and to extract text or screenshots from communication applications (Discord, Slack, Telegram, WeChat, QQ, Lark).
[REMOTE_CODE_EXECUTION]: Supports arbitrary JavaScript execution within a browser or Electron application context using the agent-browser eval command.
[REMOTE_CODE_EXECUTION]: The record-electron-demo.sh script executes arbitrary shell scripts provided as command-line arguments.
[EXTERNAL_DOWNLOADS]: Instructions include installing the third-party agent-browser CLI via npm, brew, or cargo, and using it to download browser binaries.
[PROMPT_INJECTION]: Vulnerable to indirect prompt injection (Category 8) as it ingests untrusted data from external chat messages and web pages through clipboard reading and DOM snapshots.
Ingestion points: pbpaste, agent-browser snapshot, screencapture (when combined with vision analysis).
Boundary markers: Absent.
Capability inventory: osascript (UI control), agent-browser (JS execution), curl (Network access).
Sanitization: None detected for external content processing.

local-testing