The Agent Skills Directory

[COMMAND_EXECUTION]: Shell scripts for bot testing (e.g., test-discord-bot.sh, test-slack-bot.sh) are vulnerable to AppleScript injection because user-provided messages and channel names are interpolated directly into osascript commands without proper escaping.
[COMMAND_EXECUTION]: The electron-dev.sh script is vulnerable to environment variable injection via the CDP_PORT variable, which is used to construct file paths and command arguments.
[EXTERNAL_DOWNLOADS]: Recommends installing the agent-browser CLI tool from npm, Homebrew, or Cargo, and utilizes npx to execute the electron-vite development server.
[DATA_EXFILTRATION]: Accesses sensitive system interfaces including the macOS clipboard (via pbpaste and osascript) and screen capture (via screencapture) to monitor application state.
[PROMPT_INJECTION]: Susceptible to indirect prompt injection due to the ingestion of untrusted data from external web pages and third-party chat applications.
Ingestion points: Ingests data from the browser DOM via agent-browser snapshot and from application windows via clipboard capture.
Boundary markers: None identified; external content is processed without delimiters or warnings to ignore embedded instructions.
Capability inventory: The skill can execute arbitrary JavaScript in browsers, manipulate GUI elements of native apps, and perform file system operations.
Sanitization: No validation or sanitization is performed on content retrieved from external sources before it is used to drive agent actions.

local-testing