The Agent Skills Directory

[COMMAND_EXECUTION]: The skill configures VSCode hooks in settings.json to automatically execute local scripts (notify-wsl.sh, notify-mac.sh, notify-windows.ps1) when tasks complete or fail.
[COMMAND_EXECUTION]: The configuration provided in references/hooks-config.md utilizes the -ExecutionPolicy Bypass flag for PowerShell. This bypasses system-level restrictions designed to prevent the execution of untrusted scripts.
[PROMPT_INJECTION]: The bundled notification scripts are susceptible to indirect prompt injection through shell command injection vulnerabilities.
Ingestion points: The scripts accept arguments for the notification title and message, which are intended to be populated by the agent's context during hook execution.
Boundary markers: No boundary markers or delimiters are used to isolate untrusted content in the hook configuration.
Capability inventory: The scripts execute system-level utilities including osascript on macOS and powershell.exe on Windows/WSL to display notifications.
Sanitization: The scripts lack input sanitization. In scripts/notify-mac.sh, the message variable is interpolated directly into an osascript command string inside double quotes. Similarly, scripts/notify-wsl.sh interpolates variables into a PowerShell command string inside single quotes. An attacker could potentially execute arbitrary commands by including escape characters (like quotes or backticks) in the notification text.

vscode-claude-notify