The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes osascript to interface with the macOS Calendar app. It follows security best practices by passing user-supplied data (such as event titles and descriptions) as positional arguments to the AppleScript environment. This methodology ensures that user input is never interpreted as executable code.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process natural language requests from users, which represents a potential attack surface. However, it incorporates several layers of defense:
Ingestion points: Natural language requests are first mapped to a structured JSON format by the agent before being passed to the execution script.
Sanitization: A Python validation block within calendar.sh rigorously checks all JSON fields. It enforces type constraints (e.g., integers for offsets), validates ranges (e.g., hours 0-23, minutes 0-59), and sanitizes string inputs by removing newline characters to prevent shell parsing errors.
Capability inventory: The skill's capabilities are restricted to modifying the local user's calendar and writing to a local log file (logs/calendar.log).
Boundary markers: The use of structured JSON input acts as a boundary, ensuring the agent only extracts specific data points required for the calendar operation.

macos-calendar