Pre-execution safety check for AI agent tool calls and planned actions. Screens proposed function calls, shell commands,