Impact Preview

Think of this as "terraform plan" but for AI agent actions. It intercepts operations like file writes, deletes, and shell commands, then shows you a diff preview and risk assessment before anything executes. You get four MCP tools in Claude: preview_file_write, preview_file_delete, preview_shell_command, and check_path_risk. Each one flags dangerous operations (production databases, system files) and blocks execution until you approve. There's also a full REST API and Python SDK if you want to wrap your own agent code with @require_approval decorators. The real win is seeing what will actually change, not just what the agent claims it wants to do, with an event sourced audit trail of every action.