A new open-source command-line utility called bigarrow has been released that enables AI agents to draw visible arrows and text labels on macOS screen elements, helping automated systems navigate graphical interfaces without requiring direct user interaction.

Developed by Franz Zappa and distributed under an MIT license, bigarrow functions as both a standalone tool and a skill module for AI development platforms Claude Code and Codex. The software overlays directional indicators and contextual instructions on top of application windows, dialogs, and interface components across all displays and virtual workspaces.

The tool operates by drawing transparent overlays that remain click-throughable, allowing underlying interface elements to remain functional while the visual guidance is active. Arrows automatically disappear after a set duration or when manually terminated, eliminating the need for manual cleanup by the controlling agent.

Bigarrow's architecture emphasizes minimal system impact, featuring a single compiled Swift binary with no background processes, menu-bar icons, user accounts, or telemetry systems. The project documentation confirms no embedded AI components exist within the core tool.

Use cases demonstrated in the project include guiding users through system permission workflows, such as directing attention to specific buttons in macOS System Settings or TextEdit's print dialog. The tool has been applied to highlight interface elements in diverse applications including Chrome, Keynote, Finder, and Terminal.

Technical implementation details show bigarrow supports multiple targeting methods: coordinate-based positioning (--at X,Y), element identification by label (--element "Allow"), window/tab targeting (--app "Google Chrome:Sourdough"), and accessibility-based detection. Visual properties like --shape (zigzag, spiral), --color, --size, and --style can be customized for different urgency levels or contextual requirements.

The project includes comprehensive testing protocols, with 104 automated tests verifying geometric accuracy, permission handling, and cross-display functionality. Performance benchmarks indicate minimal resource consumption, with less than 1.5% CPU usage during operation.

Developers have noted that bigarrow addresses a specific interaction gap where AI agents can identify interface elements but lack physical means to interact with them. By providing a visual pointing mechanism, the tool allows agents to communicate required user actions without assuming control over system permissions or interface state.

The release has garnered significant attention in developer communities, accumulating nearly 400 GitHub stars within two days from contributors representing major technology organizations and academic institutions. The project's documentation includes usage examples for common scenarios such as permission grant flows, copy-paste operations, and multi-step instructional sequences.

Bigarrow represents a niche but potentially valuable tool in the growing ecosystem of AI-assisted system navigation, particularly for applications requiring human confirmation in otherwise automated workflows. Its design focuses on maintaining clear separation between automated guidance and direct system modification.