Software agents have long struggled with the final mile of human-in-the-loop workflows: getting a user to click a specific button or approve a dialog. A new open-source tool called bigarrow, released by developer franzenzenhofer, solves this by allowing AI agents to paint large, click-through arrows, boxes, and text directly onto the macOS screen. The project, which appeared on Hacker News on October 8, 2026, is a small Swift binary that requires no daemon, no menu-bar icon, and notably, no AI logic of its own. It simply provides a visual interface for agents to communicate intent to humans.

Solving the 'Please Click Allow' Problem

The core utility of bigarrow is addressing scenarios where an agent identifies a necessary action but cannot or should not perform it. This includes macOS permission prompts, OAuth consent screens, 2FA inputs, and CAPTCHAs. By using commands like bigarrow point --element "Allow" --app "System Settings", an agent can highlight the exact UI element needing attention. The tool supports targeting by accessibility labels, window titles, or absolute coordinates. It operates entirely in the background, ensuring it never steals keyboard focus or interferes with the user’s current workflow, a critical feature for non-disruptive agent interaction.

Zero-Friction Architecture and Installation

bigarrow is designed for minimal footprint and maximum reliability. It installs via Homebrew with brew install franzenzenhofer/tap/bigarrow and includes a skill installer for Claude Code and Codex agents. The architecture is deliberately simple: it is a single binary that draws overlays using Core Animation. It does not need Screen Recording or Accessibility permissions to draw, though accessing specific UI elements relies on the host terminal app’s existing Accessibility grants. The tool is MIT licensed and explicitly states, 'It is the least intelligent part of your AI stack, and proud of it,' emphasizing its role as a dumb but effective pointer.

Advanced Targeting and Visual Customization

Beyond simple arrows, bigarrow offers extensive customization for visual clarity and urgency. Agents can choose from straight, bend, or zigzag shapes, and styles including arrows, rings, and boxes. The tool supports multiple simultaneous arrows, each with its own label and color, managed via --duration or explicit start/stop commands. It handles complex macOS environments, including Stage Manager, multiple displays, and full-screen apps, with automatic cleanup if a display is disconnected. The --say flag even enables text-to-speech, allowing the agent to verbally call the user back to their desk when they are distracted.

Key Takeaways

  • bigarrow is a macOS CLI tool that lets AI agents draw click-through arrows and labels on the screen.
  • It requires no daemon, telemetry, or internal AI, functioning purely as a visual overlay utility.
  • Agents can target elements by accessibility label, window title, or coordinates, with support for Claude Code and Codex skills.
  • The tool never steals focus, ensuring user workflow is uninterrupted while highlighting necessary actions.
  • It handles complex macOS scenarios like Stage Manager, multiple displays, and full-screen apps gracefully.

The Bottom Line

bigarrow is a pragmatic hack that bridges the gap between autonomous agent capabilities and human oversight requirements. By standardizing how agents point at UI elements, it eliminates the friction of vague terminal instructions like 'click the button on the left.' For the OpenClaw community and agent developers, this is a small but essential utility that makes human-in-the-loop automation actually usable.