TL;DR
A developer has introduced bigarrow, an MIT-licensed macOS command-line tool with skills for Claude Code and Codex that lets AI agents draw arrows, boxes and text over windows. The tool can point users to controls or display copyable text, but its documentation says it does not click, type or capture the screen.
A developer has introduced bigarrow, an MIT-licensed macOS command-line tool that lets AI agents draw arrows, boxes and text over open windows to guide a person through a task. The project, posted as a Show HN submission, is designed for moments when an agent needs a human to handle a permission prompt or another action; according to its documentation, it points but does not click or type.
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Bigarrow can target an interface element by label and app name, or draw at specified screen coordinates. Its commands support visual styles including rings and boxes, with options for color, arrow direction, size and duration. A text sign can accompany the pointer; the project says text enclosed in double curly braces, such as {{482913}}, becomes a copy button.
The project provides skills intended to teach Claude Code and Codex how to use the utility. Installation is offered through Homebrew or by building from source with Swift; the documentation lists macOS 14 or later and Xcode 16 or later for the source build. The developer describes the program as a single Swift binary, with no daemon, menu-bar icon, account or telemetry.
Bigarrow’s documentation says its transparent window can appear above other windows and across displays and Spaces, while mouse clicks pass through and keyboard focus stays where it was. Arrows can disappear after a set time, be removed with a stop command, or end with the agent process that created them. The tool also offers a dry-run mode that reports where it would point, and an option to render an image.
A Visual Handoff to the User
AI agents increasingly perform multi-step work in desktop applications, but some steps require a person to make a decision or supply authorization. A terminal message telling someone to approve a prompt can be easy to miss; an on-screen pointer connects the instruction to the control the agent has found.
The distinction between pointing and acting is central to the project. Bigarrow is presented as a human-guidance tool, not an automation tool: a person remains responsible for approving access, completing a payment, entering a code or handling another sensitive step. That design may also help agents explain software procedures, guide remote users, or make a screen-share instruction less ambiguous.
Those uses are described by the project author rather than established through independent testing. Whether the tool is useful in routine agent workflows will depend on how reliably it identifies the intended control and how well its overlays behave across different applications and display setups.
Products Worth Considering
Bose QuietComfort Wireless Over-Ear Noise-Cancelling Headphones, Sandstone
Apple Watch Series 12 GPS 42mm Smartwatch
LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1, DisplayPort 1.4, Black
As an affiliate, we earn on qualifying purchases.
From Terminal Prompts to Screen Guidance
The author frames bigarrow around a gap in desktop-agent workflows: an agent may locate a button but need the user to press it. Examples in the project documentation include operating-system permission prompts, OAuth consent screens, two-factor authentication, CAPTCHAs, passkeys and payment approval. In these cases, the agent can indicate where the user should act without taking that action itself.
The repository includes examples for macOS System Settings, Keynote, TextEdit, Chrome and Finder, as well as a guide to allowing screen recording. These are examples supplied by the developer; the source does not provide an independent evaluation or adoption figures. The documentation also says drawing requires no macOS permission, while describing screen recording as a separate guide use case.
The project draws a clear boundary around its purpose: it says it is not a screenshot tool and does not capture the screen. Its stated functions are to display a pointer and sign, then remove them according to the chosen duration or stop command. The repository lists an MIT license.
“It only points.”
— Bigarrow project documentation
Performance Across Mac Apps
The source material does not establish how accurately bigarrow locates controls across different applications, window layouts or changing interface labels. It also does not provide independent tests, user numbers or details on how the tool behaves in every multi-display configuration. The demonstrations and reliability statements on the project page are developer-provided claims.
The documentation says that drawing itself requires no macOS permission, but it does not set out a full security review or explain every permission requirement for the agent software used alongside bigarrow. The tool’s stated boundary—no clicking, typing or screen capture—is clear in the project description; independent verification of its implementation is not included in the supplied material.
Testing and Adoption Remain Open
The project is available to install through its Homebrew tap or build from source, so the next practical test is whether developers and Mac users adopt it in real workflows. The Show HN post may draw feedback on usability, target selection and the handoff between an agent and a person, but the supplied source does not report a release schedule, roadmap milestones or planned features.
Readers evaluating it can consult the repository’s installation instructions, commands, examples and license. Further evidence about reliability, compatibility and user uptake would clarify how well the approach works beyond the demonstrations published by its author.
Key Questions
What is bigarrow?
Bigarrow is a macOS command-line utility that draws arrows, boxes and text over windows so an AI agent can direct a user to an interface control.
Can it click buttons or enter information?
No. The project says bigarrow does not click or type. The user performs the action; the tool supplies visual guidance.
Which AI coding tools does it support?
The repository provides skills intended for Claude Code and Codex. It also describes command-line use, including targeting controls by label or coordinates.
Does bigarrow take screenshots?
The project says the tool does not capture the screen. It describes drawing an overlay and also offers a PNG-rendering option for output images.
How can users install it?
The documentation lists installation through the project’s Homebrew tap or a Swift source build. It specifies macOS 14 or later and Xcode 16 or later for building from source.
Source: hn
