TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Bigarrow is a newly presented, MIT-licensed macOS command-line tool with skills for Claude Code and Codex. It lets an agent point out interface elements with an arrow and label, but does not click, type or capture the screen. The project’s claims and examples come from its GitHub page; independent testing was not provided.
A developer has introduced bigarrow, an open-source macOS command-line tool that lets AI agents draw temporary arrows, boxes and text over application windows to show a person where to act. The project is designed for moments when an agent reaches a step requiring human approval or input; its author says the tool points but never clicks, types or captures the screen.
The GitHub project describes bigarrow as a single Swift binary that can place a transparent overlay above windows across displays and Spaces. Users can identify a target by its accessibility label and app, or provide screen coordinates. Signs can carry instructions, and the project says a value wrapped in {{double braces}} becomes a copy button. A spoken prompt is also available through the --say option.
The utility has three main modes: point for a temporary annotation, start for an overlay that remains until stopped or timed out, and stop to remove it. The project says a point disappears automatically and that a running arrow can be tied to the agent process. It lists installation through Homebrew or a release build from source, and says its integrations teach Claude Code and Codex how to use the commands.
Examples on the project page show arrows aimed at controls in System Settings, Keynote, TextEdit, Chrome and Finder. They include guidance for granting a permission, selecting a browser tab and saving a PDF. The author presents these as staged demonstrations on a Mac, not as independent verification of reliability across different systems or app versions.
A Visual Handoff to the User
Agent software can complete many computer tasks, but some steps call for a person to make a decision: approving a permission request, entering a two-factor code, completing a CAPTCHA or confirming a payment. A terminal message may be easy to miss when the relevant prompt is open elsewhere. Bigarrow’s proposed role is a visual handoff: direct attention to the control while leaving the decision and action with the user.
That boundary is also the project’s main distinction from automation that operates the interface. The author says bigarrow does not click, type or take screenshots; it only draws an overlay. If that behavior holds in use, the tool could help an agent request approval without taking over a sensitive action. It could also make instructions easier to follow for people learning an app or helping someone remotely. These are plausible uses described by the project, not measured outcomes or evidence of broader adoption.
Products Worth Considering
LG UltraGear 25G590B-B 25-Inch Full HD 1000Hz Gaming Monitor
Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
DJI Mini 4 Pro Fly More Combo Drone with DJI RC 2
As an affiliate, we earn on qualifying purchases.
From Terminal Prompts to On-Screen Guidance
The project frames bigarrow around a gap in agent workflows: an agent may locate a control but still need a person to handle a step it cannot or should not perform. Its examples span permission dialogs, sign-in codes and everyday interface choices, as well as step-by-step tutorials and demonstrations.
The project states that drawing requires no macOS permission and describes the binary as having no daemon, menu-bar icon, account or telemetry. It lists macOS 14 or later and Xcode 16 or later for building from source. Those are statements on the project page; the available material does not include an independent security review or compatibility testing.
“It can point. “Your turn.””
— Bigarrow project description
Independent Testing Is Not Provided
The source material is the project’s own GitHub page and does not establish how reliably bigarrow identifies controls across macOS versions, displays, Spaces or third-party apps. It also does not provide independent testing, adoption figures or performance benchmarks. The author’s claims about permissions, telemetry and the absence of clicking, typing or screen capture have not been independently verified here.
The Show HN prompt signals a submission to Hacker News, but the supplied material does not state when it was posted or include community responses. It is also unclear how the overlays behave in every secure or system-level prompt, or what limitations apply when an app exposes incomplete accessibility labels.
Availability and Testing Ahead
The project page lists installation with Homebrew and provides instructions to build the Swift binary from source. People interested in trying it can review the commands, compatibility requirements and license in the repository. The page also describes a skill-install command for Claude Code and Codex.
No release schedule, roadmap milestone or external evaluation is included in the supplied report. The next useful indicators will be whether users report consistent behavior across their Mac setups, whether the integrations work as described, and whether independent reviewers confirm the stated limits on screen interaction and data collection.
Key Questions
What does bigarrow do?
It draws a temporary arrow or labeled shape over a macOS interface so an AI agent can show a person where to look or act.
Can bigarrow click a button for the user?
No. The project says it does not click, type or capture the screen; the person must perform the action.
Which AI coding tools does it support?
The project describes skills for Claude Code and Codex, alongside its command-line interface.
What systems does the project list as supported?
The repository lists macOS 14 or later. Building from source requires Xcode 16 or later, according to the project page.
Has its security or reliability been independently verified?
The supplied material contains project claims and demonstrations, but no independent audit or broad compatibility test. Those questions remain open.
Source: hn
