Devotd Github
NewDeveloper tools

franzenzenhofer/big-arrow-on-the-screen v0.5.1 — on-screen guidance tool

franzenzenhofer/big-arrow-on-the-screen is a command-line tool that lets AI agents display arrows and guidance text on the macOS screen.

Repositoryfranzenzenhofer/big-arrow-on-the-screen ↗

franzenzenhofer/big-arrow-on-the-screen is a command-line tool that lets AI agents display arrows and guidance text on the macOS screen.

GitHub description: Let your AI agents paint big arrows, boxes and text on your Mac screen. One CLI, click-through, gone by itself. Skill for Claude Code and Codex. MIT.

stars
★ 826
Forks
26
Language
Swift
License
MIT
Latest release
v0.5.1
Read original source ↗View on GitHub ↗

The latest release of franzenzenhofer/big-arrow-on-the-screen is v0.5.1. This repository is a command-line tool that floats arrows, boxes, and guidance text on top of the macOS screen so that AI agents can point out locations. The provided repository materials do not record what changed in v0.5.1. As a result, the changes in this release cannot be confirmed, and the features below are the behavior documented by the project.

The tool only displays overlays on the screen, so mouse clicks pass through and keyboard focus is preserved. The arrow disappears once the specified time has elapsed or the process that displayed it ends. A location can be specified in one of three ways: coordinates, a window, or an accessibility element. An accessibility element is a screen element identified by the operating system. A skill usable in Claude Code and Codex is also provided. A skill is the unit an agent calls to perform a specific task. That said, bigarrow itself has no AI features, and its focus is on showing locations to people rather than operating the screen on their behalf.

The repository is written in Swift and uses the MIT license. The evaluation materials give it a technical score of 5.0 and a momentum score of 2.6, and the described implementation is a combination of standard libraries that does not put forward any new algorithm. These scores alone cannot be used to judge actual performance or stability. As far as can be verified, it is close to the purpose of adding guidance on the screen, and no specific changes for v0.5.1 or performance evidence are given.