bigarrow是一款开源的macOS命令行工具和AI Agent技能，可在屏幕上绘制箭头、框和文本以指向特定位置1。该工具采用MIT许可证，使用Swift编写1，旨在解决AI代理无法直接执行需要人类确认的操作（如点击授权按钮、输入双因素认证码）的问题1。通过可视化指向，用户可以清晰了解代理需要用户执行的操作对象1。
bigarrow提供多种灵活的指向方式，支持通过标签、坐标、矩形和窗口等参数进行定位1。用户可自定义样式，选择bend、straight、zigzag或spiral等箭头形状，以及多种颜色选项1。该工具可集成Claude Code和Codex等AI Agent框架作为技能模块1，并通过--say选项支持语音播报指向内容1。
在性能和兼容性方面，bigarrow包含87项自动化测试和17项行为检查，CPU占用仅1.4%（在CI运行器上测量）1。该工具支持多显示器、全屏应用、Stage Manager、Spaces等macOS特性1。
BigArrow, a new open-source macOS command-line tool written in Swift and licensed under MIT, allows artificial intelligence agents to draw arrows, boxes, and text on users' screens to indicate specific locations and actions.1 The tool addresses a critical limitation in AI agent workflows: the inability to directly execute operations that require human confirmation, such as clicking authorization buttons or entering two-factor authentication codes.1 By providing visual on-screen guidance, BigArrow enables users to clearly understand which interface elements or locations an AI agent needs them to interact with.
The tool offers flexible targeting capabilities to suit different use cases.1 Users can specify target elements by HTML tag, precise screen coordinates, rectangular regions, or entire windows.1 BigArrow also supports customizable visual styles, with options for different arrow shapes—including bend, straight, zigzag, and spiral patterns—and multiple color choices.1 Additionally, the tool can verbally announce its instructions through a speech output feature.1 The implementation is efficient, consuming only 1.4% CPU according to measurements taken on CI runners, and includes comprehensive testing with 87 automated tests and 17 behavioral checks.1 BigArrow is fully compatible with macOS features such as multi-display setups, full-screen applications, Stage Manager, and Spaces.1
The tool integrates seamlessly with major AI agent frameworks and platforms, functioning as a skill module for Claude Code and other AI systems.1 This architecture allows developers to incorporate screen-pointing capabilities directly into their AI automation workflows, bridging the gap between autonomous agents and human-confirmed actions.
评论
还没有评论，欢迎留下第一条。