A small macOS command-line tool and companion agent skill that draws large, click‑through arrows, boxes and annotated signs on screen so automated agents can point human users at UI elements without ever clicking or stealing focus. Implemented as a single Swift binary with no daemon, no telemetry and MIT license, it runs on every display and Space, requires no screen‑recording permission to draw, and integrates as a skill for Claude Code and Codex. Typical uses include prompting a user to “click Allow” on permission prompts, guiding through onboarding steps, remote help or demos, and visual debugging; it can also speak the label, render arrows into PNGs, and supports a time limit or explicit start/stop lifecycle tied to the agent process.
Controls and targets are explicit: commands let an agent point by element label/app, by coordinates, or start/stop persistent pointers for a window or tab. Arrows follow or hide when their target is covered, can bring the target window forward (or not), and accept many visual options - shape (bend/straight/zigzag/spiral), style (arrow/ring/box), sizes, colors, borders, close button, and text color. Multiple arrows manage sign placement to avoid overlap. Targets include rects, elements, mouse, windows, and Peekaboo snapshots; the tool reports exit codes and diagnostics (doctor) showing which shell app needs Accessibility permission when required.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.