The problem bigarrow solves is embarrassingly simple, which is exactly why nobody solved it well until now. Your AI coding agent can refactor a monorepo, but when it needs you to click 'Allow' in a system dialog, it prints a message into a terminal you're not watching. bigarrow gives the agent a finger — literally a giant arrow overlaid on your screen, pointing at the thing that needs human attention. The implementation is a single Swift binary. No daemon, no menu-bar icon, no account, no telemetry, no AI inside. It draws a transparent overlay window on top of everything, on every display and every Space. Clicks pass through. Keyboard focus stays put. The arrow removes itself after a configurable duration, or when the agent process that spawned it dies, or when the human submits their next prompt. The tool ships with a install-skill command that teaches Claude Code and Codex how to use it. The targeting system is where the engineering actually lives. You can point by coordinates (--at 760,500), by Accessibility element label (--element "Allow" --app "System Settings"), by window title, by Chrome/Safari tab title, or by Peekaboo snapshot ID. The --app flag raises the right window and selects the right tab before drawing, because pointing at something hidden behind your terminal defeats the purpose. When another app covers the target, the arrow hides until it's visible again. The visual craft is surprisingly deep for a pointing tool. Six arrow shapes (bend, straight, zigzag, spiral), three styles (arrow, ring, box), three sizes, full color control including hex codes, border options, and a close button. Multiple arrows can coexist on screen with automatic sign-collision avoidance. A gallery script renders every combination. The project README demonstrates real-world scenes: guiding someone through System Settings permission toggles, walking through Keynote toolbar steps, helping a parent over screen share ('Not this one, Mom'), and picking the right tab from a forest of Chrome windows. The permission model is clean. Drawing needs zero macOS permissions. Finding elements by Accessibility label needs Accessibility permission — granted to the terminal app, not to bigarrow itself. The tool never clicks, types, or captures. It only points. Every command takes --json and returns structured exit codes (0 ok, 2 bad input, 3 target not found, 4 permission missing), which is the kind of contract agents can actually work with. The project hit 361 GitHub stars in its first two days, with stars from engineers at Apple, NVIDIA, AMD, SAP, Salesforce, Palantir, CoreWeave, Weights & Biases, Stanford, Johns Hopkins, and Oak Ridge National Laboratory. It's MIT licensed, installs via Homebrew (brew install franzenzenhofer/tap/bigarrow), and builds from source with Xcode 16+ on macOS 14+. The README is itself a showcase — staged screenshots using real apps on macOS 27, every command reproducible via shell scripts. This is a tool that does exactly one thing and commits fully to doing it well. The scope discipline is the feature: no screenshot capture, no click automation, no annotation persistence. It fills a genuine gap in the agent-human handoff loop that will only widen as coding agents become more autonomous. The question isn't whether this is useful — it's whether Apple or the agent platforms ship something equivalent before bigarrow becomes the de facto standard.
