Your Coding Agent Can Refactor a Monorepo But Can't Point at a Button. This Gives It a Finger.
The gap is funny until you hit it. An agent can rewrite a migration, explain monads and restructure a monorepo, and then when it needs you to approve one permission dialog it prints "please click Allow" into a terminal you are not looking at. big-arrow-on-the-screen closes that gap with the dumbest possible fix: it lets the agent paint a giant arrow and a text sign directly on top of your Mac screen, pointing at the exact button you need to press. One Swift binary, MIT licensed, shipped as a skill for Claude Code and Codex. It hit 350 points on Hacker News and 418 stars within a day of the repo going up.
The design choices are the whole point. The arrow draws in a transparent window above every other window, on every display and every Space. Clicks pass straight through it, your keyboard focus never moves, and the overlay removes itself. Drawing needs no macOS permission at all, so the agent can point before it has been granted anything. There is no daemon, no menu-bar icon, no account, no telemetry, and the author is at pains to say there is no model inside. It is, as the README puts it, just an arrow. One command: point at the Allow element in System Settings, with a line of text addressed to the user by name.
Why this matters beyond the gag is that it names a real primitive the agent stack has been missing: agent-to-human pointing. We have spent a year building ways for humans to direct agents and for agents to call tools, and almost nothing for the moment an agent needs a human to do one physical thing in the world and has to say exactly where. A terminal line is the worst possible channel for "click here." An arrow on the screen is the obvious right one, and it is slightly embarrassing that it took until October 2026 for someone to ship it as a clean, permissionless, agent-loadable tool.
It is Mac-only and it is deliberately small, which is the correct scope. The interesting question it raises is whether this becomes a category. Once an agent can point at a pixel, it can walk a user through a settings page, a signup flow, a one-time permission grant, the exact places where today's agents stall and dump instructions into a void. The repo is a toy with a serious idea inside it, which is usually the kind worth watching.
Repo: https://github.com/franzenzenhofer/big-arrow-on-the-screen
← Back to all articles
The design choices are the whole point. The arrow draws in a transparent window above every other window, on every display and every Space. Clicks pass straight through it, your keyboard focus never moves, and the overlay removes itself. Drawing needs no macOS permission at all, so the agent can point before it has been granted anything. There is no daemon, no menu-bar icon, no account, no telemetry, and the author is at pains to say there is no model inside. It is, as the README puts it, just an arrow. One command: point at the Allow element in System Settings, with a line of text addressed to the user by name.
Why this matters beyond the gag is that it names a real primitive the agent stack has been missing: agent-to-human pointing. We have spent a year building ways for humans to direct agents and for agents to call tools, and almost nothing for the moment an agent needs a human to do one physical thing in the world and has to say exactly where. A terminal line is the worst possible channel for "click here." An arrow on the screen is the obvious right one, and it is slightly embarrassing that it took until October 2026 for someone to ship it as a clean, permissionless, agent-loadable tool.
It is Mac-only and it is deliberately small, which is the correct scope. The interesting question it raises is whether this becomes a category. Once an agent can point at a pixel, it can walk a user through a settings page, a signup flow, a one-time permission grant, the exact places where today's agents stall and dump instructions into a void. The repo is a toy with a serious idea inside it, which is usually the kind worth watching.
Repo: https://github.com/franzenzenhofer/big-arrow-on-the-screen
Comments