An AI agent works fast, and sometimes it works on the wrong thing. You need one switch that stops it, whatever it is doing. Waldo has one. He calls it the panic corner.
What is the fastest way to stop an AI agent on a Mac?
Shove your pointer into the top-left corner of the screen. Every tool, for every agent, stops until you say go. Or type waldo stop.
There is no window to find and no menu to open. Your hand is already on the mouse.
What are the three ways to stop everything?
- Push the pointer into the top-left corner of the main display.
- Run
waldo stop. - Press Stop everything in the bar or on a gate panel.
One honest note. Waldo’s README lists the corner gesture as built but not yet tried live. Until that changes, treat waldo stop and Stop everything as the sure ways. Both are one step.
What stops, and for whom?
Everything Waldo does stops, for everyone. Every tool in every session then refuses until you resume. That covers seeing, clicking, typing, keys, scrolling, dragging, and managing apps. It covers Claude Code, Codex, Gemini CLI, Cursor, and any other agent that borrows Waldo’s hands over MCP.
The switch lives in Waldo’s own folder, ~/Library/Application Support/Waldo, next to policy.json and the audit log. So it is not a setting inside one chat. It holds for every session on your Mac that uses Waldo.
One thing it does not stop is a command your agent runs in its own shell. The shell is outside Waldo. It follows your agent’s own permission settings.
How do I resume?
Resume with waldo resume or the bar’s Resume button. Until you do, every tool refuses. Your agent is told that you stopped it, that nothing was done, and to wait for you.
How do I stop one task and leave the rest alone?
Use the bar. The bar’s Stop button ends one run and leaves the rest alone. Stop everything, next to it, is the switch for every session.
What happens when a gate panel is open?
The panel has its own Stop everything button. While it shows, the tools refuse all input. No answer in two minutes is a no. So if you walk away mid-question, nothing gated happens. See how the hard stops work.
Why does an AI agent need a stop switch?
Because a model can be steered by what it reads. A web page or an email can carry hidden instructions. The OWASP project for AI security calls this indirect prompt injection (OWASP LLM01:2025 Prompt Injection). Waldo’s hard stops catch the riskiest acts: money, sending as you, passwords, and deletes. The stop switch covers everything else. It works outside the model. Every tool checks it before it does anything. Like the gates, it is a guard rail, not a jail: a session that set out to cheat still has its own shell.
Waldo is free and open source. Install him for Claude Code, or for Codex, Gemini CLI, and Cursor. The code is on GitHub.