A Kill Switch for AI Agents You Didn't Build

​Claude Code, Codex, Cursor and Copilot now run unattended on laptops and CI machines. When one goes wrong (a dropped database, a runaway loop, a surprise bill), most teams find out afterwards, with no button to press.

​In this 45-minute live session, Vivek Chand, creator of the open-source ClawMetry, shows how to put a kill switch on agents you didn't build. No SDK, and no change to the agent:

​See it: every session, tool call and dollar, read from what the agent already writes to disk

​Cap it: a budget that refuses the next model call instead of emailing you afterwards

​Stop it: pause, stop or kill a live agent process

​Hold it: a pre-tool gate that holds a DROP DATABASE for a human to approve

​Mostly live demo on real agents, plus 15 minutes of open Q&A. Attendees get an offer for ClawMetry Pro.

​For: platform, DevEx and security engineers, and anyone letting coding agents run unattended.

​Streamed on YouTube Live. The link arrives in your confirmation email and reminders.

​Can't make it live? Register anyway. Everyone gets the recording.