How it works
You give it a task. You keep the veto.
Sentinel is not a chat box that promises to be careful. It is a working agent wrapped in checks that run before each step, not after.
1. Say what you want
Type the task the way you would say it to a person: "find the cheapest flight to Lisbon next Friday and get to checkout, don't pay".
2. See the plan first
Sentinel lists the steps it intends to take before it takes any of them. You can change the task or set tighter limits before it starts.
3. Watch it work
The browser view shows the pages. The feed beside it names every step as it happens: what it opened, clicked and typed.
4. Decide the moments that matter
Anything you have marked off limits stops dead. You see the step, the reason, and the rule that fired, then allow it once or refuse it.
5. Stop whenever you like
Pause and Stop are always on screen. Stopping ends the run where it stands; nothing continues in the background.
Three outcomes
Every step ends in one of three places
There is no fourth, quieter outcome. If Sentinel cannot tell what happened, it says so instead of reporting success.
By default agents act on your task without asking, inside your limits, including sending email and placing calls.
Off by default. Set any skill to Ask and that step waits for your OK, shown in full first.
Passwords, deletions, downloads, off-limits sites. Refused, with the reason.
Other agents
It also stands in front of the agents you already use
Sentinel accepts requests from other agents and tools on your machine, checks them against the same limits, and allows, blocks, or asks you. Blocked attempts appear in the same feed with the reason.