Product

Scan
Monitor
Verify · Assay
Certify · Readygate
How it works
Pricing

Install

Add to your AI

Community

Join cohort
Watch live
Skool

Log in Scan my app →

seatrial · v0.1.0

← Back to the frameworkUse your browser's Print → Save as PDF

Seatrial · Agent Safety Quick-Start

Run this before you connect any agent to a real account.

15-Minute Quick-Start

0 / 12 done

Run this before you connect any agent to a real account.

L0 · Vet the agent

I read exactly what the agent asks permission to do.
I did not pipe a random installer into my shell.

L1 · Isolate it

The agent is connected to a separate account or browser profile, not my main one.
No .env files, SSH keys, or secrets sit in a folder the agent can read.

L2 · Scope its access

I granted read-only access wherever the task allows it.
I connected one service with the narrowest scope, not "everything" at once.

L3 · Gate its actions

The agent must ask me before it sends, deletes, pays, or posts.
Full autonomy / auto-approve is OFF for anything touching money or my inbox.

L4 · Watch it

Activity logging is on and I will review it after the run.
I set a spend / token / rate cap.
I will check for new forwarding rules and connected apps after the first run.

L5 · Recover

I know how to revoke access and rotate credentials if it goes wrong.

Every step here reduces your exposure. None of them eliminates it. Prompt injection has no robust, solved fix as of 2026 — the trifecta works because it removes a precondition for the attack, not because it makes the model impossible to fool. The goal is a smaller blast radius and a cut leg, not "follow 12 steps and you are safe."

seatrial · v0.1.0Privacy Terms Platform agreementSkool · Early AI-dopters