AI agents now write code, send emails, move money, touch customers. When one of them does something wrong — who explains it, and with what evidence?
g8s is the checkpoint system that stands between your AI agents and your systems. Every action authorized in advance. Every refusal explained in machine-readable language. Every outcome provable after the fact. Not a policy document — a working gate.
$help · doctor · submit · receipt · fence · verify · chaos · whoami| task | state | lane | receipt |
|---|---|---|---|
| 4ed2017f | WORKER_COMPLETED | cursor/composer-2.5 | 9636f68d |
| dba40b29 | WORKER_COMPLETED | cursor/composer-2.5 | e9574675 |
| 656d8e02 | WORKER_COMPLETED | cursor/composer-2.5 | 5a41fd02 |
| 98c88b6d | WORKER_COMPLETED | cursor/composer-2.5 | 8e7edea3 |
| …6,255 more rows in the dogfood ledger — every row carries its lineage | |||
9636f68d-625e-4345-826d-272bae7e2561internal/worker/**task 4ed2017f · cursor/composer-2.53300s · expires +55mfalseBefore an agent touches anything, g8s mints one of these. It names the exact paths the agent may change, expires in minutes, and is consumed by exactly one attempt. Inside your systems it is an authorization; on this page it is a promise you can read.
tilt with your cursor · the fields above are real, from today's session