Can the next agent trust what it receives?
See each protection catch a problem.
Every PAM-E handoff is signed and checked. Here you can try to change one, try to trick it with a fake claim, and check the log of everything that was handed over.
What if someone changes a handoff after it is sent?
We take a real PAM-E handoff, change one character in it, and try three attackers, each smarter than the last.
Try it yourself
Create a fresh handoff, change anything in it (a name, an amount, a single letter), then press Verify.
What if a customer tries to trick the AI?
The customer claims the last agent "approved a $500 refund", and pastes a fake "SYSTEM NOTE TO AI". We check what PAM-E passes on.
Everything that was handed over
Every handoff made in this workspace is logged with its fingerprint. Each entry is chained to the one before, so editing or deleting any entry is detectable.
Stored locally under runs/audit (not committed to git). Stored handoffs contain the conversation text that was handed over.