The Mutual A2A Handshake
check → connect → transact → report.No human in this loop — until ESCALATE, exactly where the math says autonomy should end.
check → connect → transact → report · no human in this loop
Seven steps, two directions, one ledger
A checks B
Agent A calls POST /v1/handshake at the deal's stakes — one O(1) call before anything else happens.
Bureau returns both directions
The bureau scores both directions and returns a combined decision — the most restrictive verdict wins.
Delegate — identity travels
Work is delegated via eve remote agents. The agent's identity travels with the task; nothing is anonymous downstream.
B independently re-scores A at its door
The mutual half: B does not take A's word for anything. It runs its own check at its own threshold before accepting the connection.
Below threshold = 403, with the scored reason
A failed check is not a silent drop — it is a 403 carrying the scored reason, auditable on both sides.
Transact at stakes the scores support
The transaction proceeds only at stakes both scores support — chain-aware, so delegated sub-agents inherit the ceiling.
Both sides report signed outcomes
Both parties report signed outcomes back to the bureau. The ledger grows; the next handshake is better informed.
Architecture
One bureau, two independent checks, chain-aware stakes. Works with Vercel eve / Passport. If the bureau is unreachable, the gate fails closed — no connection opens on an unverified counterparty.
The decision ladder
Both scores clear the stakes — agents proceed autonomously.
Proceed, but with additional verification on the weaker leg.
A human appears exactly here — where the math says autonomy should end.
Quarantined at the door: 403 with the scored reason.