← All practicals

Delegate down a chain of agents

Agent A hands agent B a subset of its scopes; B hands a subset to C — then kill any one and watch the whole chain die.

1. 1

   On the A2A lab, delegate from one of your active agents to another, naming a subset of the delegator’s scopes.

   Attenuation: the granted scope set can only ever be a SUBSET of what the delegator currently holds — no hop can widen access, only narrow it (RFC 8693 §4.1 nested `act` actor chains).

2. 2

   Paste the resulting token as a “parent token” and delegate again to a third agent.

   Only the agent the parent token was delegated TO may extend it, and the new grant is capped by the PARENT token’s own scope — chains can only ever attenuate further, hop after hop.

3. 3

   Introspect the token and see every hop A → B → C, all live.

   Liveness is never cached in the token — it’s re-derived on every introspection by looking up each agent id in the chain, live, through the SAME registry ai5’s kill switch uses.

4. 4

   Go kill the MIDDLE agent on the Agents lab, then re-introspect the SAME token.

   The whole chain’s token goes dead at once — active flips to false and the killed hop is flagged, even though you didn’t touch the root or the leaf. One kill, anywhere in the chain, is a kill for everyone downstream of it.