Platform
Delegation & hierarchy
The intended model is that a parent agent delegates a bounded subset of its authority to a child, and that the child can never hold more than the parent holds.
The intended model
- A parent agent holds a set of authorities.
- It delegates a named subset to a child, recorded as a signed decision.
- The child’s authority is bounded by the parent’s, always.
- Revoking the parent invalidates the child’s derived authority.
- The chain of delegation is reconstructible for audit.
This is the design that makes hierarchical revocation possible, and it is the design AWARE’s ADRs work towards.
Current status: roadmap
There is no delegation graph in the shipped implementation. Searching the source for the fields that would carry one — parent agent, delegating agent, ancestors, child agents — returns no production match. The sub-agent permission invariant is explicitly deferred in the IETF authorisation ADR.
What exists today: the coordinator that would manage such a graph, and the revocation primitives that would act on one. What does not exist: the graph itself, and the cascade that walks it.
AWARE’s own revocation research reached the same conclusion independently: before you can revoke over a delegation graph, you need a delegation graph.