Summary

  • Sumit P. Ahuja's 20 October individual Internet-Draft proposes four policy attributes for tasks delegated between agent operators. Three are bilateral; one can be checked across a delegation chain under strict conditions.
  • The draft's opaque tag takes meaning from an agreement between sender and receiver. A third party on the chain cannot read an obligation into it. The tag records a request, while evidence of the receiver's actual handling needs another record.

A tag at the company door

Consider a firm sending a task to a second firm with a short code for a residency rule. The firms have agreed what that code means. The second firm may pass the task onward. At that point the visible code and the operative agreement part company. A downstream firm outside the agreement has no basis for interpreting the code, even if the code arrives intact. If another agreement gives it a meaning, that is a new bilateral fact to establish, not a property inherited from the first hop.

That is the boundary in Ahuja's proposal. It says the receiving domain holding the agreement must apply the agreed handling. It also says a domain without the agreement must not infer meaning. The tag is attached to the delegation, but it does not establish that any handling occurred. For an enterprise buying an agent service, that last sentence carries the bill. A visible instruction can survive a handoff while the only people who can interpret it sit on one side of the next door.

The proposal is an active individual draft, not an IETF standard or a report of deployment. Its value is the precision of the boundary it draws. It does not pretend that a general label can make private contractual terms public, or that a receiver's choice can be dictated by a sender who lacks the receiver's context.

Four attributes, two very different jobs

Ahuja's grammar has four attributes. A domain may lower its priority for receiving new work without declaring itself unreachable. It may rank its own endpoints while leaving the counterparty free to select another. It may attach the opaque bilateral tag. These three express positions between parties; the draft classifies them as bilateral. None supplies a general verdict about what happened along the whole route.

The fourth attribute limits the class of onward delegation. It can support a chain-wide claim that a task was never broadened, but only if every hop made the required comparison, adjacent records are joined inside an authenticated relation, and all relevant comparisons use the same immutable profile and a relation whose transitivity is established. A mediator that ends one delegation and starts another breaks that particular chain. Those conditions are restrictive on purpose. Two individually sound comparisons do not make a continuous route unless the middle value is demonstrably the same.

This chain test still cannot certify the opaque tag. The draft excludes bilateral attributes from chain validity. A successful no-broader comparison says something about the class of delegation. It says nothing about whether a residency term was honored, a priority rule was applied, or a task was accepted by someone with the right authority. The text also says its attributes are declarations, without the enforcement point or evidence artifact that would connect them to execution.

The invisible passenger

The draft's security section identifies a further edge. A tag vocabulary agreed for one purpose can carry information for another. An intermediary outside the agreement may transport content it cannot inspect. Secrecy can be commercially necessary; blind carriage is also a poor basis for an audit claim. If a buyer is promised that a task stayed within a specified operating boundary, it needs the relevant receiver's decision and the actual onward handoff, linked to the agreement that governed them. The tag alone can show what was asked at one hop.

No cited source names a production system that mishandled such a tag. An implementation trace with the exact agreement version, tag, recipient, decision and onward transfer would change the factual picture. Until one exists, the proposal should be read for the control it defines and the proof it leaves to operators.

Sources