[Agentproto] AgentProto implementation evidence: Cross-domain dialog recovery after mediator failure (B5-2)

[Nikhil D <[e-mail]>]

Dear AgentProto colleagues,

Following my earlier discussion with Tiru regarding privacy-preserving
dialog resumption across administrative trust boundaries, I have completed
a reproducible experimental prototype examining this scenario.

I appreciate Tiru’s clarification that the general requirements are already
addressed in draft-agentic-ai-usecases-requirements-03.

Rather than proposing overlapping requirements, I focused on an
implementation question relevant to B1-6, B3-13, and B5-2:

*How can an independently operated replacement mediator safely recover an
existing dialog when the original mediator’s correlation state is
unavailable and participants use domain-scoped dialog identifiers?*

The prototype evaluates six recovery configurations, comparing shared
identifiers, durable correlation mappings, and ticket-based recovery
mechanisms.

The experiments show that domain-scoped identifiers can support recovery
when appropriate correlation information remains accessible. Where a
replacement mediator cannot access the original mediator’s state,
additional recovery material is needed, with corresponding trust and
identifier-linkability tradeoffs.

The prototype includes automated tests, failure-injection scenarios,
privacy analysis, and reproducible experiment results.

GitHub repository:
*https://github.com/nikhilred2619/agentproto-privacy-preserving-dialog-recovery
<https://github.com/nikhilred2619/agentproto-privacy-preserving-dialog-recovery>*

This is an independent, single-process research prototype with documented
security and deployment limitations. I am not claiming that it establishes
a missing use-case requirement or implements a finalized IETF protocol.

I would appreciate feedback on whether the cross-operator mediator failover
scenario and associated test cases would be useful input to the AgentProto
protocol or reference-architecture work.

If useful, I would be happy to prepare a focused technical contribution
based on the working group’s guidance.

Thank you for your consideration.

Best regards,
Nikhil Reddy Donapati
Software Engineer
NeuroNet Solutions Inc.