[Agentproto] Re: Suggested charter amendments (follow-up on my GitHub issue)

[Leonard Gebauer <[e-mail]>]

Hi Leslie,

I just want to do a quick follow up because maybe you missed or didn't
receive my rationale (or i didn't receive your answer).

Here is the email again (hope the link works):
https://mailarchive.ietf.org/arch/msg/agentproto/RaO3d-ysmPouixg84twjp8qiNXw/#

If you didn't answer because my email didn't meet the new requirements or
was unclear/not useful for you or I just missed your answer, please let me
know.

Thank you!

Best,
Leonard Gebauer

On Thu, Aug 27, 2026 at 12:12 PM Leonard Gebauer <
[e-mail]> wrote:

> And i just noticed that i didn't use STE, i'm sorry for that. I'll
> definitely use it in future.
>
> Leonard Gebauer <[e-mail]> schrieb am Do., 27. Aug.
> 2026, 13:06:
>
>> Hi Leslie,
>>
>> thank you for the feedback and for clarifying the expectations around
>> rationale.
>>
>> I understand that the current charter text reflects prior discussion and
>> rough consensus on inclusion and exclusion. My proposals are intended as
>> targeted refinements that I believe strengthen the document in the
>> direction the chairs have recently emphasized, concrete interoperability
>> problems and the minimum necessary primitives, rather than expanding the
>> scope.
>>
>> Here is the rationale for each proposal, tied to the current text:
>>
>> 1.) Human-agent communication
>>
>> The charter currently places “protocol-level mechanisms for establishing
>> sessions, negotiating modalities, and exchanging multimodal data between a
>> human user and an AI Agent” in scope, while the Out-of-Scope section
>> already excludes “the design of human to agent user interfaces, client
>> application UX, or the rendering of agent outputs on end-user devices.” The
>> proposed replacement closes the remaining gap: it keeps shared session and
>> transport mechanisms (usable by both agent-to-agent and human-agent
>> interactions) in scope, but explicitly places human-specific interaction
>> patterns such as barge-in handling or modality negotiation for end-user UIs
>> out of scope. This reduces the risk of the group being drawn into
>> application-layer UI concerns while preserving the foundational session
>> work.
>>
>> 2.) Discovery and capability advertisement
>>
>> I remain less certain about this item. Discovery is already acknowledged
>> under Coordination (INT and OPS areas) and in the Key Considerations around
>> selecting agents and tools. Adding an explicit requirements-and-integration
>> bullet to the Framework deliverable would make that coordination more
>> visible and help ensure the session protocol remains usable across trust
>> domains. However, I recognize this could be seen as moving beyond the
>> current minimal-primitives focus. I am happy to drop or defer it based on
>> the group’s and your view.
>>
>> 3.) Definition of “AI agent”
>>
>> The current definition (“an autonomous, adaptive intelligent software
>> system that uses AI models to complete a specific task”) is narrower than
>> the description that immediately follows it in the charter. The proposed
>> wording aligns the definition more closely with the systems that actually
>> require inter-domain interoperability, entities that pursue goals, make
>> decisions, and interact with other agents or tools, without materially
>> expanding the scope of the work.
>>
>> 4.) Success criteria
>>
>> The current charter does not contain explicit success or readiness
>> criteria. Adding clear, pragmatic criteria (stable architectural building
>> blocks for the Framework document; adoption of the Session Protocol by at
>> least two independent agent frameworks) provides the group with measurable
>> exit ramps and reinforces the focus on usable, interoperable results.
>>
>> 5.) Behavioral security boundary
>>
>> The existing Out-of-Scope text already places pure model-level behavioral
>> security out of scope while keeping protocol-level user confirmation in
>> scope. The proposed wording makes that boundary more precise by also
>> excluding mechanisms whose primary purpose is to mitigate model-level
>> failures (e.g., prompt injection) beyond confirmation and authorization
>> hooks. This further protects the group from scope creep into
>> model-alignment territory while preserving the confirmation mechanisms that
>> are already agreed to be in scope.
>>
>> In short, proposals 1, 3, 4 and 5 are intended to clarify and tighten the
>> existing consensus direction. Proposal 2 is the one I am least attached to
>> and where I would particularly value your and the group’s feedback.
>>
>> I am happy to withdraw any item that does not help us reach a clearer
>> consensus-ready charter, or discuss the points individually.
>>
>> Best regards,
>> Leonard
>>
>> [NOTE: The text was translated from German to English by an AI, which
>> also reworked it to be more readable. Edited by the Author.]
>>
>> Leslie Daigle (ThinkingCat) <[e-mail]> schrieb am Mi.,
>> 26. Aug. 2026, 00:21:
>>
>>> Hi,
>>>
>>> Thanks for the suggestions. As we are working on developing a text that
>>> has consensus support, and with the understanding that all the text that is
>>> in the current version of the charter is the result of several minds
>>> agreeing on both what to put in and what to leave out: could you identify
>>> why you propose those updates as an improvement in our effort to make a
>>> consensus-based charter for an IETF working group?
>>>
>>> This mailing list is for the discussion of creating a charter for an
>>> AGENTPROTO IETF working group, and carrying out  the work the group is then
>>> chartered to do.   General discussion of (AI) agents is not in scope.
>>> Discussion of specific implementations is not relevant, except where it
>>> supports a written statement proposing an update to existing charter text.
>>> The AGENTPROTO charter proposal is available here (
>>> https://github.com/ietf-artarea/charters/blob/main/agentproto/charter.md)
>>> If you have issues with any of the text, or suggestions for additional or
>>> different content in the charter, please provide your rationale.
>>>
>>> Thanks,
>>> Leslie.
>>>
>>> On 24 Aug 2026, at 17:17, Leonard Gebauer wrote:
>>>
>>> Hi all,
>>>
>>> I have opened an issue on the charter repository with five proposed
>>> changes
>>> to the current draft. I’d like to briefly present them here for
>>> discussion,
>>> because i think the discussion about proposals will take place here.
>>>
>>> I understand that the chairs have recently refocused the discussion on
>>> concrete interoperability problems and the minimum necessary primitives.
>>> Still, i want to introduce my proposal, as I believe proposals 1, 3, and
>>> 5
>>> help clarify and sharpen the scope along those lines. Proposal 4 adds
>>> success criteria that align well with this pragmatic focus.
>>>
>>> However, I'm less certain about proposal 2 (discovery and capability
>>> advertisement). I realize this might be a more contentious topic and
>>> potentially out of step with the current priority on minimal primitives.
>>> I
>>> would appreciate the group's feedback on whether this belongs in the
>>> charter at this stage.
>>>
>>> Here is the exact copy of the issue, as posted by me on Github:
>>>
>>> --- BEGIN ISSUE TEXT ---
>>>
>>> Hi all! Here are some ideas i have for the charter:
>>>
>>> 1.)
>>>
>>> Replace: „Human-agent communication protocols — specifically the
>>> protocol-level mechanisms for establishing sessions, negotiating
>>> modalities, and exchanging multimodal data between a human user and an AI
>>> Agent — are also in scope.“
>>>
>>> With: „Human-agent communication mechanisms that are specific to human
>>> interaction patterns (e.g., barge-in handling or modality negotiation for
>>> end-user UIs) are out of scope. Shared session and transport mechanisms
>>> that can be used by both agent-to-agent and human-agent interactions
>>> remain
>>> in scope.“
>>>
>>> 2.)
>>>
>>> New bullet point in Framework-Deliverable:
>>>
>>> „• Identify requirements and integration approaches for agent and tool
>>> discovery, including capability advertisement and trust-domain discovery,
>>> and coordinate with relevant existing mechanisms.“
>>>
>>> 3.)
>>>
>>> Replace: „An AI agent is an autonomous, adaptive intelligent software
>>> system that uses AI models to complete a specific task.“
>>>
>>> With: „An AI agent is an autonomous software entity that uses one or more
>>> AI models to complete tasks, pursue goals, make decisions, and interact
>>> with other agents or tools.“
>>>
>>> 4,)
>>>
>>> Add as Success Criteria: „The working group will consider the Framework
>>> document ready for Working Group Last Call once the core architectural
>>> building blocks and their relationships are stable. The Session Protocol
>>> work is considered successful when it provides a usable, interoperable
>>> session layer that can be adopted by at least two independent agent
>>> frameworks.“
>>>
>>> 5.)
>>>
>>> Replace: „• AI agent behavioral security (e.g., preventing the AI model
>>> itself from hallucinating, though mitigating the impact of hallucinations
>>> via protocol-level user confirmation is in scope).“
>>>
>>> With: „• AI agent behavioral security (e.g., preventing the AI model
>>> itself
>>> from hallucinating). Protocol mechanisms whose primary purpose is to
>>> mitigate model-level failures (e.g., prompt injection) beyond providing
>>> confirmation and authorization hooks are also out of scope. Mitigating
>>> the
>>> impact of hallucinations via protocol-level user confirmation remains in
>>> scope.“
>>>
>>> [NOTE: Original Text was in German, it was translated via AI which ALSO
>>> helped to make it more readable. From my perspective it seemed fine,
>>> however i am sorry if there is something incorrect in my proposal due to
>>> this procedure.]
>>>
>>> --- END ISSUE TEXT ---
>>>
>>> Looking forward to your thoughts!
>>>
>>> Best regards,
>>> Leonard Gebauer
>>>
>>> [NOTE: This email was translated from german into english and edited to
>>> be
>>> more readable by an AI.]
>>> ------------------------------
>>>
>>> Agentproto mailing list -- [e-mail]
>>> To unsubscribe send an email to [e-mail]
>>>
>>> --
>>> ------------------------------
>>> Leslie Daigle
>>> Principal, ThinkingCat Enterprises
>>> [e-mail]
>>>
>>