Re: Introducing the "Semantic Agent Communication" modeling notes for implementing AgentIDL

Tyson
thanks for sharing your note,  I am glad to see that you can relate your
view to other participants in this CG

Putting together your thoughts is a good start if you need feedback from a
KR point of view

We started a discussion earlier
https://lists.w3.org/Archives/Public/public-cogai/2025Jun/0022.html

Protocols in use
I know of
https://a2a-protocol.org/latest/
https://modelcontextprotocol.io/docs/getting-started/intro

an article on Linkedin
https://www.linkedin.com/posts/brijpandeyji_10-modern-ai-agent-protocols-you-should-know-activity-7354910038961860608-4RCT/

Other than for maintaining an explicit shared representation of the
protocol itself *for communication

I cannot see currently the relation between KR and the execution layer, and
semantic agents
You may have to clarify  your vision to that effect that if you could?

There are many types of AI agents, to what type of AI agent does your
architecture apply to?

Does the architecture you propose enable human readable/visual
representation of what goes on in the operating system?
Or is it supposed to  enable control of how the operating system works?

A demo or a few good examples of how your architecture relates to KR *as we
know it could be useful

The following principles may or may not be relevant to what you are
demonstrating


Snip from pdm  KR research notes
KR Principles for Semantic Agents Design

Ontological Clarity: Any semantic layer should be grounded in clear
ontological commitments. What entities exist in this domain? What
relationships matter? For agentic interaction, this likely includes action
types, consequence categories, content provenance classes, and state
dependencies.

Compositional Semantics: Simple semantic primitives that compose into
richer meaning scales better than exhaustive enumeration. Rather than
trying to label every possible action type, establish composable
properties: reversibility, scope of effect, authorization requirements, and
so forth.

Context Sensitivity: Knowledge representation for agents must account for
context-dependent meaning. The semantics of "purchase" differ between a
free trial activation and a $10,000 enterprise contract.

Graceful Degradation: The semantic layer should degrade gracefully—agents
should function (perhaps suboptimally) when semantic hints are absent, and
sites shouldn't break for users when agent semantics are present

ENDS SNIP BY PDM



On Thu, Nov 27, 2025 at 11:45 AM 陳信屹 <tyson@slashlife.ai> wrote:

> HI all
>
> I put together a collection of notes outlining the minimal algebraic
> structure I use when thinking about agency (μ/ν fixed points,
> operational semantics, and how this connects to task structures and
> grafts) in agent communication.
>
>
> https://figshare.com/collections/_b_Foundations_of_Semantic_Agent_Architecture_-_Working_Notes_b_/8171213
>
> This is not a proposal but for discussions how we think agents
> interact in the real world applications, so we can figure out the
> minimal sets of methods of an agent.
>
> AgentIDL is an open specification (may extending W3C WebIDL) to
> describe semantic agents. It provides a bridge between ontological
> definitions (RDF/OWL/JSON-LD) and executable interfaces for agent
> frameworks, allowing semantic interoperability across AI systems and
> protocols.
>
> the interface syntax looks like.
>
> ```
> [[
>   Context = "
> https://s-agent-comm.github.io/agent-ontology/context/agent.jsonld",
>   Semantic = "
> https://s-agent-comm.github.io/agent-ontology/ontologies/intent.ttl"
> ]]
> interface AgentTask {
>   [Intent="agent:ProposeContract"]
>   Promise<Outcome> proposeContract(ContractData data);
>
>   [Intent="agent:ExecutePayment", Proof="ledger:tx"]
>   Promise<Receipt> executePayment(PaymentRequest payment);
> };
> ```
>
> I have not yet decided which DSL is better tot describe interface).
> what I am more familiar with is WebIDL, but it is kinda too
> Browser-related.
>
> Please let me know if there is a better molding DSL.
>
>