Re: Special Topic Call - Agentic Linked Data

st 9. 4. 2025 v 17:14 odesílatel Melvin Carvalho <melvincarvalho@gmail.com>
napsal:

>
>
> út 1. 4. 2025 v 10:44 odesílatel Melvin Carvalho <melvincarvalho@gmail.com>
> napsal:
>
>>
>>
>> út 1. 4. 2025 v 1:07 odesílatel Rui Zhao <rui.zhao@cs.ox.ac.uk> napsal:
>>
>>> Hi all,
>>>
>>> Forgive my failure to attend in the previous call. Is there a time slot
>>> set out for the next meeting?
>>>
>>>
>>> Also regarding authorisation and delegation: indeed I would say that's
>>> no less difficult (if not more difficult) than authentication. There will
>>> be the perception of the user / owner, and also the autonomous interaction
>>> of agents; in addition, there will be disappearing of agents. I would have
>>> a lot to argue in this area, given my background in data usage control...
>>>
>>
>> We dont have a slot yet, but we can decide our own timing.  I think it
>> would be nice if one of the chairs were to attend.  Elf, did you perhaps
>> have some interest in the topic?
>>
>> I think the topics are shaping up as follows
>>
>> - Defining an Agent
>> - How Agents can interact with Solid
>> - Agent Authentication
>> - Agent Authorization
>> - Next Steps for group
>>
>> [ +Other items TBD ]
>>
>
> This just released from Google the Agent 2 Agent (A2A) spec, might be
> relevant:
>
> "A new era of Agent Interoperability"
>
> Blog:
> https://developers.googleblog.com/en/a2a-a-new-era-of-agent-interoperability/
>
> Source: https://github.com/google/A2A
>

I particular, have a look at Agent Card:

https://google.github.io/A2A/#/documentation?id=agent-card

An *Agent Card* describes the capabilities, interfaces, and usage
characteristics of an agent. It serves both human users and other agents
wishing to interact with it.
------------------------------
🧾 Core Components 🔹 Basic Info

   -

   *Name*: A human-readable name (e.g., *"Recipe Agent"*).
   -

   *Description*: A brief summary of the agent's purpose and functionality
   (e.g., *"Helps users with recipes and cooking"*).
   -

   *URL*: The endpoint or base URL where the agent is hosted.
   -

   *Provider* *(optional)*: Information about the service provider
   (organization name and homepage).
   -

   *Version*: A string identifying the version (format up to provider,
   e.g., *"1.0.0"*).
   -

   *Documentation URL* *(optional)*: A link to technical or usage
   documentation.

------------------------------
🛠 Capabilities

An agent may support a variety of capabilities:

   -

   streaming: Supports Server-Sent Events (SSE)
   -

   pushNotifications: Can notify the client of updates
   -

   stateTransitionHistory: Can expose a history of task status changes

These allow clients to understand interaction patterns, performance
expectations, and integration possibilities.
------------------------------
🔐 Authentication

Specifies how to authenticate with the agent:

   -

   schemes: One or more auth mechanisms (e.g., Basic, Bearer)
   -

   credentials *(optional)*: Any credentials required to access private
   Agent Cards

This is aligned with OpenAPI authentication models for easy integration.
------------------------------
🔄 Input/Output Modalities

Defines which *MIME types* the agent accepts and produces by default:

   -

   defaultInputModes: e.g., text/plain, application/json
   -

   defaultOutputModes: e.g., text/html, application/json

These can be overridden at the *skill* level.
------------------------------
🧩 Skills

Each *skill* is a modular unit of the agent's functionality.

   -

   id: Unique identifier for the skill
   -

   name: Human-readable label (e.g., *"Suggest Recipe"*)
   -

   description: What the skill does and how it’s useful
   -

   tags: Keywords to classify the skill (e.g., cooking, customer support)
   -

   examples: Example prompts that illustrate typical use cases
   -

   inputModes / outputModes *(optional)*: Skill-specific overrides for
   supported MIME types




>
>
>>
>>
>>>
>>> Rui
>>>
>>> *From:* Melvin Carvalho <melvincarvalho@gmail.com>
>>> <melvincarvalho@gmail.com>
>>>
>>> *Sent:* Monday, March 31, 2025 at 1:16 PM UTC+1
>>>
>>> *To:* Joshua Cornejo <josh@marketdata.md> <josh@marketdata.md>
>>>
>>> *Cc:* public-solid <public-solid@w3.org> <public-solid@w3.org>
>>>
>>> *Subject:* RE: Special Topic Call - Agentic Linked Data
>>>
>>>
>>>
>>> ne 23. 3. 2025 v 11:42 odesílatel Joshua Cornejo <josh@marketdata.md>
>>> napsal:
>>>
>>>> Re: point 2) – OIDC already has a lot of work in that area, and a gap
>>>> is acknowledged:
>>>>
>>>>
>>>>
>>>> https://mayakaczorowski.com/blogs/ai-agent-authentication
>>>>
>>>>
>>>>
>>>> There is also the topic of authorisation and delegation (OIDC’s answer
>>>> to my question in January – *“**Some of what you need to do hasn't
>>>> been invented yet.**”*)
>>>>
>>>
>>> Of possible interest (thanks to spotter: Michiel!):
>>>
>>> The Next Phase of the Web: Better, Safer AI
>>>
>>>
>>> https://www.inrupt.com/blog/the-webs-inventor-on-building-ai-that-benefits-everyone
>>>
>>>
>>>>
>>>>
>>>> Regards,
>>>>
>>>> ___________________________________
>>>>
>>>> *Joshua Cornejo*
>>>>
>>>> *marketdata <https://www.marketdata.md/>*
>>>>
>>>> smart authorisation management for the AI-era
>>>>
>>>>
>>>>
>>>> *From: *Melvin Carvalho <melvincarvalho@gmail.com>
>>>> *Date: *Sunday, 23 March 2025 at 05:30
>>>> *To: *public-solid <public-solid@w3.org>
>>>> *Subject: *Special Topic Call - Agentic Linked Data
>>>> *Resent-From: *<public-solid@w3.org>
>>>> *Resent-Date: *Sun, 23 Mar 2025 05:29:56 +0000
>>>>
>>>>
>>>>
>>>> Hi all,
>>>>
>>>> There’s been some interest around the idea of *Agentic Linked Data* —
>>>> think of it as Social Linked Data, but with a focus on *Agents*:
>>>>
>>>> 📄 https://hackmd.io/@jSWt69rrQAKeUTsa_s49Qg/H16RCGRv1g
>>>>
>>>> After chatting with Tim and the chairs, they kindly suggested we could
>>>> explore the topic in a *Special Topic Meeting*. Unlike a regular CG
>>>> call (where we might get ~10 minutes), this would give us a full hour or
>>>> more to dig in.
>>>>
>>>> We’re probably a little early to set a date just yet — but I’d love to
>>>> start gathering *topics* for the agenda. Elf suggested 4–6, so here
>>>> are two to kick things off:
>>>> 🔹 Special Topic Suggestions
>>>>
>>>>    1. *How Agents can interact with the Solid ecosystem*
>>>>    Tim noted that every Agent should have an owner. There are also a
>>>>    few real examples of Agents in Solid we can build from.
>>>>    2. *Authentication of Agents*
>>>>    Jesse raised Agent Auth as a key point — it’s come up a lot in
>>>>    discussions and deserves focused time.
>>>>
>>>> ------------------------------
>>>>
>>>> *What other topics would you like to explore on a call like this?*
>>>> Feel free to share ideas, and we’ll shape the agenda around them. And
>>>> if we don’t get to everything, we can always figure it out live on the call.
>>>>
>>>> Looking forward to hearing your thoughts or any links you would like to
>>>> share!
>>>>
>>>> Best,
>>>> Melvin
>>>>
>>>>
>>>>
>>> --
>>> Rui Zhao
>>> Postdoc @ EWADA (Ethical Web And Data Architecutres in the Age of AI)
>>> Human-Centred Computing, Dept. Computer Science
>>> University of Oxfordhttps://me.ryey.icu
>>>
>>>