14:50:12 <RRSAgent> RRSAgent has joined #webmachinelearning
14:50:17 <RRSAgent> logging to https://www.w3.org/2026/05/14-webmachinelearning-irc
14:50:17 <Zakim> RRSAgent, make logs Public
14:50:18 <Zakim> please title this meeting ("meeting: ..."), anssik
14:50:18 <anssik> Meeting: WebML CG Teleconference – 14 May 2026
14:50:23 <anssik> Chair: Anssi
14:50:30 <anssik> Agenda: https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-05-14-cg-agenda.md
14:50:34 <anssik> Scribe: Anssi
14:50:39 <anssik> ScribeNick: anssik
14:50:50 <anssik> Present+ Anssi_Kostiainen
14:50:54 <anssik> RRSAgent, draft minutes
14:50:55 <RRSAgent> I have made the request to generate https://www.w3.org/2026/05/14-webmachinelearning-minutes.html anssik
14:54:15 <JeffWhelpley> JeffWhelpley has joined #webmachinelearning
14:55:45 <anssik> Present+ Jeff_Whelpley
15:00:22 <anssik> Present+ Guru_Singh
15:00:23 <anssik> Present+ Mike_Wasserman
15:00:35 <anssik> Present+ Wenson_Ssieh
15:00:53 <Mark_Foltz> Mark_Foltz has joined #webmachinelearning
15:01:00 <anssik> Present+ Chris_Needhan
15:01:05 <Mark_Foltz> Present+ Mark_Foltz
15:01:23 <cpn> cpn has joined #webmachinelearning
15:01:27 <anssik> Present+ Mark_Foltz
15:01:52 <anssik> Present+ Ben_VanderSloot
15:02:06 <anssik> Present+ Iris_Johnson
15:02:06 <bvandersloot> bvandersloot has joined #webmachinelearning
15:02:12 <domfarolino> domfarolino has joined #webmachinelearning
15:02:19 <anssik> Present+ Dominic_Farolino
15:02:32 <anssik> Present+ Yongjun_Zhang
15:02:41 <brwalder> brwalder has joined #webmachinelearning
15:02:46 <Victor> Victor has joined #webmachinelearning
15:02:47 <anssik> Present+ Victor_Huang
15:02:48 <cpn> present+ Chris Needham
15:02:50 <gurusingh> gurusingh has joined #webmachinelearning
15:02:51 <bvandersloot> present+ Benjamin_VanderSloot
15:02:59 <AlexN> AlexN has joined #webmachinelearning
15:03:07 <anssik> Present+ Alex_Nahas
15:03:07 <iris> iris has joined #webmachinelearning
15:03:19 <anssik> RRSAgent, draft minutes
15:03:21 <RRSAgent> I have made the request to generate https://www.w3.org/2026/05/14-webmachinelearning-minutes.html anssik
15:03:38 <anssik> Anssi: as a reminder, we'll use IRC-based queue management in this meeting:
15:03:42 <anssik> -> https://www.w3.org/guide/meetings/zakim.html#speakerqueue
15:03:49 <gurusingh> present+ Guru_Singh
15:03:56 <anssik> Anssi: to suggest agenda topics, use Agenda+ label, e.g.:
15:04:02 <anssik> -> Agenda+ https://github.com/webmachinelearning/webmcp/labels/Agenda+
15:04:18 <anssik> Anssi: welcome to the new participants to the group:
15:04:23 <anssik> ... Sarah Drasner from Google
15:04:28 <anssik> ... Guru Singh and Cathy Li from Meta
15:04:38 <anssik> ... Chandi Cumaranatunge and Phillip Long from Arizona State University
15:04:44 <anssik> ...  Arham Awan, Sean Brunnock, Thomas Jones and Simeon Vincent as individuals
15:04:51 <anssik> ... welcome all to the WebML Community Group!
15:05:06 <anssik> Topic: WebMCP
15:05:11 <anssik> gb, this is webmachinelearning/webmcp
15:05:11 <gb> anssik, OK.
15:05:16 <anssik> Subtopic: WebMCP early wide review
15:05:38 <anssik> Anssi: due to strong interest in WebMCP at large, we are planning to do an early wide review of the spec to gather feedback from a broader audience
15:05:56 <anssik> ... this includes W3C's horizontal groups such as Accessibility, Internationalization, Privacy, Security, and the TAG
15:06:04 <Ehsan> Ehsan has joined #webmachinelearning
15:06:19 <anssik> ... input from the broader web community, web developers and users, will be considered by the group as well
15:06:49 <anssik> ... and I ask all participants to help with gathering such feedback through their channels and bring it to the group for consideration, including experimental implementations and prototypes
15:06:55 <anssik> ... we are tracking related issues in the WebMCP repo with the following labels:
15:07:01 <anssik> -> https://github.com/webmachinelearning/webmcp/labels/a11y-tracker
15:07:05 <anssik> -> https://github.com/webmachinelearning/webmcp/labels/tag-tracker
15:07:10 <anssik> -> https://github.com/webmachinelearning/webmcp/labels/i18n-tracker
15:07:16 <anssik> -> https://github.com/webmachinelearning/webmcp/labels/privacy-tracker
15:07:16 <alispivak> alispivak has joined #webmachinelearning
15:07:20 <anssik> -> https://github.com/webmachinelearning/webmcp/labels/security-tracker
15:07:49 <anssik> Anssi: I encourage the group to help triage issues with these labels, and to bring up any issues that you think should be part of the early wide review
15:08:04 <anssik> ... I will work with the editors on a plan how to reach out to the groups when we are ready to start the reviews
15:08:28 <anssik> ... many of these horizontal groups expect some pre-work, and we will need to prepare for that, e.g. make sure the explainer is up to date, and that we have a clear list of issues that we want to get feedback on
15:08:34 <anssik> ... I will reach out to the chairs of the respective horizontal groups to give a heads up on our plans
15:09:06 <anssik> ... W3C Process does not require CG work to go through a wide review, but I think it's important to get early feedback for this spec, given the potential impact on the web platform and the strong interest from the broader web community
15:09:27 <anssik> ... do folks have a preference with respect to the priority order of the reviews?
15:09:29 <whsieh> whsieh has joined #webmachinelearning
15:09:33 <anssik> q+
15:09:35 <anssik> ack anssik
15:09:37 <domfarolino> q+
15:09:41 <anssik> ... we already have engaged with the TAG, Security IG and i18n WG, so we could start with those groups
15:09:44 <anssik> ack domfarolino
15:10:06 <anssik> Dominic: we had discussion with the TAG, I'd like to get a formal review from them for the spec
15:10:46 <anssik> ... Johann may know about security threat model for WebMCP, that artifact exists in our heads, we can document that and work with the Security IG
15:10:59 <anssik> Johann: no update yet on formal threat model
15:11:09 <anssik> Dominic: do we have a draft yet?
15:11:13 <anssik> Johann: not yet
15:11:15 <johannhof> johannhof has joined #webmachinelearning
15:11:24 <anssik> Dominic: let's coordinate on that
15:12:35 <anssik> Anssi: Johann would you like to own the threat modeling exercise?
15:12:48 <anssik> Johann: I understood Simone was interested in working with us
15:12:49 <JakeA> JakeA has joined #webmachinelearning
15:12:57 <anssik> ... Victor could also be interested
15:13:19 <anssik> Victor: I would be interested, I did understand this is a new exercise
15:14:23 <anssik> Johann: we could move the S&P Considerations markdown that Victor initiated is already pretty solid, we could use it as a starting point for the Security review and threat model
15:16:38 <anssik> BenjaminV: could help fill in S&P questionnaire
15:18:51 <domfarolino> +1
15:18:57 <anssik> RESOLUTION: Group starts WebMCP early wide review with the TAG, Security IG/WG, Privacy WG.
15:19:31 <anssik> Subtopic: Consider making registerTool() return a Promise
15:19:35 <anssik> Anssi: issue #175
15:19:37 <gb> https://github.com/webmachinelearning/webmcp/issues/175 -> Issue 175 Consider making `registerTool()` return a Promise (by domfarolino) [Agenda+]
15:19:48 <anssik> ... Domenic proposed that registerTool() should return a Promise due to cross-origin iframe integration, already asynchronous by nature
15:20:05 <anssik> ... most importantly, this would help with forward compatibility, enable reporting on failure cases ergonomically
15:20:21 <anssik> ... alternative proposed design would be signal failures through the ontoolchange attribute
15:20:27 <anssik> ... yet per issue discussion everything points to the Promise-based design as the more ergonomic and forward-compatible approach
15:20:28 <anssik> q?
15:21:02 <anssik> Dominic: I think myself and Francois are aligned, we want to hide sync failures behind a microtask, ontoolchange is not at odds with this proposal
15:21:08 <anssik> q?
15:21:20 <Victor> Is the S&P questionnaire needed for TAG review?
15:21:26 <domfarolino> +1
15:21:56 <Victor> +1
15:21:59 <anssik> RESOLUTION: Make registerTool() return a Promise for forward compatibility and improved ergonomics. (issue #175)
15:22:15 <anssik> Subtopic: Hint for reversible or consequential actions
15:22:19 <anssik> Anssi: issue #176
15:22:20 <gb> https://github.com/webmachinelearning/webmcp/issues/176 -> Issue 176 Hint for reversible or consequential actions (by johannhof) [Agenda+]
15:22:33 <anssik> ... currently read v write actions are differentiated with readOnlyHint only
15:22:41 <anssik> ... Johann points out this does not align too well with the Human in the Loop principles
15:23:05 <anssik> ... per HCI guidelines, humans like to undo things when they explore user interfaces, the agents should be able to do the same
15:23:32 <anssik> Johann: I'd like to hear from Ben and Victor on their thoughts
15:23:40 <Victor> q+
15:23:44 <anssik> ... we seem to agree reversible is the correct signal
15:23:54 <anssik> ... destructive isn't too far off was another comment
15:23:55 <anssik> q?
15:23:57 <anssik> ack Victor
15:24:01 <bvandersloot> q+
15:24:22 <anssik> Victor: I think destructive, per historical discussion, it can from a bucket of things, he just made it happen
15:24:37 <anssik> ... CRUD construct motivated that design
15:25:12 <anssik> ... on the front of whether reversible is the right thing for this API, consequential could be too loosely applies
15:25:27 <anssik> ... reversible is bigger could, consequential is a subset
15:26:01 <anssik> Johann: nuances in each of these hints, example for each that won't perfectly fit
15:26:07 <anssik> ... example use case: drafting an email is reversible, while sending is not
15:26:23 <anssik> ... what is creating a calendar invite that goes to other people?
15:26:42 <anssik> ... a ton of nuance on these examples
15:27:38 <anssik> reversible seems the best match, could also argue we want to go with what MCP spec uses
15:28:00 <anssik> Victor: calendar use case could be both consequential and destructive
15:28:08 <anssik> ... combined together
15:28:50 <anssik> ... I think MCP has openWorldHint, does not make sense for the web that is always open world
15:29:28 <anssik> ... maybe we bring in destructive and add a similar semantic such as "open world"
15:29:43 <anssik> Johann: make it something like action is "noticeable by other peoples"
15:30:03 <anssik> ... not sure if there's enough use cases for destructive in the narrowest sense
15:30:18 <anssik> q?
15:30:42 <anssik> Dominic: we want to reduce actions the agent can do
15:31:19 <anssik> ... maybe there's a use case for all the actions, semantics or hints are hard to get apply to all use cases
15:31:49 <anssik> ... lifting the user interaction action from the tool's implementation to the registration seems clearer to me
15:32:30 <anssik> Johann: I think the idea is user agents may make some rules based on security state, having that available at runtime so the website could decide if it should confirm with the user
15:32:54 <anssik> Victor: pushing people to use requestUserInteraction seems wrong
15:33:04 <anssik> ... "I have made my decision with the agent"
15:33:21 <anssik> ... pushing people to do user interaction, you force developers to a specific UX
15:34:06 <Victor> q+
15:34:11 <anssik> Dominic: makes sense, maybe this is covered by tool description, "this calendar event may be booked with someone important that cannot be taken away"
15:34:47 <anssik> ... embed hint to description, agent would get more context this way filling the user's intent
15:35:00 <anssik> ack bvandersloot
15:35:03 <johannhof> q+
15:35:22 <anssik> BenjaminV: I want to push back on the idea everything is open world, for example, adding to the shopping cart
15:35:47 <anssik> ... question is do you try to mimic the regular MCP trying to get some level of synergy there
15:36:12 <anssik> ... I like Dominic's alternative of readOnlyHint imply "no user interaction"
15:36:35 <anssik> ... lifting the context out and making it explicit what it does makes more sense to me, unless you want to stick to the regular MCP interface
15:36:36 <anssik> q?
15:36:41 <anssik> ack Victor
15:36:51 <bvandersloot> q+
15:36:57 <anssik> Victor: either boolean or enum
15:37:14 <anssik> ... you could filter all tools and do only readOnlyHint research
15:37:36 <anssik> ... you can deterministically do that, boolean or something that can be iterated on
15:37:49 <anssik> ack johannhof
15:38:04 <anssik> Johann: pushing back on free text to the agent, this is not meant for the model
15:38:34 <anssik> ... we are in an environment were we trust the model less, not show you the tool that do certain actions
15:38:35 <anssik> q?
15:38:37 <anssik> ack bvandersloot
15:39:18 <anssik> BenjaminV: we should try to have a property that requestUserInteraction inside a call means the user will always going to see it
15:39:28 <anssik> ... abusive site could be hammering away
15:39:28 <anssik> q?
15:40:01 <anssik> Victor: I should narrow down on the set we want to add
15:41:59 <anssik> Johann: there will be agents with no capacity of big tech first, we want to allow also simpler agents to be compliant
15:42:10 <domfarolino> +1
15:42:14 <Victor> +1
15:42:16 <reillyg> Present+ Reilly_Grant
15:42:18 <AlexN> +1
15:42:28 <anssik> Subtopic: Support cross-frame tool enumeration and composability while handling name collisions
15:42:32 <anssik> Anssi: issue #160
15:42:33 <gb> https://github.com/webmachinelearning/webmcp/issues/160 -> Issue 160 Support cross-frame tool enumeration and composability while handling name collisions (by markafoltz) [Agenda+]
15:42:46 <anssik> ... this is how to allow enumeration of tools across frames without name collisions
15:42:51 <anssik> ... MarkF did the initial investigation and proposed three possible solutions
15:43:05 <anssik> ... in follow-ups I see Dominic suggesting a postMessage() inspired security model
15:43:37 <anssik> Dominic: "getTools() returns a flat list of tool objects, each containing at least the WindowProxy or origin of the source that registered the tool."
15:43:51 <anssik> MarkF: tool enumeration steps should address this issue
15:44:04 <reillyg> q+
15:44:08 <anssik> Dominic: I will update the spec and explainer accordingly
15:44:08 <anssik> q?
15:44:24 <anssik> ack reillyg
15:44:54 <anssik> Reilly: does not matter for in-browser agents, but for in-page agents, do we want to include that tool enumeration has explicit opt-in for cross-origin tools?
15:45:10 <anssik> ... so that need to opt-in to possible malicious cross-origin tools?
15:45:27 <anssik> Dominic: we try to match postMessage semantics, similar security model applies
15:45:52 <anssik> ... no need to specialize for this, filter inside getTools(), I think
15:46:10 <anssik> Reilly: I agree
15:46:41 <anssik> Dominic: we mirror the target origin, and give information who registered the tool
15:47:48 <reillyg> +1
15:47:59 <domfarolino> +1
15:48:01 <anssik> RESOLUTION: Adopt a postMessage() inspired security model for cross-frame tool enumeration, getTools() returns a flat list of tool objects with their source information. (issue #160)
15:48:02 <brwalder> +1
15:48:09 <anssik> Subtopic: Human in the Loop support for non-browser clients
15:48:12 <anssik> Anssi: issue #165
15:48:13 <gb> https://github.com/webmachinelearning/webmcp/issues/165 -> Issue 165 Human in the Loop support for non-browser clients (by MiguelsPizza) [Agenda+]
15:48:23 <anssik> ... Alex raised a question about how non-browser clients can support Human in the Loop interactions
15:48:45 <anssik> ... where HITL is interpreted as "the agent lives in the context of the browser", the agent could be browser's agent, extension agent, iframe agent
15:48:59 <anssik> ... the initial design is focused on HITL and as a consequence not on automation use cases
15:49:09 <anssik> ... we're seen emerging interest in automation use cases, including new implementation experience
15:49:22 <anssik> ... Cloudflare for example experimentally supports WebMCP in their headless Chrome product aka "Browser Run"
15:49:27 <anssik> -> https://developers.cloudflare.com/browser-run/features/webmcp/
15:49:48 <anssik> Anssi: Alex also presented a proposal how a 1P integration between WebMCP and a non-browser-native client such as Claude Desktop, Gemini CLI etc. might look like
15:49:51 <anssik> ... in this scenario, quoting Alex:
15:50:05 <anssik> ... "The user might be present but driving from outside the browser, so does this fall into our Human-in-the-Loop domain?"
15:51:04 <anssik> Alex: thanks Anssi, we see more and more browser automation use cases happen with browser harnesses, E2E testing, a use case worth looking
15:51:28 <anssik> ... it is difficult to do, if you want to be roped in at some point and get the user's input, not clear how to do that
15:51:36 <anssik> ... I've seen some deployments of this
15:51:51 <anssik> ... Cloudflare implementation provides an iframe that allows the user to interact
15:52:03 <anssik> ... would be great to have a programmatic solution to this
15:52:24 <anssik> Anssi: BrandonW asks "what the requestUserInteraction primitive would look like in the remote scenario?"
15:52:49 <anssik> BrandonW: an open question, seems to me the reasonable starting point would be similar to MCP elicitation API
15:53:17 <anssik> ... we have a similar arrangement, MCP Server does not have UI directly accessible to the user, and that's where elicitation comes in
15:53:48 <anssik> Dominic: is requestUserInteraction adequate for this if harnesses hook into e.g. Discord
15:54:19 <anssik> BrandonW: the way requestUserInteraction is spec'd, it is up to the page to present UI and collect that input in local input and do something with that information
15:54:44 <anssik> ... to make it useful in this scenario, requestUserInteraction would need to carry more information to the agent, know what UI to pop up
15:55:17 <anssik> Dominic: directly reaching out to the agent would  be one possible path
15:55:59 <anssik> BrandonW: requestUserInteraction is handled within the page, and agent does not need to know that interaction happened, interplay between the page and agent, could use the same code path
15:55:59 <anssik> q?
15:56:16 <anssik> Dominic: requestUserInteraction does not do back to the agent, is that the problem here?
15:56:59 <anssik> BrandonW: yes, to give an example, imagine having a single too, order a product, go to checkout process, form gets presented to the user in the middle
15:57:09 <bvandersloot> q+
15:57:11 <anssik> ... agent just gets the final result
15:57:26 <anssik> ... from the agent's point of view it is just a tool call with user intervention
15:57:42 <anssik> Dominic: if user interaction is required, the tool tells the agent what it needs
15:58:24 <anssik> BrandonW: Kushal shared, a page asking user attention too often can be user by malicious page, a more structure mechanism would help address this malicious use
15:58:26 <anssik> q?
15:58:36 <anssik> ack bvandersloot
15:59:32 <anssik> q?
16:01:13 <domfarolino> +1
16:01:13 <brwalder> +1
16:01:28 <AlexN> +1
16:01:30 <anssik> RESOLUTION: Form a concrete design for user interaction to satisfy non-browser client Human-in-Look use cases.
16:02:14 <anssik> Topic: Prompt API
16:02:18 <anssik> gb, this is webmachinelearning/prompt-api
16:02:18 <gb> anssik, OK.
16:02:41 <anssik> Anssi: I wanted the group to pay attention to the TAG review issues with recent activity
16:02:57 <anssik> ... and focus on issues that are both actionable and have a possible interoperability impact
16:03:01 <anssik> ... let's look at tag-tracker issues with activity in the last 60 days:
16:03:04 <anssik> -> https://github.com/webmachinelearning/prompt-api/issues?q=is%3Aissue%20state%3Aopen%20updated%3A%3E%40today-60d%20label%3Atag-tracker
16:03:34 <anssik> Anssi: all issues:
16:03:37 <anssik> -> https://github.com/webmachinelearning/prompt-api/labels/tag-tracker
16:04:08 <anssik> MikeW: thanks for bringing these topics, I hope we can have more discussion in upcoming meeting on this API
16:04:14 <anssik> ... we have made most progress on parameters
16:04:52 <anssik> ... finding categories to allow clients to find creative behaviours and allow implementers to align on implementations that fit that category with topK etc.
16:05:05 <bvandersloot> q+
16:05:16 <anssik> ... developers are interested in tuning parameters, so we took to heart TAG feedback on this design point
16:05:36 <anssik> ... we're done basic design and invited developer feedback to explore this categorical preset approach
16:05:49 <anssik> Anssi: this was issue #170
16:05:50 <gb> https://github.com/webmachinelearning/prompt-api/issues/170 -> Issue 170 [Tag Review] - Interoperable model parameters (by etiennenoel) [tag-tracker]
16:05:52 <anssik> q?
16:05:58 <anssik> ack bvandersloot
16:06:39 <anssik> BenjaminV: we had Jake in the room who wrote standards position for this API, the concerns around model tuning and parameter tuning, people designing for one particular model
16:07:49 <anssik> Anssi: I'd invite feedback to be provided in the spec
16:07:51 <anssik> -> https://github.com/webmachinelearning/prompt-api/labels/tag-tracker
16:08:23 <anssik> s/in the spec/in the spec repo
16:08:24 <anssik> q?
16:08:52 <anssik> RRSAgent, draft minutes
16:08:54 <RRSAgent> I have made the request to generate https://www.w3.org/2026/05/14-webmachinelearning-minutes.html anssik
16:09:53 <anssik> s/we've done/we've done
16:10:32 <anssik> RRSAgent, draft minutes
16:10:33 <RRSAgent> I have made the request to generate https://www.w3.org/2026/05/14-webmachinelearning-minutes.html anssik
16:12:02 <anssik> s/… we're/… we've
16:12:04 <anssik> RRSAgent, draft minutes
16:12:06 <RRSAgent> I have made the request to generate https://www.w3.org/2026/05/14-webmachinelearning-minutes.html anssik
16:12:51 <anssik> s/these topics/these topics for group discussion
16:13:54 <anssik> Present+ Brandon_Walderman
16:14:35 <anssik> s/Present+ Chris_Needhan/Present+ Chris_Needham
16:15:30 <anssik> s/Present+ Wenson_Ssieh/Present+ Wenson_Hsieh
16:15:55 <anssik> Present+ Sarah_Drasner
16:16:09 <anssik> Present+ Ehsan_Toreini
16:16:31 <anssik> Present+ Ali_Spivak
16:16:43 <anssik> RRSAgent, draft minutes
16:16:44 <RRSAgent> I have made the request to generate https://www.w3.org/2026/05/14-webmachinelearning-minutes.html anssik
16:18:22 <anssik> s/Ben_VanderSloot//
16:18:37 <anssik> RRSAgent, draft minutes
16:18:39 <RRSAgent> I have made the request to generate https://www.w3.org/2026/05/14-webmachinelearning-minutes.html anssik
16:21:11 <anssik> s/initiated/initiated to the spec, it
16:21:41 <anssik> s/Domenic/Dominic
16:22:18 <anssik> s/be signal/be to signal
16:23:49 <anssik> s/can from/came from
16:24:05 <anssik> s/he just made/we just made
16:24:27 <anssik> s/loosely applies/loosely applied
16:24:53 <anssik> s/bigger could/a bigger set
16:25:16 <anssik> s/of nuance/of nuances
16:25:50 <anssik> s/other peoples/other people
16:26:12 <anssik> s/get apply/get to apply
16:33:28 <anssik> i/Subtopic: Support cross-frame tool enumeration and composability while handling name collisions/RESOLUTION: Identify use cases for hints. (issue #176)
16:33:38 <anssik> RRSAgent, draft minutes
16:33:40 <RRSAgent> I have made the request to generate https://www.w3.org/2026/05/14-webmachinelearning-minutes.html anssik
16:35:28 <anssik> s/seen emerging/seeing emerging
16:37:25 <anssik> s/not do back/not go back
16:37:38 <anssik> s/single too/single tool
16:38:19 <anssik> s/user by/used by
16:39:01 <anssik> s/Human-in-Look/Human-in-the-Loop
16:39:04 <anssik> RRSAgent, draft minutes
16:39:06 <RRSAgent> I have made the request to generate https://www.w3.org/2026/05/14-webmachinelearning-minutes.html anssik
18:32:06 <Zakim> Zakim has left #webmachinelearning