14:49:10 RRSAgent has joined #webmachinelearning 14:49:14 logging to https://www.w3.org/2026/06/11-webmachinelearning-irc 14:49:14 RRSAgent, make logs Public 14:49:15 please title this meeting ("meeting: ..."), anssik 14:49:23 Meeting: WebML CG Teleconference – 11 June 2026 14:49:37 Chair: Anssi 14:49:41 Agenda: https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-06-11-cg-agenda.md 14:49:45 Scribe: Anssi 14:49:55 ScribeNick: anssik 14:49:58 Present+ Anssi_Kostiainen 14:50:03 RRSAgent, draft minutes 14:50:05 I have made the request to generate https://www.w3.org/2026/06/11-webmachinelearning-minutes.html anssik 14:56:56 brwalder has joined #webmachinelearning 14:57:31 Present+ Brandon_Walderman 14:59:29 Present+ Victor_Huang 15:00:19 Present+ Benjamin_VanderSloot 15:00:42 benvds has joined #webmachinelearning 15:00:55 Present+ Mike_Wasserman 15:01:13 kush has joined #webmachinelearning 15:01:18 present_ 15:01:22 present+ 15:01:29 Present+ Sam_Bhattacharyya 15:01:54 Present+ Reilly_Grant 15:02:11 Present+ Cathy_Li 15:02:37 present+ Thomas_Steiner 15:02:38 alispivak has joined #webmachinelearning 15:03:08 Anssi: as a reminder, we'll use IRC-based queue management in this meeting: 15:03:12 -> https://www.w3.org/guide/meetings/zakim.html#speakerqueue 15:03:12 Julia has joined #webmachinelearning 15:03:25 Anssi: to suggest agenda topics, use Agenda+ label, e.g.: 15:03:29 -> Agenda+ https://github.com/webmachinelearning/webmcp/labels/Agenda+ 15:03:31 Present+ Benjamin_VanderSloot 15:03:38 Anssi: please welcome our latest new participants 15:03:42 ... Severin Ferrand from Google 15:03:52 ... Tyler Duprey, Neelakandan NC and Sam Bhattacharyya joining as individual contributors 15:03:57 msw has joined #webmachinelearning 15:03:57 ... welcome! 15:04:04 Topic: Announcements 15:04:08 Present+ Mike Wasserman 15:04:12 Subtopic: TPAC 2026 15:04:13 iris has joined #webmachinelearning 15:04:15 sambha has joined #webmachinelearning 15:04:19 Anssi: TPAC 2026 is in Dublin, Ireland on 26-30 October 2026 15:04:28 ... during that week, the W3C groups gather to resolve challenging issues and discuss future directions for the web platform 15:04:33 ... TPAC group meetings take place Mon, Tue, Thu and Fri 15:04:34 cpn has joined #webmachinelearning 15:04:36 ... Wed is for breakout sessions and social events 15:04:38 present+ 15:04:42 ... I believe this Community Group wants to meet during the TPAC week 15:04:46 ... in the past we have met on Tuesday 15:04:56 ... assuming also other groups stick with their earlier schedules, that is my default preference to avoid conflicts 15:05:00 ... but I'm open to other suggestions 15:05:04 ... some of the groups have shared their schedule preferences already: 15:05:08 -> https://github.com/w3c/tpac2026-meetings/issues 15:05:22 Anssi: TPAC 2026 website is not yet live, I'll share more information when it becomes available 15:05:33 Anssi: questions, comments? 15:05:33 BenGreenstein has joined #webmachinelearning 15:05:54 q? 15:06:29 johannhof has joined #webmachinelearning 15:06:58 Dominic: I don't have a TPAC plan yet, Tuesday sounds good 15:07:15 Topic: Prompt API 15:07:19 gb, this is webmachinelearning/prompt-api 15:07:19 anssik, OK. 15:07:23 Subtopic: Model neutrality 15:07:31 Anssi: issue #169 15:07:32 https://github.com/webmachinelearning/prompt-api/issues/169 -> Issue 169 [Tag Review] - Model selection and availability (by etiennenoel) [Agenda+] [tag-tracker] 15:07:40 ... we will review and discuss the proposed solution to the model neutrality issue 15:07:50 ... this was one of the two issues Mozilla brought to the group's attention 15:08:02 ... Reilly proposed a solution to address the model neutrality issue by requiring implementations to use either: 15:08:09 AlexN has joined #webmachinelearning 15:08:19 Simba has joined #webmachinelearning 15:08:41 Reilly: high-level point, when an API required a model, implementations that don't have their own they should be able to use the platform provided model 15:08:50 ... when there's an open source model implementation can use that 15:09:41 Benjamin: Jake raised a point that models often come with T&Cs 15:10:00 ... having T&Cs tied to the Web API was one of the Mozilla's standards position concerns 15:11:25 Reilly: IANAL, but I think the goal of using open-source models is to help with T&Cs 15:13:48 q? 15:14:06 Tom: should we mention open-weights explicitly? 15:15:37 ryan-roemer has joined #webmachinelearning 15:15:48 RESOLUTION: Prompt API specification to require conformant implementations to use a permissively licensed model provided either by the implementation or the underlying platform. As further work, determine what criteria makes sense exactly for such models. (issue #169) 15:15:57 Topic: WebMCP 15:16:01 gb, this is webmachinelearning/webmcp 15:16:02 https://github.com/webmachinelearning/prompt-api/issues/169 -> Issue 169 [Tag Review] - Model selection and availability (by etiennenoel) [Agenda+] [tag-tracker] 15:16:02 anssik, OK. 15:16:08 Subtopic: WebMCP wide review 15:16:15 Anssi: I've kept the horizontal group chairs up to date with our progress 15:16:29 ... for privacy review specifically there was a suggestion for this group to highlight to the reviewers specific questions or technical issues 15:16:36 ... for example: "what does Privacy WG think about X approach to Y" 15:16:46 Victor has joined #webmachinelearning 15:16:57 ... "does anyone in Privacy WG have experience or suggestions for how we could achieve X in a privacy-preserving manner” 15:17:21 ... this helps focus reviewers' attention to the most important issues, and also helps to get more specific feedback from the reviewers 15:17:35 Anssi: I called for a final review of the updated materials before we submit for the wide review, to ensure we have consensus on the content of the submission 15:17:42 -> Explainer https://github.com/webmachinelearning/webmcp/blob/main/README.md 15:17:51 -> S&P Considerations https://webmachinelearning.github.io/webmcp/#security-privacy 15:17:58 -> S&P Questionnaire https://github.com/webmachinelearning/webmcp/pull/195 15:17:59 https://github.com/webmachinelearning/webmcp/pull/195 -> Pull Request 195 Add security and privacy questionnaire (by victorhuangwq) [Agenda+] 15:18:06 Anssi: I believe the updated explainer is in a good shape 15:18:11 ... similarly, S&P considerations section is in a good shape 15:18:17 ... S&P Questionnaire seems to be soon ready to be merged as well 15:18:28 Anssi: for the S&P Questionnaire PR, I'd like to check the few open discussions there: 15:18:44 -> 07. Do the features in your specification expose information about the underlying platform to origins? https://github.com/webmachinelearning/webmcp/pull/195#discussion_r3375026898 15:18:54 Anssi: Johann, Benjamin? 15:19:02 q? 15:20:24 Johann: I don't think I have a strong position on this, it is channel similarly to other APIs, would be appropriate to call the API out 15:20:50 Benjamin: agree with Dominic's point 15:21:23 ... this is just part of browsing with this type of user agent, does not change the threat is a potential approach 15:22:23 Johann: theoretically could create a deterministic script that uses WebMCP to input to the tool, don't think this should block 15:22:37 Benjamin: agree this should not block the PR 15:23:47 Johann: also want to avoid make this look like this would be a novel concern with the API 15:24:12 -> 09. Do features in this specification enable access to device sensors? https://github.com/webmachinelearning/webmcp/pull/195#discussion_r3375059500 15:24:15 Ehsan has joined #webmachinelearning 15:24:16 q+ 15:24:21 ack Victor 15:24:45 Victor: both 07 and 09 questions are pretty similar 15:26:11 ... my thought on this is that, I'd try to answer the questions in a way they are intended to be answered to 15:26:39 Johann: happy to resolve as "no" and say user agents can already expose such information to websites 15:26:44 q? 15:26:57 Anssi: with those two discussions settled, we can merge #195 15:26:58 https://github.com/webmachinelearning/webmcp/pull/195 -> Pull Request 195 Add security and privacy questionnaire (by victorhuangwq) [Agenda+] 15:27:22 Anssi: and execute the resolution from our last meeting to stage the review requests for final look before submission 15:27:41 Victor has joined #webmachinelearning 15:27:54 ... the spec editors are free to stage those review requests at their discretion, but I propose we do it in a timely manner to not delay the start of the review process 15:27:58 ... the editors can grab the markdown templates from: 15:28:02 -> TAG review request ("Early design/incubation review") https://github.com/w3ctag/design-reviews/issues/new?template=000-incubation-review.yaml 15:28:06 -> Privacy review request template ("Request review for another type of spec") https://github.com/w3cping/privacy-request/issues/new?template=request-review-for-another-type-of-spec.md 15:28:11 -> Security review request template ("Request review for another type of spec") https://github.com/w3c/security-request/issues/new?template=request-review-for-another-type-of-spec.md 15:28:47 Anssi: specifically in the TAG's "You should also know that..." and Privacy and Security WG's "Other comments" sections are expected to highlight to the reviewers specific questions or technical issues we want feedback on, as I mentioned earlier 15:29:07 ... any questions? 15:29:20 q? 15:29:45 Brandon: the ask is to stage the templates in our WebMCP repo and when group has consensus submit to TAG, Security and Privacy WGs 15:30:48 q? 15:31:03 domfarolino has joined #webmachinelearning 15:31:12 RESOLUTION: The group approves the updated review materials (explainer, S&P considerations, S&P questionnaire). The spec editors will stage the review requests in WebMCP repo for the TAG, Privacy WG and Security WG, ensuring that specific questions or technical issues are highlighted to the reviewers in the appropriate sections of the review requests. (issue #195) 15:31:33 Subtopic: Human in the Loop support for non-browser clients 15:31:39 Anssi: issue #165 and PR #204 15:31:39 https://github.com/webmachinelearning/webmcp/pull/204 -> Pull Request 204 Replace requestUserInteraction with requestUserInput (by bwalderman) 15:31:39 https://github.com/webmachinelearning/webmcp/issues/165 -> Issue 165 Human in the Loop support for non-browser clients (by MiguelsPizza) [Agenda+] 15:31:57 ... earlier we resolved to form a concrete design for user interaction to satisfy non-browser client Human-in-the-Loop use cases 15:32:18 ... Brandon contributed a proposal for requestUserInput() "interactive", "form" and "url" modes inspired by the MCP Elicitation spec 15:32:27 ... this replaces the current requestUserInteraction() 15:32:37 ... these three modes are designed to cover a range of Human-in-the-Loop use cases for non-browser clients 15:32:48 ... Alex seems to be in support of the three modes 15:32:58 ... Dominic suggests in PR #204 that Chrome is interested in supporting the "interactive" mode initially 15:33:46 Brandon: interactive mode is still for browser client even if this was filed for non-browser clients 15:34:10 ... consensus to focus on these three modes in some order, and "interactive" seems to be the first priority 15:34:25 ... questions what happens for in-page agents 15:34:48 ... when user input is called, what's not clear is how an in-page agent gets a notification when UA is called 15:35:08 ... those are the simplest, page registering the tool takes input and fullfills the request 15:35:29 ... "form" and "url" are more complicated, the agent is responsible for rendering the response and pipe it back to the tool 15:35:44 q+ 15:35:47 ... my latest comment I posted a few mechanisms asking for feedback from group 15:35:50 ack domfarolino 15:36:25 Dominic: I was originally more into "form" mode that keeps the user interaction in UA 15:37:02 ... I think I've become more sceptical of this one, in fully remote flows, browsing happening in the server, agent UI connected to e.g. Codex 15:37:41 ... most of the tools are equipped for local or remote, and making websites aware of this complicates this somewhat 15:38:04 ... driving things via VM somewhere, requires input from the user 15:38:30 ... need to expose this via an agent harness 15:38:40 q? 15:39:11 q+ 15:39:11 q+ 15:39:21 Dominic: trying to keep the browsing use case in mind, while not stifling the remote use cases 15:40:10 ... critique with "form" and "url" modes is the tool can already choose to navigate already to drive interaction on a different site 15:40:43 q? 15:41:08 ack brwalder 15:41:43 Brandon: I do see the critique of the "form" mode, I have received similar feedback from others, agents optimized for browsing the web, they need how to solicit information from the user 15:41:59 ... agents have their own way to get this information 15:42:50 ... the "url" mode, however, in local browsing scenario the web page can open a new tab or a new window, but "url" mode helps when running on a server, user interacting indirectly via a mobile chat app, for example 15:43:44 ... "interactive" mode first priority, "url" mode second 15:43:45 q? 15:43:47 ack AlexN 15:44:20 Alex: would be helpful to talk why elicitation is in MCP spec, the reason is security, secure credentials you put in the agent 15:44:52 ... in elicitation, you input credentials in a form, not into the agent's context 15:45:05 q+ 15:45:09 ... "url" solicitation makes sense 15:45:17 ack brwalder 15:45:48 Brandon: about security, in MCP security, "url" is for sensitive flows as Alex explained, flows that do not go via agent 15:46:23 ... for "form" elicitation, however, the frontend agent renders HTML and the form collects information and passes that back to the tool, never go via the LLM's context 15:46:42 ... the tool listening for information, the agent can see this information even if the LLM does not 15:47:26 ... from tool server's point of view the "form" does not resolve all security issues, however, "url" solves the sensitive information issue 15:47:50 Dominic: "form" makes sense for MCP tools because they don't have for forms similarly to HTML 15:48:08 ... our version of "interactive" makes sense as an analog to MCP's "form" mode 15:48:29 ... I see value in giving "url" back to the user 15:49:12 ... in the remote case, if the tool says I need to type in information into the form, I trust the agent's UI 15:49:31 ... we don't need to give URL back to the user, but navigate to a different tab 15:49:32 q? 15:49:49 q+ 15:50:09 q+ 15:50:17 Dominic: I see value in having the user visit something outside of the space where agent is operating 15:50:22 ack AlexN 15:50:43 Alex: good point, originally I thought about Cloudflare's implementation 15:51:07 ... guidance how to achieve how they are approaching the issue would be helpful 15:51:08 q? 15:51:55 Alex: if there was URL once visited to load credentials, surfaces as a regular solicitation, maybe we don't need a separate channel to send the URL through 15:51:57 q? 15:51:59 ack brwalder 15:52:43 Brandon: one possibility to add, say we add "url" mode, when the tool requests, this is a suggestion to the browser that can decide based on context whether to open the URL if it knows the user has in-browser agent open 15:53:00 ... if this is remote agent, then forward the request to the remote client to open the URL 15:53:54 Dominic: the browser does not always know an agent is driving the experience to be able to pass back 15:54:24 q? 15:56:03 +1 15:56:50 +1 15:56:55 +1 15:57:01 RESOLUTION: The group agreed to initially specify the "interactive" mode in PR #204 and will continue discussion on "form" and "url" modes. (issue #165, PR #204) 15:57:08 Subtopic: Register agent skills 15:57:11 Anssi: issue #161 15:57:11 https://github.com/webmachinelearning/webmcp/issues/161 -> Issue 161 Proposal: Skills — workflow-level context for tool composition (by Idan-Levin) [Agenda+] [backlog] 15:57:19 ... this is a proposal to let pages register skills 15:57:31 ... "tools can tell agents what a site can do, but skills tell agents how to do it well." 15:57:37 ... strawman proposal navigator.modelContext.registerSkill() 15:57:42 ... independent of registerTool() 15:57:59 ... registerSkill() takes as input a dictionary with name, description, instructions, tools, context, annotations 15:58:21 ... an experiment by Andre demonstrated you can also create the skills-as-tools by registering each skill as a zero-argument tool 15:58:37 Anssi: Andre suggests there's value on helping agents to differentiate skills from tools 15:58:43 ... David points out a tool with readOnlyHint could do the same 15:58:48 ... how does the group prefer to proceed with this, any comments? 15:58:54 q? 15:59:37 Dominic: Chrome thinks tools could allow orhestration, we don't know how granular tools will be written, or whether we need this orchestration layer 15:59:58 ... would like to get feedback from OT on this 16:00:27 ... potentially a lot of value in such an explicit skill registration mechanism 16:00:37 q? 16:01:00 Subtopic: Manage agent memory 16:01:04 Anssi: issue #29 16:01:05 https://github.com/webmachinelearning/webmcp/issues/29 -> Issue 29 Allow developers to manage Agent's memory (by khushalsagar) [Agenda+] 16:01:09 ... an issue opened a while ago by Khushal based on Nitin's idea 16:01:21 ... MCP resources can provide contextual information to agents beyond current DOM data 16:01:29 ... this enables access to implicit application memory such as shopping cart items or other user-relevant context 16:01:38 ... the first question to the group is should WebMCP allow developer's to manage agent's memory? 16:01:53 ... if there's interest, I'd welcome contributions to the issue on specific use cases and design proposals for how this memory management could work in practice 16:01:53 q? 16:02:11 Khushal: this was filed early on, not actively working on this right now 16:02:30 ... a lot of this can be represented by tools 16:02:56 ... tools such as "get cart items" 16:03:30 ... prefer to look how developers use this, learn via OT feedback, and if developers indicate interest we can look at a proper abstraction for this feature 16:03:57 RRSAgent, draft minutes 16:03:58 I have made the request to generate https://www.w3.org/2026/06/11-webmachinelearning-minutes.html anssik 16:04:54 Present+ Ali_Spivak 16:05:10 Present+ Ben_Greenstein 16:05:36 Present+ Chris_Needham 16:05:44 Present+ Christian_Liebel 16:05:53 Present+ Dominic_Farolino 16:05:59 Present+ Ehsan_Toreini 16:06:06 Present+ Johann_Hofmann 16:06:14 Present+ Iris_Johanson 16:06:19 Present+ Khushal_Sagar 16:06:48 Present+ Sam Bhattacharyya 16:07:12 Present+ Shubham_Gupta 16:07:44 RRSAgent, draft minutes 16:07:46 I have made the request to generate https://www.w3.org/2026/06/11-webmachinelearning-minutes.html anssik 16:08:08 Present+ Alex_Nahas 16:08:30 Present+ Ryan_Roemer 16:08:59 Present+ Guru_Singh 16:09:04 RRSAgent, draft minutes 16:09:05 I have made the request to generate https://www.w3.org/2026/06/11-webmachinelearning-minutes.html anssik 16:09:21 s/present_// 16:11:34 s/either:/either a model with a permissive license or a model provided by the underlying platform 16:11:54 s/required a/requires a 16:16:02 s/are expected/we are expected 16:23:27 s/my latest/in my latest 16:24:27 s/navigate already/navigate 16:25:03 s/need how/need to know how 16:26:47 s/have for forms/have forms 16:27:32 s/surfaces/surfaced 16:30:13 s/developer's/developers 16:31:06 s/use this/would use this 16:31:24 RRSAgent, draft minutes 16:31:25 I have made the request to generate https://www.w3.org/2026/06/11-webmachinelearning-minutes.html anssik