Office Hours · 19 November 2026

OpenLint Office Hours #8

Summary

The eighth OpenLint office hours agreed that the specification should drive the toolbox, not the other way round, and that this is December’s priority, ahead of new toolbox features. A separate spec meeting and a chat space for the project are coming.

Status. Two toolbox releases this week: [email protected] split the rulesets into packages and added the opt-in schema cache from SAP’s Open Resource Discovery team, and [email protected] pinned the spectral:* compatibility rulesets at 1.0.0. The spec is at 1.0. A new roadmap label on openlint/openlint collects the bugs and housekeeping that are up next, and toolbox changes are on the changelog.

MCP server. Miguel Quintero’s @openlint/openlint-mcp offers five read-only tools: list rulesets, validate a ruleset, lint a document, list rules and explain a rule. Its docs are on Context7. The first outside pull request reports slow linting on large documents (62 seconds for a 2.7 MB OpenAPI document, past most clients’ 60-second timeout) and proposes a size limit. Miguel will merge it and then freeze until after apidays Paris.

Performance and document size. Kin will lead performance benchmarks, so that merges like the schema cache are measured, not just trusted. Mike Kistler pointed out that how split OpenAPI documents compose (where a referenced security scheme is resolved, for example) is still ambiguous and worth raising with OpenAPI. Kin sees rulesets describing different document “shapes” that organizations can choose from.

How the project is organized. The diagram on openlint.org shows the roles, repos and how changes land (community#41). Joost Farla asked for clear rules on who approves and merges pull requests. Each lead sets their repo’s rules first; Kin will propose a baseline (such as two reviewers) next week. Andre said it’s hard to follow what’s happening and how to contribute, and suggested dedicated sessions per area.

Communication. The project will get its own chat space with channels per area. The APIs You Won’t Hate invite failed for Mike. Kin will compare Slack, Discord and Matrix on cost and accessibility for a vote next week.

The spec drives the toolbox. Joost is finding behaviour the spec now states that neither OpenLint nor Spectral implements (alphabetical ordering, cycles through extends), so the conformance tests fail, and Spectral can’t be fixed from here. Kin’s position: the spec is authoritative, OpenLint fixes its bugs, and the project files issues on Spectral and vacuum. Today the spec is checked against the toolbox to keep the Spectral promise; once the spec is stable, the toolbox’s CI should fail when it breaks the spec. New features, such as the schema cache’s new option, should start in the spec. The pipeline spans the spec, conformance and the toolbox, with downstream builds like the MCP server. Joost will propose a time for a weekly spec meeting, possibly kicking off in person in Paris.

apidays Paris (1–3 December). Talks from Matthias Keckl, Kin, Dimitri van Hees, and Miguel (on API developer experience, Wednesday 2 December). Miguel is writing an article for apidays on why OpenLint. The logo concepts in community#47 are open for votes. Kin, Phil Sturgeon and Joost will join the Nordic APIs LiveCast in January. Next office hours, 26 November, is the last before Paris.

Agenda and discussion on GitHub · Watch on YouTube

Transcript

00:00:00Before the start: The recording opens with a short video Kin made for apidays Paris on API governance: get your OpenAPI, AsyncAPI and JSON Schema contracts in order; generate and validate MCP tools from them; index A2A cards and agent skills; publish an API catalog, APIs.json and llms.txt at the root of your domain; and lint all of it with OpenLint, which is growing from OpenAPI, AsyncAPI and Arazzo to MCP, A2A cards, skills and discovery formats.

00:02:18Before the start: Hellos and audio checks while people join. Kin asks Miguel about the lifecycle proposal on the OpenAPI side; Miguel hasn't checked it lately and will this weekend. Dimitri can't make it today. Kin pings Phil, who missed last week because of time zones.

00:04:50Kin Lane: Oh, cool, there's already a pull request on the MCP server.

00:05:00Miguel Quintero: Someone emailed me: they'd fired up the npm package, found a couple of issues, and were stoked about it. I told them to submit a PR.

00:05:27Kin Lane: I'll share my screen and we'll slow roll into this. We're slowly getting a little more formal about our approach, and more organized in how we track things. Sitting in on other spec meetings, the code of conduct comes up first, so I'm going to weave it into the opening: we have a code of conduct, it's in all the repos and on the website, to keep things organized and respectful as we get going.

This is the regular weekly office hours for OpenLint. Right now it covers the spec and the toolbox, which may change. We recap what's happening and then dive into a handful of issues. The previous week's recording, summary and full transcript are on the meetings site and YouTube, and linked from the office hours issue, which is always pinned on the community repo.

00:07:27Kin Lane: This week we had two releases: 1.2 and 1.3 of the tool. 1.2 split the packages and added the opt-in schema cache from Open Resource Discovery. 1.3, which came just a little while ago, pinned the spectral:* rulesets at 1.0. So the rulesets are pinned at 1.0, the tooling is at 1.3, and the spec is at 1.0. That's fairly clean, but we have three targets to pay attention to, and for me that's about communication: which ones are changing and where they're going.

00:08:34Kin Lane: The MCP server is out. Do you want to talk about it, Miguel?

00:08:40Miguel Quintero: Sure. I did some work last weekend and tested locally. All of the tools are read-only, just to get started. I briefly looked at the PR that just came in. They were complaining about document size and performance, which is good, I'll take a look. We'll see where the problem is; maybe it's in one of the core packages. If it's worth it, I may run a test where I download all of APIs.guru, thousands of APIs, and see what performance looks like. To give you an idea, they said 0.17 MB takes 2.7 seconds, and a 2.74 MB OpenAPI document takes 62 seconds. That's slow, and it times out at the 60 seconds most clients use.

But it's really cool. Once you plug it into Claude Code or OpenCode, you can ask "what rules are you using? I have this API, what ruleset should I use? Lint this document." It's amazing how well it works. So give it a try, send more PRs, and let it rip.

00:10:39Mike Kistler: This sounds great. What tools does the MCP server provide?

00:10:45Miguel Quintero: List rulesets, validate ruleset, lint document, list rules, and explain rule.

00:11:09Miguel Quintero: And I took the liberty of publishing the OpenLint MCP docs in Context7, so agents using Context7 get the latest documentation: "oh, you're going to lint a document, let's look over here." It needs a little improvement. Search for OpenLint there and you'll find it. We've got to work on performance, of course.

00:11:50Kin Lane: Thanks for the contribution. I've plugged it in but haven't played with it much, so I'll get you feedback. It also says we need to start on performance testing and benchmarks alongside conformance testing: is it the tool, is it MCP, is it the ruleset? And, like last week, where's our recommendation for how big an OpenAPI document should or shouldn't be?

00:12:35Miguel Quintero: That's right. In fact, the PR that was submitted suggests a maximum size, a new flag on one of the tools. I'll take a look.

00:12:45Kin Lane: This is one of those cutting-edge features that differentiates the toolbox from Spectral.

00:12:57Mike Kistler: You've mentioned a couple of times how big an OpenAPI document should be. Maybe this is for the OpenAPI folks, but there's been a lot of angst over how you split definitions into different files and the rules for how that works. Henry in particular has been working on cleaning that up. For example, if an operation is defined in another document and refers to a security scheme, where do you look for that security scheme? In that document, or back in the original? I think there's still some ambiguity about how OpenAPI documents compose. That would be worth identifying and giving feedback to the OpenAPI folks. I'd be surprised if all the problems are solved.

00:14:10Kin Lane: Agree a hundred percent. It's a tough area, and it's one of the ways I see rules playing: rules can define different shapes of what governance means, as optional packages. Part of the problem is what you described: people don't know how to assemble documents, and others know how and have their own opinions, and there isn't always a right or wrong way. I spent ten months arguing with Stripe and Twilio over how they generated their OpenAPIs. Stripe was adamant there should only ever be one OpenAPI document, the source of truth. I went through the same with Salesforce. You'll come up against architects with strong opinions. So rules that define different shapes: here's one approach, and these rules look for refs, structures and sizes in a certain way; here's another. That would be very useful, and then we bring it to OpenAPI, AsyncAPI and the other camps.

00:16:10Kin Lane: Changelogs. I'm trying to get better at communicating. I added changelogs to capture what we release with 1.2 and 1.3, and make sure it's labelled. I'm doing the same with the roadmap: I just created a roadmap label and applied it to the bugs Joost has been adding (sorry, I keep mispronouncing your name, and everyone mispronounces mine, so I'm hyper aware of it) and to Phil's housekeeping items.

00:17:00Kin Lane: I added two proposals this week, based on feedback from Joost's spec work. One breaks down the roles, how they go into governance, and the repo structure. The result is this diagram, which is on the homepage of the site now. I'm the acting administrator, the circus ringmaster. The toolbox lead is Phil. The specification lead is Joost. Tool leads are per tool: the MCP server is the first, with Miguel; if you develop a tool, you become its lead, with a certain level of responsibility. And it would be nice if each of us had a plan B, so the pressure isn't all on one person.

The spec is at v1 with a set of conformance tests, and Dimitri and Joost are working on the v2 draft. That's directly related to the core tooling, the parsers, the functions, and the toolbox releases and changelog, all working as a unit. Then the add-ons: the MCP server is the first package where things are getting real, with Miguel leading. There's the language server, which I'll probably lead, because Daniel from adorsys is really interested in going beyond VS Code, to IntelliJ. I'm going to try to get the person who created the VS Code extension to bring it into the org. All of these boxes are clickable.

Then the rulesets: we're versioning them. They overlap with Spectral and OpenAPI, so there are two sets: the ones we iterate on, and the frozen spectral:* ones that never change. There's an aggregate across them at rules.openlint.org, a searchable directory, but each has its own repo and npm package, and we'll add new rulesets the same way. Then the conformance site, which is wider than the spec's conformance tests: it's about tooling across the community, and there are wrinkles to iron out, as Joost has brought to the table. Learn is separate from Docs: Docs is the toolbox documentation, Learn is any related concept, so it's much bigger. And the main website, with the blog and the meetings.

And how changes land: discussions, issues, pull requests, releases. We have various org owners; if you want to be added, I'm happy to. As the leads take shape we'll lock things down: lock repos, limit who's an administrator, and I'll govern that. That's my response to Joost saying we need order in the household. I still want to keep things somewhat loose, but now we have a more formal way of talking about it, and I'll update this every week.

00:21:36Joost Farla: I really like it. For some topics we need more detail on the process: who has rights to approve a pull request, who merges, the creator or the lead maintainer. That's important to set up in the short term, because we're creating more and more issues and pull requests.

00:22:00Kin Lane: Agreed. Some of that will be defined per repo. On openlint/openlint you need multiple reviewers before you can merge; others don't have that. That should be on each lead to set in motion, strict or not, and I'll aggregate across them, look for common patterns, and note the differences. We'll review it every week: Phil, you set this for the toolbox; Miguel, on MCP you decide whether you need a separate reviewer and who you pull from; same for you, Joost.

00:23:15Joost Farla: But maybe we have to decide on a baseline.

00:23:20Kin Lane: Right now I want each of you to do it for yours, then I'll propose a baseline across them, and we can negotiate from there. Andre?

00:23:27Andre: One thing for me, and maybe I've not been engaged enough recently, is that I find it hard to follow what's happening here and there, and I can't find it in the issues. Also, in this session we discuss the general things, but the spec and the tool are big, important areas. There may be a benefit in a dedicated discussion every now and again, for the spec, say. I reviewed a couple of pull requests, and it's even hard to figure out how to contribute sometimes.

00:24:15Kin Lane: I agree. Half of this diagram is starting to pull that together: where you go to get involved in core work. These are clickable; the work should happen there. Discussions for general community things, issues per repo. I should probably put a bounded box around that, so you go to each area or start a discussion, and each area has its own communication. That's on openlint.org. How we aggregate the activity is what the changelog is for: a changelog for the toolbox, for the docs, none for the spec yet, so I'll add that, and the roadmap. That's all still coming together, and needs lots of focus.

There's also the aspect of coming regularly. I've experienced it with OpenAPI, Arazzo, AsyncAPI and MCP: if I don't show up regularly, I'm much more lost. And pay attention at the repo level: look at the issues, then the pull requests, and start chiming in. If there's a better way to surface this, aggregated on the README or somewhere else, let us know.

00:26:20Kin Lane: Back to the diagram. Under "how changes land" I'd add a block for how we govern and keep things stable: pull requests require two reviewers as the baseline, and the things Joost just asked for. I'll start from the current settings in the toolbox, the spec, the MCP server and the ruleset repos, how many reviewers each needs, and propose a baseline to talk about next week.

00:27:59Joost Farla: Maybe we also need a channel of communication between these weekly updates, something like a Slack team, so we can ping each other instead of waiting for the next office hours.

00:28:15Kin Lane: I'd recommended the APIs You Won't Hate OpenLint channel. We hesitated to set up another Slack because a lot of people have Slack fatigue. I haven't added it to the communication list on the website, but I'll add it. If we feel we need our own, we can. JSON Schema, AsyncAPI and OpenAPI each have their own.

00:29:00Kin Lane: Oh, and you had problems getting in, Mike, so maybe that's a roadblock.

00:29:10Mike Kistler: I don't know if it was specific to me, but I got one of the invite emails, and when I clicked the link it failed with a circular reference error or something like that.

00:29:25Kin Lane: I feel that. I've been trying to get into the A2A governance meeting for a month and a half, through the Linux Foundation. So maybe we should have our own, and branding-wise that might make sense. I could reach out to Phil and see why the invite failed.

00:30:00Joost Farla: The nice thing about a dedicated team is channels for specific topics.

00:30:10Kin Lane: Right, and that gets at what Andre was saying: one for the spec, one for the tool.

00:30:20Andre: Another question: is Slack actually the right place? Or Discord, or Matrix?

00:30:30Kin Lane: I haven't heard of Matrix.

00:30:35Andre: It's open source. I understand the issue with something like Matrix, which is new: yet another tool, another cognitive load. I'm just throwing the question out.

00:30:50Kin Lane: There are pros and cons to each. Let me research cost and accessibility, and we can vote next week, asynchronously, or both, and then fire one up. Slack tends to reach the widest audience; Discord tends to be younger. I struggle with Slack: my API Evangelist Slack went from nine dollars a month to five hundred dollars a month in July. But I agree with Joost: in the OpenAPI Slack there are channels for the spec, overlays, Arazzo, and that might help with your concern, Andre. I'll make suggestions next week.

00:32:16Kin Lane: The other proposal was on the spec: how to keep the conformance tests' source of truth in the spec and check them out as part of a pipeline in the conformance suite I use to test the wider tools. We're working through it. Anything you want to add, Joost?

00:32:40Joost Farla: I posted a reply yesterday, because I feel we need to reconsider the role of the conformance runner in the spec repo. I'm working through the open issues, and I'm finding unexpected behaviour in both current implementations, OpenLint and Spectral, which in my opinion are bugs more than new proposals. Like how alphabetical ordering works with numeric literals; I raised issues for that. Another new rule is that extends must not create a cycle, which is a no-brainer and should have been there already, but it never surfaced. So I'm making those aspects explicit in the specification, and now the conformance tests fail, because the bugs are still in the implementations. For OpenLint we can fix them; we should decide what comes first, the spec change or the implementation, or whether the spec waits for the implementation before it gets approved. But Spectral is the other one we test against, and we can't influence that. It will probably keep failing unless they follow our issue tracker and fix it too. We've created a dependency chain that makes the current process hard to work with.

00:34:30Kin Lane: Per what Andre was saying, you and I should probably start a spec meeting, weekly or on a schedule, and really deep dive into these issues. I'll leave it to Phil, Jakub and others to spearhead a separate tooling meeting if they want. This meeting stays the overview. My answer is that the spec should drive this. That's why we have a spec now: it should be authoritative and opinionated. Then it's up to all of us, with me leading, to take what you're finding and get it done in OpenLint. That's one of the trust-building things with OpenLint versus others: we're actively working to a standard. And we should drive not just OpenLint: I should submit issues on Spectral and bring issues to vacuum. The spec should be loud, opinionated and precise in driving this, in my opinion.

00:36:00Joost Farla: We probably don't have time to discuss this now, but let's take it to an issue, or Slack, or whatever.

00:36:10Kin Lane: I'll wrap my head around it and come up with a suggestion in a few days. Come up with a time for a meeting that works for you, because you're the lead on this. You lead on the time and space, make a suggestion, and we'll set up the chat channel too.

00:36:30Contributor of #39 [?]: On that topic, I was also wondering about the spec, and how it's handled. Normally the spec defines the requirements, and OpenLint implements them, so it's important to make clear that the spec drives this. I also noticed that my contribution's pull request was merged without any changes, and I'm not sure the spec is aware of that implementation. Thank you for trusting me, but that pull request was about performance; was it measured, or tested on your side, or was it more a trusted merge?

00:38:38Kin Lane: It was a trusted merge. Phil, who isn't here, went through it, and I left it to him: he's the toolbox lead, and he's the original developer, so I trust that he reviewed it. But, as I said last week and again this week, we don't have any performance benchmarks or testing to back that trust. We need someone to lead that, and I'm probably going to take it on, because we shouldn't blindly take things in. That's also part of what Miguel raised with the MCP server.

00:39:40Joost Farla: On the roles of the spec and the toolbox: because we started from a fork of Spectral, what we do now is have the Spectral fork and implementation, with a spec created next to it. The spec is still in draft; we're still figuring out a lot of details. Meanwhile we run automatic checks against the current OpenLint code base, to make sure the spec doesn't describe something that breaks the current implementation, because one of our promises is that every ruleset that works in Spectral today keeps working in version 1 of OpenLint. Eventually it must be the other way around. Once we have a stable version of the spec, the toolbox should have a CI step validating that it still conforms to the specification. That's part of what Kin and I are discussing about where the conformance runner lives and what role it takes. Eventually the toolbox build should fail if it violates the specification.

00:41:30Contributor of #39 [?]: Yes, if that chain is validated automatically. The main point of the spec is to define: a new concept, feature or optional flag is discussed as a pull request into the spec, in the theoretical way, the group decides it's fine, it's merged into the spec and set as a requirement for the toolbox, then implemented and checked there again. A pipeline that makes it reliable downstream. That's the approach I would prefer too. The performance contribution was an optional flag, not breaking anything, and I measured it locally on the projects where we found the potential, so it isn't hurting much. It's about having a structure for bringing new features in, from the spec first.

00:42:50Joost Farla: I hadn't seen that PR before, but now I look at it: it adds a new parameter to an existing function. You're right, that should be in the specification too, and it's missing now.

00:43:15Kin Lane: Everything you just described is why we forked this, but we still have to do the work to get to that pipeline. Until now, Spectral was a tool, and that's it: no spec, just a configuration file. Now OpenLint has a tool and a spec, but the tool is still wagging the spec, and we need the spec to wag the tool.

00:44:03Miguel Quintero: So who's working on the pipeline? We need a pipeline to validate backwards compatibility, so that OpenLint does 100% of what Spectral does today. That's the first green. Once we're solid, any PR can go through, but that original baseline can't break, because the MCP server depends on it as well. Then when the spec is ready, you add another step to the pipeline: green for the original stuff, then validate the new stuff. Is anybody working on that?

00:45:10Kin Lane: That's what's on the table. There's the spec pipeline, the conformance suite pipeline and the tool pipeline, and the pipeline in my opinion spans all three. It's a combined effort between me, Joost and Phil right now.

00:45:40Miguel Quintero: OK, that makes sense. I'm downstream, so any time there's a change to a core package, because you find a bug or whatever, it should kick off the MCP build before we release.

00:46:00Kin Lane: That's why Joost and I are focusing on this. And thank you for bringing up the performance testing, and that adding properties should go through the spec. This is a top priority. Joost, suggest a time and ping me; I'll invite Andre and others, bring it to the group. And we'll get a chat channel up to keep the conversation going. For the spec itself, the spec drives the pipeline to the tool, but backwards compatibility gets really muddled: if a bug is in both Spectral and OpenLint, and the spec says otherwise, and we can't get those changes into Spectral, it gets messy. We have to find a path forward, and the spec has to drive it. I wanted to wait until there was pressure before setting up another meeting, but the spec is getting pretty important now.

00:47:50Joost Farla: There's a discussion on GitHub if you have additional input on this.

00:48:00Contributor of #39 [?]: From the contributing side it would help to see: this exists in the spec; then, to extend something, start on the spec side with a snippet or a parameter, say we want it integrated into the toolbox, then go to the engine itself. A contributing pipeline.

00:48:40Kin Lane: Correct, we're working towards that, and we're fully aware there's confusion. That's why we forked Spectral: it didn't exist before.

00:49:06Kin Lane: Moving on to the toolbox. We've got fifteen minutes, so a little faster. Versioning: I think we've got it now. We need a meeting to get the spec 1.0 rubber-stamped, since we were loosely there. Phil got the 1.0 of the rulesets stamped and pinned, and the tooling has its versioning in place. Some of the spec issues are creating bugs, and I'll work to triage those with development on the tool. There are some housekeeping items in the queue. The docs have a small bug to reconcile before they build again. I'm organizing the next release considerations, including the performance work that already got merged, and I'll keep communicating what's next in the tool. This is more of the pipeline we just talked about: right now the tooling has been driving the spec.

00:50:30Kin Lane: We went through most of the spec items, and we'll spin off another meeting to go deeper. Anything on v2, Joost?

00:50:40Joost Farla: Dimitri and I are working on it; nothing to show yet.

00:50:45Kin Lane: And Joost and I will work out how conformance works at the spec level, the wider community tool level, and the OpenLint tool level.

00:51:05Joost Farla: And if there are comments on the specification's pull requests, feel free to comment.

00:51:15Kin Lane: It'll probably take a week or two to work out the rhythm of the spec meeting. Here, at the top level, we'll keep spec things high level and how they relate across things, and make sure people know to come to the spec meeting to dive deep into each issue, because there's a pretty massive list of spec issues aside from the ones you mentioned.

00:52:04Joost Farla: Maybe we can do the first kickoff session in Paris.

00:52:08Kin Lane: There we go, do it in person. It's in two weeks. And that's a good segue to administration. I went through how the project's organized; if you have feedback on that, let me know, and the chat channel is part of it. Rulesets for other specifications: I brought the repo to the OpenAPI group about helping maintain and give feedback on their ruleset. I mentioned it at the AsyncAPI and Arazzo meetings too, and I'll keep bringing it up. We'll keep our repos as the central place in our org. As we consider new formats, MCP and A2A, I'll bring those to those groups too, to drive governance interoperability.

00:53:00Kin Lane: December 1st to 3rd, many of us are in Paris. Matthias is talking, I have two talks on December 2nd, so three talks, and Dimitri and I have two on the 3rd. If there are any others, let me know; I'd like to include them. I have two press releases scheduled.

00:53:30Miguel Quintero: I have a talk too. It's not really on OpenLint; it's on API developer experience at a strategic level. But I'll definitely take out my Spectral slide and put in OpenLint. It's on Wednesday as well.

00:53:54Kin Lane: I'll add it. And Mark said he was talking to you about something else, an interview?

00:54:00Miguel Quintero: Good that you bring that up. He asked me to write an article on why to use OpenLint, how it differs from Spectral, and where we're heading.

00:54:15Kin Lane: Excellent. I'll add your talk and your article. I did a video short for them on governance and mentioned OpenLint.

00:54:30Miguel Quintero: Oh, and day one too.

00:54:35Kin Lane: Thanks, I forgot about day one. Day one is APIOps, and there's a lot. I'll make sure everyone's talks are referenced, because I want them in the press releases, even if it's just one slide.

00:55:20Kin Lane: Last thing: the evolution of the logo. This is what I quickly generated when I set up the org, and I'm playing with different incarnations of it, different ways of interlocking it with OpenAPI, Arazzo and others. They're separate approaches; vote and comment on each one. I started combining, putting the words in, then went out of the box away from the current logo, and started thinking about badging: if our logo, branding and stickers looked like what our conformance badges and buttons might be like, there might be some benefit. Comment, make suggestions, come up with your own. I'm not stuck on the brand, and I'm not a graphics person, so I'm completely dependent on Google Gemini for this. I'd rather use human beings than AI; I wish we had some graphics folks.

00:57:15Kin Lane: We're slowly communicating more, and we'll slowly get a marketing group, or a segment of this office hours. In January, Bill at Nordic APIs has invited us to do a LiveCast, and Phil, Joost and I will represent. I haven't done more funding work; I haven't had the time. We have one more office hours next week to prepare for Paris. There'll be lots of talks in Paris, and I think we can do some live work, so I'll bring my cameras and record any live working conversations, maybe stream them and invite people. So the weekly meeting in the first week of December might be disrupted, but there might be other opportunities. And depending on what Joost suggests for the spec meeting, we might change up the calendar that week or into December.

00:58:27Kin Lane: That's the list. More progress, especially on the spec front; thank you for pushing on the spec. The spec needs to start standing on its own legs now. I think the tools are ready, open for business; there'll always be bugs and housekeeping. But getting the spec and the pipeline driving the tool comes before we jump into the roadmap. I was going to propose we start adding features to the roadmap for the tool, but after hearing what you said, getting the spec house in order, the pipeline and conformance, should be December. Anything else before we call it?

00:59:45Miguel Quintero: I just put a comment in the chat. There's one PR on the MCP server; I already replied. I think it's a good PR and I'll probably merge it, and then freeze: I won't merge anything else before the conference, so you can go crazy with demos. I don't want to introduce anything new.

01:00:11Kin Lane: Sounds good. I appreciate you all, thanks for showing up.