Office Hours · 29 October 2026

OpenLint Office Hours #5

Summary

The fifth OpenLint office hours took stock of a project that now feels open for business, and turned to how rules are written, organized and shared.

Status. OpenLint has an npm package, a docs site for the tools (docs.openlint.org) and a spec site (spec.openlint.org). Miguel Quintero’s TypeScript 7 upgrade has merged. The Dutch government team is moving its tooling from Spectral to OpenLint and announcing it on its blog, and Kin has switched his own tools. The aim is an official launch at API Days Paris, with a few migration stories behind it.

Rulesets and education. Jakub Rożek proposed moving the built-in rulesets out of the main repo, because Spectral’s tests were coupled to them and small rule changes broke unrelated tests. Joost Farla asked for a standard way to organize rulesets across several standards and versions, and for modular, reusable parts, like JSON Schema references. The group’s biggest friction is writing the regular expressions and JSONPath for a rule; Dimitri’s example was a servers rule that also has to handle response bodies and callbacks. The plan is to start lean: publish common snippets on openlint.org, document how rules, inheritance and aliases work today, and grow that into an education section and, later, a registry of rules, aliases, regex and JSONPath that the spec and skills can reference, ideally co-owned with standards bodies such as OpenAPI, AsyncAPI, Arazzo, PSD2 and FHIR.

Agents. Agents will write most rules, so the knowledge has to be published where models and tools can find it. Miguel will look at getting OpenLint into Context7. Dimitri described the Dutch “validator in the loop”, where a skill lints the ruleset an agent writes, which fits the MCP server (#13). Joost described an AI-ready standards project in the Netherlands.

Editors and roadmap. People are waiting on the VS Code extension. Dimitri and Joost offered to look at a language server, alongside Ivan’s extension. Markdown and agent-skill linting is a top roadmap item. Dimitri will compare vacuum’s rules with OpenLint’s.

Spec and funding. The working draft is in good shape. Kin has small PRs open, and Joost will iterate next week. Open Source Europe is the fiscal host, with $1,000 in the account. Kin is preparing Sovereign Tech Fund and other European grant proposals, including the test harness and education, and asked for core-tooling ideas.

Agenda and discussion on GitHub · Watch on YouTube

Transcript

00:00:00Before the start: Small talk while people join: winter arriving in New York, and the clocks changing in Europe a week before the US.

00:02:02Kin Lane: I think we can get rolling and let others join. Phil is fighting other battles, so I'm not sure he'll make it today. I feel like we're right at the edge of this being real and open for business. We have an npm package, a docs site for the tools, a spec site, we meet weekly, and I've switched all my tools over to it without any major problems. How's everyone else feeling?

00:02:49Dimitri [?]: We're in a similar process of migrating all our stuff to OpenLint, with no issues so far. We'll probably release our upgraded tooling today or tomorrow.

00:03:09Miguel Quintero: I wrapped up TypeScript 7, the minimal suggested version. Jakub and Phil gave me a hand yesterday and the PR was merged, so we're at the latest and greatest as far as TypeScript is concerned, and ready to move on to business features rather than TypeScript features.

00:03:51Kin Lane: We've got some tidying up to do continually, but I think we can start working on the roadmap. I went through the spec docs again and found a couple of little things. I kept working through Joost's issues; there's a long tail of maybe ten I still haven't got my head around. But the spec site is great. I feel like we have a spec now, even with all the problems on the table.

00:04:36Joost Farla: I've seen the review requests, and I'll get to them in the next few days.

00:04:50Kin Lane: Jakub, anything from you and Phil, or Miguel, from the tool side? Concerns, thoughts, work on the table?

00:05:05Jakub Rożek: One decision to make is whether to keep the rulesets in this repo or outside. Not the rules themselves, but the built-in ones: OpenAPI, AsyncAPI and so on. Phil created an issue for it in OpenLint. It doesn't really impact users, but I'd lean towards outside, because it makes it easier for people to author rules. In the past, some of the tests in Spectral relied on those rulesets, so if you tweaked or removed an OpenAPI rule, a suite of unrelated tests started to fail. The tests were coupled to the core rules. It wasn't a big issue, but it was annoying when you just wanted to fix a typo in a description and suddenly had several tests to update. That's the motivation for splitting it out. It's something Spectral should have done in the past, and we should do it now that we have our own project.

00:07:32Miguel Quintero: All my PRs have merged; thank you, Jakub, for your feedback. I've started looking at the feedback on the MCP issue, and I'll look at it more closely. Next, Jakub or Phil, maybe you can help me work out how the MCP server takes advantage of the rulesets. Jakub, did you see my issue 13, with my initial thoughts on where the rules should live and how they're managed so the MCP server can access them?

00:08:27Jakub Rożek: I looked at it but haven't replied yet; I will. I'd probably follow more or less what the extensions do. I'm not sure the VS Code extension still works, and we had a discussion about moving it under the OpenLint umbrella. The interface isn't for humans, but the behavior, how rules and everything work, should use the same mechanism underneath. I'll leave a note in the issue.

00:09:26Kin Lane: Take your time. I want to spend more time understanding the current state of how rules are managed, how they're part of the testing, how the rules builder will work, how the core tools use them. Then how you, Dimitri and Joost, manage your rules. And I want to interview a couple of enterprises, Ford and a couple of banks and insurance companies, about how rules management should work aside from the tooling: here's what we have with the tools and how rules are managed, here's how different businesses are doing it, and a way of visualizing and understanding that. Dimitri, where's the pain for you with rules?

00:10:46Dimitri: We don't really have pain with rules at the moment. The rules are in a YAML file and they're part of the official government standard. That's a different pain, how standards are managed within the government, which isn't very agile. It's quite old-fashioned, and we can't solve it in this group. It's also why I was so happy we finally upgraded to OpenAPI 3.1, officially announced yesterday; it took ages. For example, there's a rule that every server should start with https, and the regular expression was written without taking into account that servers could also be a property of a response body. We want to fix the regular expression, but they aren't very eager on patching formal standards. In terms of OpenLint rules, we don't really feel pain yet, perhaps only the custom-rule part. Joost?

00:12:41Joost Farla: I wouldn't call it pain, but we've built a lot manually around Spectral. For example, how we organize rulesets: we lint documents against several standards, and every standard has several versions, draft and stable. We built a system to organize the rulesets. What I'd like as a future feature is something like a ruleset catalog, a standardized way to organize your rulesets, so it can go into the standard tooling and we don't have to do it manually any more.

00:13:49Kin Lane: There were a couple of threads in the discussions about naming files to help with organization. We need to be thoughtful and cautious about adding properties, but we could think about extensions to help with this. I'm going to step back and map how it currently is, so I can communicate it, because I don't think people understand composition and inheritance. I still stumble on it: what's the default set, how do you extend it, how should you build your rulesets, how should you name them, where should you store them. I have my own practices and I don't think they're all sound. So I need to document what the tool currently supports, build from there, create educational resources, and then talk about the roadmap.

00:15:11Joost Farla: It would also be cool to have a more modular approach, like JSON Schema, where I can use references to parts of rules. Ideally, the OpenAPI Initiative could maintain at least some snippets: if you want to address the servers part, they provide the regular expression, and it can be reused and shared, so you don't reinvent the wheel. That affects the spec itself, of course. Right now, the Dutch government API Design Rules only specify errors. In our own tooling we also want to give warnings for best practices that aren't part of the official standard, which is political, but we can provide them in our checker. And for upcoming versions, we use a generator to create rulesets for our checker that add warnings saying "watch out, this becomes an error in the upcoming version of the API Design Rules." It's not easy to extend those things at the moment, so we made a custom generator. It works. Perhaps it's possible already, but then it needs more documentation.

00:17:49Kin Lane: I like the approach of standards maintaining their own. We'll always have to maintain it in the beginning, with some governance, but getting OpenAPI to help own the rules for all of their properties, down to the regex, would be good, and the same for Arazzo, Overlays and AsyncAPI. The same for industry ones, PSD2, FHIR: own it ourselves to start, then coordinate with the standards bodies to help own it.

00:18:35Dimitri: We can also start lean and mean: a page on openlint.org with the most used snippets, regular expressions or JSONPath parts. If you want to address all the property names in a schema, use this snippet. There was a page on the Stoplight site with some of these, but it didn't always work. Perhaps that's the education or documentation part, but we could really easily improve the developer experience of building rules yourself.

00:19:25Kin Lane: Agreed. I'd like to see this in the education part, linked from the spec. As part of my search engine I'm already starting to gather standard rules and aggregate the most used regex and JSONPath, keeping the references, so they point back to Adidas or others, and seeing which companies just use the default ruleset. Alignment with each standards body will be important too. I'll document the baseline first, because that's the first part of the education we have to get out of the gate: what's the current state, what's the default set. When I look at companies' pipelines, most just use the default set. They haven't extended it, and I suspect that's because people don't understand why and it's not easy for them to do.

00:20:53Miguel Quintero: Maybe one of the top three points of friction is what Dimitri mentioned, the regular expression. When I started hacking with Spectral, my biggest problem was: here's my style guide, I'm codifying it in Spectral, what's the regular expression for this? I found some examples on the internet, but a collection of those regular expressions on the learn site or the spec would help. And if you're in the OpenAPI world, we can point people there.

00:21:47Dimitri [?]: And aliases will help here: a standard alias for all server objects, in this case.

00:22:05Miguel Quintero: Good idea, but where would that go? Now we're talking scale, like an alias registry. Where does it go, on the learning side of OpenLint? Or are there basic sets enabled by default in the tool, a default package? Because that's the thing: you're writing Spectral rules and go "what's the expression?" That's one of the bigger problems. Times have also changed; people don't write regular expressions any more, or almost don't write code. If an API program were starting from scratch, I'd take their PDF or wiki, feed it to an LLM and say "write all the OpenLint rules for this." But the LLM would have to know where to infer those regular expressions from, and for that to happen it has to be published somewhere, on the learn or spec site, so that when models get trained a couple of months from now they find it.

00:23:43Dimitri: This is our actual problem, on screen now. First we wanted to address the OpenAPI servers property. Then we found that our own API, a collection of APIs, has a servers property on every API resource in the response, and it affected that as well. We fixed that, and then we found that callbacks in OpenAPI can also contain servers. I wouldn't want to find that out myself. It would be nice to say: if you're using this OpenAPI version and want to do something with servers, use this snippet.

00:24:47Kin Lane: And this is where aliases come in: an alias that abstracts this away, so you'd just say "servers". Back to what Miguel was saying: we've got to document the base state of rules. Then there's how you create rules in the spec, then an education section on how to create rules, and then a rules registry that the spec and the education reference, with a base URL like rules.openlint and a URL structure others can reference in their docs and stories. The same for aliases. I need to document the current state, because to be honest I knew aliases existed but I've never used them, and I never saw documentation on how they work in Spectral. So: document the baseline, an education section on aliases, and an alias registry. The same with JSONPath and regex. We need a canonical place to learn about all of these.

00:26:23Joost Farla: I fully agree, and I'm glad you mentioned the alias registry. A registry with predefined rules helps people create better APIs. But if I want a different rule, say every server starts with ftp, which probably wouldn't be a default rule in the registry, I'd still like the flexibility to make my own rules while reusing the aliases, or at least the regular expressions or given snippets.

00:27:23Kin Lane: And having regex and JSONPath tooling on our education site. I have little tools I use regularly to derive JSONPath. There's an important difference between JSONPath and JSONPath-Plus in the libraries you use. It took me a while to realize, and it's a big complaint people have about JSONPath. So built-in tooling for building regex and JSONPath, as well as the libraries, will be pretty critical.

00:28:01Miguel Quintero: And a good skill available, because I don't think many people will write those rules by hand.

00:28:14Kin Lane: That begs the question: do we need agent skills for some of these things? Some people will just use Claude or ChatGPT to do this.

00:28:21Miguel Quintero: For that to work, you need the knowledge base first, the registries or documentation, so anyone writing a skill, in Codex or whatever, can infer from it. For example, Context7 is one of the first things I add to my own agent harnesses, so it picks up the latest package documentation. Which reminds me: I'll open an issue and talk to Phil about making sure OpenLint shows up in Context7, so that someone building an MCP server or skills picks up OpenLint and can do rules automatically.

00:29:12Dimitri [?]: It could be as easy as starting without a full registry or external references, just publishing snippets on a Markdown page.

00:29:40Kin Lane: I think that becomes the education section. We start with the snippets, and it becomes the registry and the education section as the snippets come together and we see what we have. And I agree with you, Miguel. I'm an old database guy, a SQL guy, and I see Claude as just a SQL statement: you've got to have the database you pull that context from, and it's just a harness. So we gather those snippets, as Dimitri says, us humans organize them as we gather them, and eventually we craft a set of skills on top that pull from our registry and education as context. I'm sad I don't use my regex book any more. I used to have it on my desk and write regex by hand. Not any more; natural language.

00:30:51Dimitri: In addition, I've been on vacation and ill for a while, so I'm not sure this was mentioned in the last weeks: in our approach in the Dutch government, the way we work with agents is a "validator in the loop", as we call it. Once an agent creates a ruleset for you from our knowledge base, we should also provide a linter, so the skill can say "after every feedback loop, call this linter to check the ruleset is a valid OpenLint ruleset." I'm not sure there's an issue for that already.

00:31:46Kin Lane: Miguel's working on the MCP design, so that linter could be exposed as an MCP tool, as one possibility. It could also be an API call. Then we look at what other distributions of a linter we'd want.

00:32:09Dimitri [?]: We could just add the OpenLint ruleset as a format for OpenLint itself, in the CLI.

00:32:23Kin Lane: Definitely. I'd like to interview you both. I'll go through most of your style guides and processes, because you're very public with what you have, but I'd love to ask how distribution works, how teams get this, what the guidance is, and the AI guidance, which is important. I'm doing the same with John Musser at Ford, who agreed to an interview, and a couple of others at banks and insurance companies: how do they currently use Spectral, craft rules, educate people, run their pipelines, and how is AI changing that?

00:33:12Miguel Quintero: That's a good strategy. To me the context window is an agent's workbench. There's a lot in it: MCP, the tooling accessible to the agent, RAG. Let's not forget RAG; you can still RAG your entire ruleset or your APIs and ask "are we missing some linting across our APIs?" You've got skills, you've got memory. Within all of that context, OpenLint has to be in there somewhere, so it works at the scale of Ford.

00:34:02Kin Lane: And at the input and output level, as Dimitri said, things have to be linted and validated. If you use something structured as an input, you should be able to validate it with JSON Schema going in, and when it comes out, use JSON Schema and OpenLint to check it's valid and complete. So I feel we're pretty close. Thank you for helping with housekeeping on the repos, keeping the website, the repo READMEs and the GitHub org aligned and saying the same thing. I feel we're open for business in a beta way. I'm switching my tools over, you're doing it for the Dutch government, and I want to find a couple of other companies who can commit to switching, or at least begin the process, while I find companies to interview about their processes. I'm going to reach out to the Italian government, because they have a really nice tool suite and ruleset, and the German government said they're willing to help. The next step is gathering interviews and getting some people migrating and giving us feedback on the migration path. I'm going to start talking more publicly about it and telling more stories on a daily basis, so people know we're migrating and things are happening, and work towards API Days Paris as our official launch, a line in the sand. Hopefully a couple more migration customers by then give it credibility. The other part: Open Source Europe is our fiscal partner, and we have a thousand dollars in the bank. API Evangelist is the first sponsor of OpenLint. I have a list of European grants to go after, the Sovereign Tech Fund and a couple of others. I'll also talk to Mehdi about a formal talk slot. One proposal is building our education setup, and another is funding the test harness. If you have other needs or ideas, especially around the core tooling, let me know, because I want to pay Phil. How are we on the spec side?

00:37:50Joost Farla [?]: The spec side is in pretty good shape. The working draft is in really good shape.

00:38:06Kin Lane: I've submitted a couple of PRs linking issues to existing discussions, and there are a few issues without discussions that I need to wire up and get going. I found a couple of small corrections, so we have two or three pull requests to go through. Any feedback would be great. I've gone through it once and I'm going through it again, and I'm pretty happy with it. Submit your feedback.

00:38:43Joost Farla: I'll be iterating on it next week. And maybe Dimitri and I could take a first pass at the organizational, modularization part. We have some ideas on how to approach that.

00:39:09Kin Lane: Please do, both of you. That would be great. Back to the spec: I'm trying to keep my pull requests modular and small, one batch of things at a time, so they stay separate. I'm pretty happy with the spec and where the tools are, and we have an npm package, so as far as I'm concerned we're open for business. By API Days we'll say it's officially open for business, first version.

00:39:48Dimitri: Perhaps it's also nice to mention that we're working on another project at the moment, a test bed, an investigation track, about how we make standards, written the way the Dutch government standards are, AI friendly: AI readiness for standards. It's also a test environment to see what the outcomes are, so perhaps we could take it back here as an example. Joost?

00:40:35Joost Farla: I'm actually in a full-day sprint today and tomorrow on this AI-ready standards project, so it might give some nice insights into how we can apply it to OpenLint. Once we have something to share, we will.

00:41:01Miguel Quintero: Is that related to OpenAPI, or more general, like MCP?

00:41:13Joost Farla: It's currently from a more geospatial perspective. We have all kinds of geospatial base registries in the Netherlands. If a citizen asks a question in ChatGPT, it probably won't base its answer on authoritative sources; it will just search and try to reason, and most often it won't consult authoritative sources. We want to steer that, so we're improving how we publish these registries and their metadata, so AI agents can do the work more efficiently and trustworthily. One possible outcome is that whenever a government organization publishes a standard, it also provides a skill file, for example. We're building a framework of artifacts and conventions a standard publisher follows to make their standard more digestible for AI. We've just started, so if I have results relevant to OpenLint, I'll get back to you.

00:43:02Dimitri: And I hope to publish a blog post on our website in a couple of hours saying we've moved from Spectral to OpenLint, which will make a lot of concerned people happy. Within the government, as far as I know, a lot of people use either our checker, which uses Spectral under the hood and OpenLint from now on, or the Spectral command line. I'll also ask them to test OpenLint immediately and, if it works, switch. That will probably bring more contributors to the project.

00:44:09Kin Lane: A lot of happy people. Me included. Thank you.

00:44:17Dimitri: Yesterday we had another meeting where we mentioned this project, and people are really waiting for a fix of the VS Code plugin. I don't know exactly what the outcome of that discussion was, but I still think we should create an LSP. Joost and I discussed yesterday that we're also willing to look into that.

00:44:41Kin Lane: Yes, please. Let me see if I can find it: Luke Park, with OpenRouter, has been working on OpenAPI LSP tooling and wants to get involved. So we have other people interested in that too. I'll do some storytelling around it. I'd also like to talk about your AI guidance, because I've done a lot of work on apis.io around artifacts: lots of tagging, linking to business capabilities, for example actual telecom subscriber-management capabilities, and conversational overlays attached to each operation and path, like "can I delete a conversation service", phrased the way people will speak to ChatGPT. These are all available as artifacts on the page: JSON-LD, overlays on the OpenAPI. I'm augmenting APIs with a rich set of artifacts and running tests on how ChatGPT, Claude and others ingest them. I get a lot of bot traffic and run tests to see which bots come in and what GPTBot consumes daily, then I ask them those questions. So I'd love to learn how you publish the geospatial data, and maybe contribute some ideas.

00:47:03Dimitri: Sure, let's talk.

00:47:06Kin Lane: On VS Code, we'll get that on track too. VS Code and the LSP work need to be prioritized; that's pretty critical, and I suspect VS Code will come up as super critical in the interviews. Ivan's working on a VS Code extension, though not an LSP, so we need to start pushing down that road too. I think we can do both in tandem and align them. The other early roadmap item for me is skills: we need Markdown linting. That will be one of the top roadmap items I push for, and it doesn't do it today.

00:48:18Dimitri: From our side, we're also extending our checker apps, as in the links I posted in the chat, with support for RDF. We're not sure how to approach it yet, but we'll figure it out, and hopefully that also leads to something we can contribute back.

00:48:45Kin Lane: There's a lot to think about. Busy times, but I feel good. We did it: we figured out the name, we forked it, OpenLint is up and running, it's a spec now, we have docs for the tool and docs for the spec, an npm package, and we're starting to migrate. We need more interviews and storytelling and more migrations. API Days Paris is coming with multiple talks referencing it. I gave a talk at API Days Australia the night before last, over Zoom, that included OpenLint and how I use it on apis.io. API Days Paris is an important line in the sand. We have a financial partner and some money in the bank. This feels real now.

00:49:57Dimitri: Next week I also have an appointment with someone from the government standardization board. I'm pitching OpenLint to him and he's very enthusiastic. The problem with the government is that it can't give out-of-pocket money, but there's a good chance they can assign Joost and me to this project for more commitment.

00:50:42Kin Lane: You as cornerstone contributors and maintainers of OpenLint will be critical: your style guide and rules management, and your AI-readiness work is another critical piece. If I can expand this to the German government, who are very eager too, and get a couple of enterprise customers... Pavel from SAP is very interested in supporting, but we need a couple of other big corporate customers doing the migration. That's a top priority for me. If we have this in motion by API Days Paris, going into the new year we can probably find serious funding and grants to stabilize, maybe get someone to help me run this meeting and the organization, and pay people to do the work. Anything else high-level? We've got eight minutes.

00:52:11Dimitri [?]: Shall I investigate the differences between the vacuum rules and the spec and the current OpenLint rules, the incompatible vacuum rules? I really want to know why they're different, and if there's a good reason, perhaps we could implement them.

00:52:39Kin Lane: Yes, please. A bunch of people use vacuum in their pipelines for good reasons, and it would be nice if they could migrate to OpenLint rulesets too. Whatever we find, we can take to Dave Shanley. He wasn't very interested in this spec being a standard, but if it brings him users, and he knows I can bring him users and adopters, since I've helped him grow his platform... Please do, and share anything you find. I'm going to step back and do the base ruleset, base aliases, and the regex and JSONPath work we talked about. That's where I'll focus, because I don't have my head fully around the foundation of how rules are used and communicated as part of the product, and I need to make sure it's documented properly in our education center. I'll share that next week. I think we're in business. Keep working on the core tools; I love seeing all the PRs and conversation. Keep working on the spec, and please review it. Phil got the docs for the tools up, so docs.openlint.org is available for the tools and spec.openlint.org for the spec. Keep the feedback coming. I'll work on some blog posts too. Dimitri, tag me in anything you post so I can amplify your blog post.

00:54:49Dimitri: I will. It has to be in Dutch, unfortunately; that's policy.

00:54:55Kin Lane: That's not a problem. I'll amplify it in English. Thank you all for your support and your work. This is great. Enjoy the rest of your week, and we'll see you next week.