Show HN: Git-Backed Skill Sharing for Non Techies

Alex Strick van Linschoten, engineering lead at ZenML, built Skillburst, a git-backed skill-sharing platform for non-technical teams, after finding existing tools like Anthropic's Skills, OpenAI's custom GPTs, GitHub, and Notion each only partially solved the problem of distributing AI workflows across an organization. Skillburst provides a neutral catalog with version control and governance, allowing skills to be created, reviewed, and automatically updated in whatever AI tool each person uses, bridging the gap between engineers and the rest of the company.

← Back to blog /blog/ Why I Built Skillburst At ZenML I had AI skills I wanted to hand to the non-technical side of the company. Every tool I tried got one part right and the rest wrong. So I built the missing piece. I run engineering at ZenML https://zenml.io , so the AI part came easily to the engineers. We live in Claude Code and Cursor all day, and when one of us works out a good workflow, we share a file and move on. The rest of the company was a different story. Our sales, marketing, and ops people wanted the same thing, and they were good at it. Someone in sales worked out a really effective way to turn call notes into a follow-up that sounded like her, next steps included. It was better than what most of the engineers would have written. The problem was never talent. It was distribution. That workflow lived in her head, and in one Slack message. Two weeks later, nobody could find it. I kept hitting the same wall. I had a handful of these skills, really just named, repeatable instructions that make the model do one job well, and no clean way to get them to people who don’t open a terminal. So I went looking for something that already solved it. I assumed something would have. It hadn’t. Not really. What everyone got half-right To be fair to the tools I tried, each of them got a real piece of this right. Anthropic and OpenAI both have something now, but it’s tied to their own models. Claude has Skills and Projects, OpenAI has custom GPTs, and you can share them across a workspace. That’s real progress. But each one is a single-vendor silo. Claude’s org skills only reach Claude. Workspace GPTs only reach ChatGPT. When half your team is in Claude Code and the other half is in Cursor or Codex, nothing spans them. And even inside one vendor, you get sharing without much governance: no approval step, no “everyone’s on the reviewed version,” no clean rollback when something goes wrong. Sharing is mostly solved. A neutral catalog with real version control, across the tools people already use, is not. GitHub is where our engineers already keep everything. So the instinct was to just put the skills in a repo. For the engineers, that’s right: version control, pull requests, review, history, all there. But you are never going to get someone in sales to clone a repo, and you shouldn’t ask them to. A repo is the right backend and the wrong front door for two-thirds of the company. Notion, Confluence, a shared drive. This is where most teams actually put this stuff, and it fails in a simple way: docs can’t run. A skill in Notion is a description of a skill, not the skill itself. People copy it, paste it, tweak it, and now there are nine versions and nobody knows which one is current. Good for finding things, useless for running them. So that was the map. Vendor tools that share within their own walls. A developer backend most of the company won’t touch. Doc tools that store text but can’t run it. Everyone had a piece. Nobody had the middle. The piece that was missing What I actually needed was pretty boring to describe. A place where a skill gets created once, reviewed by someone who owns quality, and then shows up automatically, always current, inside whatever AI tool each person already uses. Engineers keep working in GitHub, because that’s right for them. Everyone else never thinks about GitHub, or versions, or where the file lives. They just have the skill. That’s the whole product, really. A distribution and governance layer between the person who builds a skill and the rest of the org that should be using it. So I built it, because I wanted to use it. That’s Skillburst. How it actually works The shape follows the two audiences, because that was the whole problem. Someone builds a skill, either from their AI or by pushing it from a GitHub repo. It lands in your org catalog as a proposal. A lead reviews and approves it, and every skill is backed by a real GitHub repo underneath, so you get version history, pull requests, and rollback for free. That’s the governance half: the part that lets you say yes to the whole company using AI without it turning into chaos. Then the distribution half: once a skill is approved, it shows up inside everyone’s AI Claude Code, Cursor, Codex, Claude.ai, anything that speaks MCP and stays on the latest approved version without anyone lifting a finger. The person in ops doesn’t install anything or learn anything. The skill is just there. And when they have feedback “this should include a next-steps section” , it goes back through the same connection. The skill gets updated once, and everyone’s on the new version the next time they use it. Engineers keep their PR flow. Everyone else gets an assistant that works a bit better than it did last week. Nobody had to become a prompt engineer. What I actually believe Every company already has its best AI workflows. They’re not missing. They exist right now, in the head of the one person on each team who’s really good with this stuff. The gap was never capability. It was that there’s no easy way to move what that person knows to everyone else, and to keep it current as they improve it. Give people that, and a few things change. Your best person stops being a help desk and starts curating the catalog. The skeptical team lead sees hours saved instead of a slide deck. The new hire gets the org’s accumulated AI knowledge on day one, instead of three months of learning how things are done here. I built Skillburst because I wanted it at ZenML. I’ve spent the last two weeks in San Francisco for the AI Engineer World’s Fair, and the same problem came up in conversation after conversation: someone’s best AI work stuck with one person, or copied across tools that never stay in sync. It felt like the right time to stop keeping it to ourselves and put it out there. If your skills are stuck in one person’s head, or scattered across Slack, a repo, and a Notion page nobody’s opened in months, it’s probably time to give them somewhere to live. You can try it free https://app.skillburst.ai/register . It connects to Claude, Cursor, Codex, and Gemini, and you’ll have your first skill shared across your team in about ten minutes. If it’s useful, tell me what’s missing. I’m still building it. Want to equip your team with AI skills? Skillburst helps organizations discover, create, and deploy AI skills across their workforce. Get started → https://app.skillburst.ai/register