FULCRUMAXE - An autonomous dev team that lives in your GitHub repo

Hi everyone — I released fulcrumaxe today. It's an autonomous development team that works on itself. Agents pull work from a backlog, write the spec, implement it, review each other's PRs, and merge. Nothing lands until code review, acceptance and security all pass. It's been running since April: 1,071 merged PRs and 958 design discussions, across 42 days of actual work. Almost all of it is work the system specced, built, reviewed and merged on itself. I've adapted it for countless projects and it's become my daily driver for how I work with AI and code. I've tried a lot of other tools and pulled everything I learned from them into this. Nothing else has quite felt like this — to me, anyway. One thing I didn't expect: it coexists with whatever spec system you already use. I run epics, tasks and subtasks on some of my projects, and that sits alongside GitHub Discussions as state management perfectly well. You just have to prompt-engineer your way into the adaptation. You don't have to throw out how you already work. I'm open-sourcing it because I want it to become the best it can be, and I don't think that has to be me. Me building on it, someone else building on it, or someone else taking the idea somewhere better — all three are good outcomes. Where I'd genuinely love help: - Coldstarts — smoother behavior when the harness meets a brand-new project - Cross-project sync — harnesses updating from a central source instead of drifting apart - Fleet — running teams across more than one machine It's AGPLv3, you bring your own Claude subscription or API key, and there's nothing hosted and nothing to sign up for. You can read the live queue — including the bugs it filed against itself — and browse the whole codebase as an interactive graph at fulcrumaxe.dev. If you want to contribute, or just run it and tell me what broke, I'd like to hear from you — fulcrumaxe.dev, johnproblems.dev, or right here. I'll be around all day to answer whatever I can.