Autonomous AI development studio that orchestrates Claude agents to implement features on a Kanban board. My first open-source project, built with a team of 4. Over 3,000 stars in 2 months.
AutoMaker is an autonomous AI development studio. Users describe features on a Kanban board, and when moved to "In Progress", a Claude agent automatically implements the feature in an isolated git worktree. You review the changes, approve or reject, and move on to the next one.
This was my first open-source project. A team of four of us started it for fun and it took off way faster than we expected. We hit 3,000 GitHub stars within two months. Contributing every day, pushing rapid changes, and watching the community grow was an incredible experience.
Building AutoMaker taught me things no tutorial ever could. How much time open-source maintenance actually takes. How hard it is to keep a codebase healthy when multiple people push changes daily. How to support cross-platform users with completely different environments. How to coordinate a team around a fast-moving project without everything falling apart.
I no longer have time to contribute actively, but AutoMaker remains one of my proudest projects. Not because of the star count, but because of everything I learned building it.
npm workspace monorepo with 2 apps (React+Electron frontend, Express+WebSocket backend) and 8 shared libraries (types, utils, prompts, platform, model-resolver, dependency-resolver, git-utils, spec-parser). AI agents run through the Claude Agent SDK with autonomous tool use. Each feature executes in its own git worktree for isolation. File-based JSON storage in .automaker/ directory, no database required.
Solution: Set up CI pipelines, code review processes, and clear contribution guidelines. Learned the hard way that moving fast without process leads to breakage, but too much process kills momentum.
Solution: Built a platform abstraction layer (libs/platform) for path management and security, with Electron builds via electron-builder and Docker support for multi-arch deployments.
Solution: Each feature runs in its own git worktree, giving the agent a completely isolated copy of the repo. Users review changes before anything touches the main branch.
Desktop app for running and managing up to 12 AI coding assistant sessions in parallel, with per-session terminals, git worktrees, and MCP configs.
Personal anime browser and tracker. Search, track, and organize your watchlist with a built-in browser, Discord Rich Presence, and community bot.
A local-first music player for your own library. Like Spotify, but for files you already have. Live at shiranami.app.