bechsteindigital/claude-shiploop

claude-shiploop is a set of skills for Claude Code, the AI coding assistant, designed to turn a short project pitch into a working minimum viable product with as little back and forth as possible. Instead of you typing instructions at every step, the suite takes over after one initial interview and runs the rest of the process on its own. The flow starts with an onboarding interview where you describe your idea and answer a small number of questions about scope and goals. Once you approve the resulting brief, the system moves through setup, which researches the right technology choices, scaffolds the project, and breaks the work into cards. It then starts a loop where several role based AI agents work together: one plays a CEO making portfolio decisions, one plays a product owner cutting work packages, two development agents implement code in parallel, and a reviewer checks each change before it counts as done. Decisions the agents make along the way are written down as a log rather than left unexplained. You are only pulled back into the process in specific situations that you agree to in advance, such as a decision involving money, legal or security concerns, or a task that has failed twice in a row. Otherwise you only see milestone reports, and a final report once the agreed minimum viable product criteria are met. To use it, you install it as a Claude Code plugin or run an install script inside your target project, then start a Claude Code session there in a non interactive permission mode so the agents can write files and run commands without stopping for approval each time. The project stores its own working files, such as the brief, current state, decision log, and backlog of work cards, inside the target project itself, so a run can be paused and resumed later. The project is released under the MIT license.