akigarage/autonomous-project-run-skill

autonomous-project-run is a Skill, meaning an add-on capability for an AI coding agent, that tries to take a multi-part GitHub project from a rough idea all the way to a finished, verified result while needing as little human oversight as possible. Instead of a person checking in after every step, it coordinates the whole chain: writing a specification, breaking work into dependent tickets, running each implementation task in isolation, testing it, having it reviewed by AI, checking continuous integration, opening pull requests, merging them, closing issues, and finally auditing the whole project before calling it done. It only interrupts a human when a decision would meaningfully change the project's direction, scope, or do something that cannot be undone. It works through one ticket at a time in a fresh task each time, verifying that ticket before moving to the next, and it refuses to touch production systems, spend money, handle credentials, run destructive operations, force-push code, or bypass a repository's protection rules. This skill depends on another skill collection from a separate author, Matt Pocock's workflow skills, which provide the underlying specification, test-driven-development, and code review pieces this tool coordinates. If that companion set is not already installed and configured in a target project, this skill automatically triggers the official setup process for it before continuing. Using it requires a GitHub-backed project, the GitHub command line tool signed in and authenticated, and a coding agent runtime that supports Agent Skills. Installation is two commands using a package manager called skills, one to add the companion workflow suite and one to add this skill itself. The project is pre-stable at version 0.4.0, is not affiliated with or endorsed by Matt Pocock despite building on his work, and is released under an open license with attribution details in the repository.