Turn a feature spec into reviewed, merged code with bounded AI agents (github.com)

A new tool called pm-go has been introduced as a comprehensive control plane for AI-assisted software delivery. By leveraging feature specifications and repositories, pm-go transforms these inputs into structured plans, executing tasks through scoped implementation agents within isolated work environments. It emphasizes a durable workflow that records plans, approvals, and audits in a Postgres database, allowing for resumable and inspectable operations. Unlike traditional agent coding tools that rely on chat interactions, pm-go promotes bounded autonomy for agents and ensures that tasks are only marked complete when all checks align, thereby enhancing reliability in the development process. This innovation is significant for the AI/ML community as it facilitates a disciplined approach to AI integration in software development, addressing common challenges related to version control and task management. Technical components include a Hono control-plane API, Temporal workers for managing workflows, and an Ink-based terminal interface for task monitoring. The tool ensures a local-first development loop and provides adaptability with various runtimes for different operational needs. With these capabilities, pm-go aims to streamline software delivery processes, reduce manual intervention, and maintain high-quality standards through structured audits and approvals.