Show HN: Agent that refuses to run commands without human approval (github.com)

Fewshell has emerged as a new SSH copilot designed for mobile and desktop use, addressing the challenges of remote command execution without compromising security and privacy. This tool allows on-call devops, MLOps professionals, AI researchers, and sysadmins to manage infrastructure safely by ensuring that no command is executed without human approval. Key features include secure defaults with mandatory SSH and secrets management, cross-device synchronization, and a user-centric approach that prioritizes human oversight over AI functionalities. The significance of Fewshell lies in its unique stance on AI command execution, minimizing security risks associated with automated infrastructure management. Built using Flutter, it offers a lightweight and performant solution, with capabilities such as session archival for post-mortem analyses and a command snippet library for streamlined operations. Fewshell operates under the premise that while AI enhances productivity, maintaining human oversight is crucial in environments where infrastructure integrity is vital. This project is open-source under the AGPL license, fostering community engagement while emphasizing user control over deployments.