Loop Engineering CLI Lets Developers Score and Design Agent Automation Systems

A new open-source project called Loop Engineering, published by developer Cobus Greyling, argues that the real bottleneck in AI-assisted coding is no longer prompt writing but the design of automated systems that prompt agents continuously. The project provides a CLI toolset installable via npm that scaffolds skills, state, and budget files, then generates a 'Loop Readiness Score' for a given project. It breaks agent control systems into five building blocks — automations, worktrees, skills, plugins, and sub-agents — supported by persistent memory and state outside any single conversation. Seven production patterns are included, each rated by cadence and token cost, with the README explicitly flagging high-cost options like PR Babysitter and CI Sweeper. The project targets developers using tools such as Claude Code, Cursor, Codex, and Grok, and recommends a phased rollout starting with observation before enabling any automated fixes.