# Loop Engineering

A pattern library and CLI for operating AI coding agents around a codebase on a recurring basis, scaffolding patterns such as daily triage or a PR babysitter with a graded, report-first rollout.

Loop Engineering is a pattern library and set of command-line tools for operating AI coding agents around a codebase on a recurring basis, rather than issuing prompts to an agent one at a time. A user designs a loop that discovers work, hands it to an agent, verifies the results and persists state between runs, instead of typing the next instruction by hand each time.

Its CLI, installed and run through npx, scaffolds a chosen pattern, such as daily issue and dependency triage, a PR babysitter, a CI sweeper, a changelog drafter or post-merge cleanup, against a chosen agent tool, including Claude, Grok, Codex or OpenCode, with each pattern documenting its cadence, a conservative report-only first week, and a relative cost level.

A maturity model grades a loop from level 1, report-only, through level 2, assisted, to level 3, unattended, weighting how recently a loop's state file has actually been used rather than simply whether one exists on disk, and the project documents failure modes, anti-patterns and safety guidance for running loops unattended.

Website: https://cobusgreyling.github.io/loop-engineering/