loop-engineering

Loop Engineering is a pattern library plus CLI toolkit for operating AI coding agents around a repository. Rather than typing each prompt by hand, you design a loop: a system that discovers work, hands it to agents, verifies results and persists state. Getting started uses npx: `npx @cobusgreyling/loop init . --pattern daily-triage --tool claude`, followed by `doctor` and `cost` commands. The `--tool` flag defaults to claude and can be swapped for grok, codex or opencode. Week one is report-only. Documented patterns include Daily Triage, Thin loop, PR Babysitter, CI Sweeper, Dependency Sweeper, Changelog Drafter, Post-Merge Cleanup and Issue Triage. Each entry lists cadence, week-one level and relative cost. An interactive pattern picker and a registry.yaml index are provided. Rollout guidance suggests progressing L1 report, then L2 assisted, then L3 unattended only after the verifier has been reliable for a week. A Loop Ready score weights recent runs more heavily than files on disk. Examples are organized by tool: Claude Code (including a plugin), Grok, Codex, OpenClaw, Opencode and GitHub Actions. Operating material covers failure modes, anti-patterns, safety and operating loops, with stories that include failures. The README warns that token costs can grow and that unattended loops make unattended mistakes. A unified CLI front door offers init, doctor, status, audit and cost; older packages such as loop-init and loop-audit remain supported. Companion repositories are listed as optional and advised only after a loop has actually run. Licensed MIT.