This repository provides practical patterns, starter kits, and CLI tools for "loop engineering," shifting AI coding agent interaction from manual prompting to system design. It targets developers using agents like Grok, Claude Code, and Codex, enabling the creation of automated, iterative agent workflows and elevating leverage from prompt crafting to system orchestration.

How It Works

The core is the "loop": a recursive goal where AI agents iterate using sub-agents, verification, and state management until completion or human handoff. Key primitives include Automations/Scheduling, Worktrees, Skills, Plugins/Connectors, and Sub-agents, all underpinned by a durable Memory/State layer. This approach focuses on designing control systems for agent orchestration over time.

Quick Start & Requirements

Usage centers on npx commands for CLI tools: loop-init for scaffolding, loop-cost for token estimation, and loop-audit for readiness scoring. Development from source requires npm ci. Prerequisites include AI coding agents (Grok, Claude Code, Codex). Links to an interactive pattern picker and documentation are available.

Highlighted Details

• Three core CLI tools: loop-audit, loop-init, loop-cost.
• Seven production loop patterns (e.g., Daily Triage, PR Babysitter) with starter kits.
• Automated CI workflows validate patterns and audit readiness on pushes/PRs.
• Supports phased rollout: L1 (report) to L3 (unattended automation).

Maintenance & Community

Maintained by Cobus Greyling, with conceptual contributions from Addy Osmani. Community interaction via GitHub Discussions. Automated CI provides continuous health signals.

Licensing & Compatibility

Released under the permissive MIT license, allowing broad adoption and integration into commercial projects.

Limitations & Caveats

Loop engineering amplifies judgment; unattended loops risk unattended mistakes. Token costs can escalate, and verification remains the user's responsibility. Results may vary, and comprehension debt increases if loop outputs aren't reviewed.