loop-engineering: a framework for designing automated control systems that orchestrate AI coding agents

What it solves

Loop Engineering shifts the focus from writing individual prompts to designing automated control systems (loops) that orchestrate AI coding agents. It solves the problem of manual, repetitive prompting by creating recursive goals where the AI iterates on a task—using sub-agents, verification, and external state—until a goal is achieved or human intervention is required.

How it works

The system uses five core building blocks to automate agent behavior:

• Automations/Scheduling: Handles discovery and triage on a set cadence.
• Worktrees: Provides isolated environments for parallel execution.
• Skills: Maintains persistent project-specific knowledge.
• Plugins & Connectors: Integrates with external tools via the Model Context Protocol (MCP).
• Sub-agents: Implements a "maker/checker" split for implementation and verification.
• Memory/State: Uses a durable spine (like STATE.md) to maintain context outside of a single conversation.

Who it’s for

Developers using AI coding agents such as Grok, Claude Code, Codex, and Cursor who want to move from manual prompting to designing autonomous system-level workflows.

Highlights

• Production Patterns: Includes 7 pre-defined patterns such as Daily Triage, PR Babysitter, CI Sweeper, and Dependency Sweeper.
• Tooling Suite: Provides CLI tools for scaffolding (loop-init), estimating token costs (loop-cost), and auditing loop readiness (loop-audit).
• Phased Rollout: Recommends a safety-first approach moving from report-only (L1) to assisted fixes (L2) and finally unattended automation (L3).
• MCP Integration: Supports the Model Context Protocol for extending agent capabilities.