Verified project record

cobusgreyling/loop-engineering

Practical patterns, starter kits, and CLI tools for designing, orchestrating, and auditing automated loops for AI coding agents. It transitions teams from constant manual prompting toward system-driven control, while requiring external AI coding agents to execute the generated loops.

Project overview

Provides seven production patterns with specific cadences, token costs, and autonomy levels, alongside mechanical gate enforcement and state drift detection for long-running agent loops.

Project type: AI Coding · Workflow · Prompt Engineering
Use cases: Coding & Development · Automation
Deployment: Refer to project documentation
License: MIT

Best for

Developers and AI engineers seeking to implement structured, system-driven control over AI coding agents using defined cadences and autonomy levels.

Key capabilities

Scaffolds skills, state, and budget files, then prints a Loop Ready score and first loop command.
CLI tool that evaluates a project and suggests improvements for automating agent loops.
Estimates token spend for a specified loop pattern and cadence.
Detects drift between STATE.md and LOOP.md.
Stateful memory manager and circuit breaker for long runs.
MCP runtime lookup for patterns, skills, state.
Manages isolated git worktrees per fix attempt.
Mechanical enforcement of the path denylist and auto-merge allowlist.

Limitations and risks

Unattended loops make unattended mistakes, requiring user verification of the output.
Token costs can explode when using sub-agents and long-running loops.
Comprehension debt grows faster unless the user reads what the loop ships.
The data boundary is not documented; local execution of orchestration tools does not prove a local-only data boundary for the external AI coding agents required to run the loops.

Getting started

Setup is evaluated as easy because npm packages can be run directly via npx without cloning the repository.
Run the provided CLI commands to initialize a loop pattern, estimate token costs, and audit the project for readiness.

Alternatives and comparisons

Bridges UI design tools with compatible coding agents by providing standardized skills for design generation and code conversion.
Provides a test-based, multi-stage iterative flow intended to improve LLM performance on code-generation problems.
An AI development environment that orchestrates multiple parallel CLI coding agents side-by-side in isolated worktrees.

Project comparisons

Evidence and sources

GitHub project description: Practical patterns, starters & CLI tools for loop engineering with AI coding agents.
README: Loop engineering replaces you as the person who prompts the agent — you design the system that does it instead.
README: loop-init scaffolds skills, state, and budget files, then prints your Loop Ready score and first loop command.
README: loop-audit npm
README: Seven production loop patterns with cadence and token cost

Reviewed at: 2026-07-18T05:11:02Z

Open original GitHub project