Agent Autoloop

Overview

Agent Autoloop is a very lightweight automation hub for command-line agents such as Claude Code and Codex. It turns schedules, GitHub Issues, and service alerts into standard events, then uses local rules to decide which agent receives each event and in which directory it runs. An agent works one round and exits; only when a decision is needed does it send a card, and an answer resumes the same conversation to finish the work. It does not depend on a cloud server or public port and can run on a personal Mac or Linux machine.

Key Features

• Runs on a single computer with one command, without a cloud server or open public port. Rules in rules.json decide which event goes to which agent and directory, with scheduled and external events sharing the same rules. While waiting for a decision, the process exits, so no

Use Cases

• Daily brief: on a schedule, organize local notes and Git commits and write them to briefs/YYYY-MM-DD.md. GitHub Issue follow-up: check for new Issues on a schedule, change code locally, and notify you to review; it does not commit, push, or reply to the Issue. Production alert

Getting Started

• On macOS or Linux, install with the project's install.sh; both installation and agent-autoloop update use the latest Release. Run agent-autoloop try to walk through a real loop with a signed-in local agent: ask, exit and wait for a reply, decide, and finish the work. Open

Deployment & Requirements

• Supported on macOS and Linux. A Windows version is pending, and Windows use is not available now; if waiting is not an option, WSL2 can be used with the installation notes. Before running, you need Node 24, Git, curl, and any signed-in command-line agent; no GitHub account is

Before You Adopt

• License: MIT. Review its terms before using, modifying, or distributing the project.