Agent Island: 7-State AI Coding Agent Monitoring Taxonomy

A new open-source project called Agent Island is introducing a structured framework to monitor AI coding agent sessions. This framework defines seven operational states an AI coding agent session can occupy. It also includes a desktop application for macOS and Windows, currently at version 1.7.1. The goal is to help developers understand exactly what an AI coding agent is doing at any given moment. The taxonomy distinguishes between events and states, arguing that many existing tools conflate the two, leading to misleading results. For example, a completion event doesn't prove code is correct. The seven defined states are: inactive, running, waiting_for_user, completed, blocked, stalled, and unknown. Each state requires specific evidence to be assigned. This matters because it provides a reliable way for developers to get at-a-glance session awareness without uploading their data.