agentmon-live

Claude Code mod that checks whether "done" actually means done We've been working on AgentMon Live, a plugin that puts a live monitoring pane beside your Claude Code chat. The main reason we built it: Claude can report that work is finished even when a test failed, the last passing test predates the latest edit, or a subagent left something unfinished. The plugin adds a "done gate" that checks for those conditions before Claude stops. It sends unfinished items back to Claude once. A second model, which doesn't write code, checks whether every part of your request was answered. The pane also shows: - Running tools and subagents - Files edited and read - Latest test, lint, and build results - Context usage and token consumption - Items needing attention, including repeated failures and stuck subagents You can turn the done gate off and customize which sections appear. Install in Claude Code: /plugin install agentmon-live --marketplace codenotary/agentmon-live Requires Claude Code 2.1.308 or later. Apache 2.0 licensed. GitHub - Open source under Apache 2.0.: codenotary/agentmon-live Walkthrough: codenotary.com Curious what you'd want checked before Claude calls a task finished. Failed tests are straightforward. What else do you regularly catch during review?