kunchenguid/lavish-axi

Lavish is a command line tool built as a richer editor for HTML files that AI coding agents generate. The problem it addresses: when an AI assistant produces an HTML file, such as a web page, a plan, or an interactive document, the usual way to give feedback is to take a screenshot and describe the change in a text message. That approach throws away the interactivity that makes HTML useful in the first place. Lavish opens the generated HTML file in a local browser window and layers annotation tools on top of it. You can click specific elements on the page, highlight ranges of text, or type feedback into a chat panel, and that feedback is turned into a structured prompt that tells the agent exactly what to change and where. The agent updates the file, and the browser view reloads automatically to show the result. Everything runs on your own machine. There is no cloud service involved, and the HTML file itself is never modified in a special way, so it still looks and works normally if you open it directly in any browser outside of Lavish. The tool is designed to be called directly by AI agents rather than by a human typing commands. The project describes this pattern as an AXI, a command line tool built specifically for agent use, meaning an agent can open a session pointed at a file path, poll for human feedback, and receive the annotated instructions back without any special setup or training. Lavish also ships prebuilt playbooks that guide agents toward producing good results for common artifact types, including diagrams, tables, slide decks, comparison views, and interactive forms. It installs through npm and works on macOS, Linux, and Windows.