kunchenguid / lavish-axi

The public lavish skill tells an agent when to reach for the CLI: whenever it's about to hand back a plan, comparison, diagram, table, code diff, or report — anything easier to review visually than as prose — it opens the file in a local browser window where a human can annotate elements or Mermaid nodes and send feedback straight back to the agent. A second skill, lavish-design, is an internal brand-and-UI-kit reference for the maintainers' own product; the README says it stays hidden from default listings unless a reader sets an internal-skills flag, so most people who install this package will only ever see the one public skill.

This is for someone already using an agent that writes HTML artifacts — plans, diagrams, reports — who wants a faster review loop than screenshots and back-and-forth prose for revisions. It needs no account: the lavish skill just shells out to the lavish-axi CLI on demand, and the CLI itself handles the browser session, layout checks, and export.

READ THE FULL ANALYSIS＋

Sharing goes through a third party, but only if you ask for it. The README is explicit that the core review loop is local-first with no cloud dependency, but the CLI's own share command publishes an export to ht-ml.app, a separate hosted service, public by default unless a password is set — worth knowing before sharing an artifact that shouldn't leave your machine.

What we have not checked. The README documents the export pipeline, the layout-audit gate, and a session-start hook for Claude Code, Codex, OpenCode, and GitHub Copilot CLI in detail, but we have not run the CLI ourselves to confirm the annotation loop behaves as described end to end.