lavish-axi: A Local-First Editor for Agent-Generated HTML Artifacts
HTML is the new markdown. Lavish is the new editor for your HTML artifacts.
At a glance
- What is it?
- lavish-axi is a CLI-based editor that opens agent-authored HTML files in a local browser, lets humans annotate elements and edit Mermaid diagrams, and sends feedback back to the agent. It is an AXI, meaning no setup beyond a skill install or direct npx invocation.
- Who is it for?
- Adopt lavish-axi if you regularly review agent-generated HTML artifacts and want a local, token-efficient feedback loop that preserves interactivity. Do not adopt it if you need a full WYSIWYG editor or if your workflow relies on cloud-based collaboration; the core loop is local-first with opt-in sharing.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- Mainly JavaScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 14, 2026, and from our analysis. They are not legal advice.
Star history
- kunchenguid/lavish-axi4,577 ★ · +1,634
| Repository | 2026-08-22 | 2026-11-14 |
|---|---|---|
| kunchenguid/lavish-axi | 2.9k | 4.6k |
1,634 Stars in the last 84 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
The Problem: Human-Agent Loops on HTML Are Stuck in Screenshots
Agents can generate rich HTML artifacts, but the human-agent collaboration loop around them is weak. The README states that feedback often falls back to screenshots and long responses for "tell me what to change," which loses HTML's key strength: interactivity. This project targets that gap. It is for engineers and product managers who review agent-produced plans, diagrams, tables, or reports and need to give precise, contextual feedback without leaving a browser. The intended user is someone already working with an agent harness like Claude Code, Codex, or Copilot CLI, and who wants a local, low-friction way to annotate and iterate.
How It Works: An AXI That Opens a Browser and Sends Feedback
lavish-axi is an AXI, which the README defines as a CLI any capable agent can run without setup. The flow is simple: the agent writes an HTML file, then the CLI opens it in a local browser UI. The human can pinpoint elements and selected text ranges, edit Mermaid whiteboard diagrams, and send messages back to the agent. The CLI is designed for agent ergonomics: it uses TOON output, long polling, and contextual disclosure to keep token usage low. The skill and hooks only handle discovery; agents learn the AXI by using it, not by reading long instructions. This means the CLI itself is the interface, not a server or a plugin that exposes MCP endpoints. The README explicitly says it declares no MCP server.
Getting It Running: Skill, Hook, or Direct npx
The recommended path is to install the Lavish skill via `npx skills add kunchenguid/lavish-axi --skill lavish`. That is the entire setup, no npm install needed. The skill teaches the agent to run `npx -y lavish-axi`, so the CLI comes on demand. It stays a short stub that sends the agent to `--help`, `design`, and `playbook` for current instructions, so it cannot go stale. For ambient context, you can install the CLI globally and run `lavish-axi setup hooks`, which installs a SessionStart hook for Claude Code, Codex, OpenCode, and GitHub Copilot CLI. This hook surfaces open sessions and visualization playbooks at the start of each session. Alternatively, you can install the CLI globally and run `lavish-axi setup plugin` to register it with VS Code, Cursor, and GitHub Copilot CLI as an Agent Plugin. The README warns to restart your agent session after installing the hook.
The Local-First Trade-Off: No Cloud in the Core Loop
The project is local-first by design. The README says there is no cloud dependency in the core feedback loop; hosted sharing through third-party ht-ml.app is explicit and opt-in. This is a genuine strength for privacy-sensitive work, but it is also a limitation. If you need to collaborate with remote teammates on the same artifact, you cannot rely on this tool's core loop. You would have to use the opt-in sharing service, which is third-party. For a solo engineer iterating with an agent, local-first is fine; for a distributed team, it is the wrong tool unless you are willing to add that external dependency.
Maintenance and Upgrade Cost: Versioned Releases and Self-Repairing Hooks
The project is actively maintained, with three releases in three days (v0.1.94, v0.1.95, v0.1.96) up to the last push on 2026-10-29. The README describes a mechanism to avoid staleness: the skill is a stub that points the agent to the latest CLI instructions, so an installed copy cannot go stale against a newer CLI. The plugin setup is idempotent and repairs the registered path after a reinstall or relocation, which reduces upgrade friction. However, the hook installation requires a restart of the agent session, and the README notes that Codex and ChatGPT install plugins only from marketplace sources, so Codex users must use the session hook instead. This means upgrade cost is low for skill users but slightly higher for hook users who must restart sessions. The license is MIT, which allows free use and modification with attribution, but this is not legal advice.
Limitations and Failure Modes: Not a WYSIWYG Editor, and Setup Depends on Your Harness
Lavish's core loop is reviewing an artifact: pinpointing elements and selected text, editing Mermaid whiteboards and sending feedback to the agent. A failure mode is that the tool only works if your agent harness can run npx commands and supports the skill or hook mechanism. The README says any capable agent can run the CLI directly, but the skill installation targets specific formats like Agent Skills and Agent Plugins. If your harness does not support those, you must fall back to telling the agent to use `npx -y lavish-axi` manually, which still works but loses the discovery features.
Alternatives: Screenshot-Based Feedback vs. Full Editors
The obvious alternative is the current practice the README criticizes: screenshots and long text responses. That approach is simpler but loses interactivity. Another alternative is to use a full HTML editor like Visual Studio Code with a live preview plugin, which gives you direct manipulation but no agent feedback loop. The difference is that lavish-axi is built specifically for the agent collaboration loop, not for human-only editing. A third alternative is to use an MCP server that exposes browser automation, but the README explicitly avoids that by declaring no MCP server. The project's approach is to make the CLI itself the interface, which is a different architectural choice from an MCP-based tool. If you already have a MCP-based workflow, this tool may not fit without adaptation.
Who Should Adopt It, and What to Verify
Adopt lavish-axi if you are an engineer or PM who works with agent-generated HTML artifacts and want a local, token-efficient way to annotate and iterate. Do not adopt it if you need a full WYSIWYG editor or if your team requires cloud-based collaboration in the core loop. Before adopting, verify that your agent harness can run npx commands and supports either the skill or hook mechanism. Check the latest release notes for any changes, since the project is actively versioned. Also verify that the `npx skills` command is available in your environment, as that is the recommended installation path. If you plan to use the hook, be prepared to restart your agent session after setup. The project is MIT-licensed, so you can inspect and modify it freely, but you should review the code yourself if you need to trust its security posture.
Editorial conclusion
Adopt lavish-axi if you regularly review agent-generated HTML artifacts and want a local, token-efficient feedback loop that preserves interactivity. Do not adopt it if you need a full WYSIWYG editor or if your workflow relies on cloud-based collaboration; the core loop is local-first with opt-in sharing. Before adopting, verify that your agent harness supports the AXI pattern (e.g., Claude Code, Codex via hook) and that you are comfortable running an npx-based CLI. Check the latest release notes for any changes to the skill installation path or hook behavior, since the project is actively versioned (v0.1.96 as of the last push).
Frequently asked questions
What is lavish axi?
It is a CLI and local browser UI for reviewing HTML artifacts that an AI agent produced. Given a file path it opens the file in a local browser, lets you pinpoint elements and selected text, edit Mermaid whiteboard diagrams, and send the feedback back to the agent. It is MIT-licensed, written in JavaScript, and requires Node 22 or newer.
How do I install lavish-axi?
The documented setup is one command that installs the skill rather than the program: npx skills add kunchenguid/lavish-axi --skill lavish, with no npm install step, because the skill runs the CLI through npx on demand. For the session hook or the plugin you install the package instead with npm install -g lavish-axi and then run lavish-axi setup hooks or lavish-axi setup plugin. Building from source needs pnpm, a frozen lockfile install, a build and a link.
What is Lavish AI?
The package name is lavish-axi, the command is lavish-axi, and the public skill is simply named lavish, and the README calls the product Lavish Editor. There is also a second, internal skill called lavish-design that ships in the repository but is hidden from normal discovery unless INSTALL_INTERNAL_SKILLS=1 is set.
Does lavish-axi need an MCP server?
No, and the project says so directly: the CLI itself is the agent interface. Because of that, installing the plugin brings the same skill the skill install would give you, which is why the two are documented as alternatives rather than something to stack. Installing both buys you a second registration path, not extra capability.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/kunchenguid-lavish-axi)Star history chart for your README
[](https://hysenlabs.com/projects/kunchenguid-lavish-axi)Was this useful?