What is the Lavish skill for Claude Code?

Reviewing a long agent response in a terminal is the worst part of agentic coding. You scroll, you lose the thread, you paste fragments back with "change this one". Lavish moves that review into a browser page you can click on, and sends your annotations back to the agent. It is one of the more unusual entries in the UI design sub-niche.

Key takeaways

• The lavish skill has 16,472 installs on the kunchenguid/lavish-axi repo, which carries 3,329 GitHub stars and was last pushed 5 September 2026 (skills.sh registry, GitHub).
• It writes an interactive page under .lavish/, opens it with npx -y lavish-axi, and keeps a long-poll session open so element-level feedback reaches the agent without a copy and paste round trip.
• Seven playbooks ship with it: diagram, table, comparison, plan, code, input, and slides.
• It reads your design system first. If the project uses Tailwind CSS plus daisyUI, the generated artifact matches it rather than inventing a look.
• Two smaller siblings exist on the same repo: lavish-design (103 installs) and lavish-axi (49 installs), both far behind the main skill.

What problem does Lavish actually solve?

It solves review bandwidth. Agent deliverables arrive as prose, and prose is a bad format for a comparison table, a migration plan, or a diff. Lavish converts that output into rich HTML artifacts developers can review, annotate, and refine in the browser (skills.sh registry). The feedback path is the interesting half: annotations, queued prompts, and layout warnings travel back over a long poll, so the agent gets structured corrections instead of a paragraph describing them.

What is the best skill for reviewing agent output visually?

Lavish is the clear pick, at 16,472 installs against 103 for lavish-design and 49 for lavish-axi on the same 3,329-star repo (skills.sh registry, GitHub). Install it with:

npx skills add https://github.com/kunchenguid/lavish-axi --skill lavish

The browser session needs no global install; the skill launches it with npx -y lavish-axi and serves the artifact from a local Express server using relative asset paths, so the page stays portable if you send it to someone. Ranked alternatives across the whole design surface sit on the frontend and design pillar.

How do the seven playbooks work?

Each playbook is a layout contract for a kind of output. Diagram, table, comparison, plan, code, input, and slides each render differently, so a migration plan does not get squeezed into the same shape as a code diff. The skill picks the playbook from what the agent produced, then matches your user, project, or Tailwind plus daisyUI design system before writing any HTML (skills.sh registry). That ordering matters: styling decisions come from your repo, not from the skill's taste.

Where does Lavish fit next to other design skills?

It is a review layer, not a generator. The skills that produce the interface itself, ranked by installs, are covered in our guide to designing a UI with Claude Code. Lavish sits after them: build the screen, then use Lavish when you need a human to mark up an artifact the agent produced. Its own catalog classification is Design and UI/UX, which places it in the frontend and design category alongside 59,500 catalogued skills overall (Skillselion catalog).

A complete starter stack

1. lavish (16,472 installs) as the review surface for every non-trivial agent deliverable.
2. A design-system skill so the artifact inherits real tokens rather than defaults; see the design systems hub.
3. lavish-design (103 installs) only if you want the design-focused variant; it is early and thinly installed.

Common pitfalls

• Leaving .lavish/ untracked and then wondering where the artifact went. Decide early whether those pages are throwaway or part of the review record, and set your gitignore accordingly.
• Using it for short answers. A three-line reply does not need a browser session; the overhead only pays off on diagrams, tables, plans, and diffs.
• Expecting it to work without a design system in the repo. It matches user, project, or Tailwind plus daisyUI conventions, so a project with none of those gets a generic page.

Install it before your next long planning session and review the plan as a page instead of a scroll; the annotation loop is the part that changes how you work.

Common questions