ML

mouse-lin/finesse-brief

Developer tools

This page was translated automatically. Technical details may differ from the original. View the original

124 stars Quality 64 Trend 22MIT

The "Workspace Architect" skill for AI coding assistants. It doesn't write interfaces — it answers *what exactly this thing is*, then outputs a ready-to-start copy to finesse-ui or any AI that can write pages.

Overview

The "Workspace Architect" skill for AI coding assistants. It doesn't write interfaces — it answers *what exactly this thing is*, then outputs a ready-to-start copy to finesse-ui or any AI that can write pages. One skill, only one thing: organize a vague requirement ("pet ownership," "I want to build a CRM") into a complete workbench definition—positioning, core modules, home screen structure, page hierarchy, data model, and the first sentence you see when you open it, all written into .workbench/spec.md. The output is always ** (H5 in a mobile browser or a web console in a computer browser), not an app. It covers two scenarios: - —— Daily follow-up page for each person, one field (menstruation · Muscle gain · Pet care · Bookkeeping · Setting up a stall · Learning with children): Say a word and get a complete solution - —— Multi-module workbench centered around a business object (CRM · ERP · AI Agent Console · Data Kanban · Smart Factory · Project Management · E-commerce Backend... ): Ask a couple of questions before giving the full specs Once you get the spec, hand it to finesse-ui (or Cursor / v0 / Tongyi / any AI that can write pages) and you can directly build the interface. If you want to understand the methodology behind it (how to find evidence, how to divide the 11 structures, how to bind fields, common dead zones), check skills/finesse-brief/references/, a total of 9 articles. After installation, say in the dialogue "Help me get a workbench," "Raise a pet," or /workbench to trigger it. —— See installation & tool support below. I recommend npx skills; a single command applies to all agents (which tool you can choose to install at runtime). Manual installation is in the folded area below. The rules are hung on .workbench//spec.md, or you can directly @ it in the conversation.

Overview (machine translated)

Chinese | English

Think carefully about what to do first, then think about how to look good

The “Workspace Architect” skill for AI coding assistants. It doesn’t write interfaces—it answers what exactly is this, then outputs a Workbench Spec ready to start working right away, handed to finesse-ui or any AI that can write pages.


What is it?

One skill, only one thing: organize a vague requirement (‘pet ownership’, ‘I want to build a CRM’) into a complete workbench definition—positioning, core modules, home screen structure, page hierarchy, data model, and the first sentence you see when you open it, all written into .workbench/spec.md.

The output is always webpages (H5 in mobile browsers or web consoles in computer browsers), not apps. It covers two scenarios:

  • Personal Domain — Daily follow-up pages for each person, one field (menstruation · Muscle gain · Pet care · Bookkeeping · Setting up a stall · Accompanying children in learning): Say a word and provide a complete plan
  • System Domain — A multi-module workbench (CRM · ERP · AI Agent Console · Data Kanban · Smart Factory · Project Management · E-commerce Backend… ): Ask a couple of questions before giving the full spec

Once you get the spec, hand it over to finesse-ui (or Cursor / v0 / Tongyi / any AI that can write pages) and you can directly create the interface. If you want to understand the methodology behind it (how to find evidence, how to divide the 11 structures, how to bind fields, common dead ends), see skills/finesse-brief/references/, a total of 9 articles.


How to use it

After installation, say in the dialogue “Help me get a workbench,” “Keep a pet,” or /workbench to trigger it. You can install without Claude Code — see the installation & tool support below.

| You said | What will happen | |—|—| | Domain terms like ‘pet ownership’ | Directly give you a complete plan to reject it, without first asking you three questions in reverse | | “Show me the effect first / Let’s just build it and let me see” | Go sketch: Skeleton → spec → Directly hand it over to finesse-ui to build the interface | | “I want to make a workbench” (didn’t say what it was about) | Just one question: What is this console mainly responsible for | | “I want to make a CRM / Agent console / factory dashboard” | Ask a couple of questions (what are you going around · Who uses it · What do you do most often every day), and once again, give the full spec | | Completely hard to say | Using evidence method: ask about the app you’ve already been using Notes/Excel to get by, or the app you deleted after ten days |

Command

Command What to do
sketch [两个字] Fast Lane — Skeleton → spec → Create interface directly, marked as the starting point
discover Starting from zero, when you don’t even have a direction

This page was translated automatically. Technical details may differ from the original.

Recommended Tools

Try a different keyword or remove a filter.

Install

npx skillfish add mouse-lin/finesse-brief
New to MCP? Start with the mcpskill.org guide →