HalfAI Workbench

The Problem: From Fragmented Inputs to a Structured Workbench

Many users juggle scattered notes, tasks, reference images, or customer information and wish to organize them into a usable personal or work management interface. The primary hurdle is the technical overhead involved: selecting tools, designing page structures, handling responsive layouts, and managing data flow.

Skill Core: Two Paths and Key Steps

The halfai-gzt skill acts as an AI Agent that takes a user's minimal input—a sentence, list, or screenshot—and transforms it into a genuine, multi-page web workbench. It offers two distinct starting paths:

• Template Path: Suited for users starting from scratch who prefer guidance. It uses a series of non-technical questions (e.g., "What do you mainly want to manage?" "What's the first hassle you want to solve?") to gather intent, then guides the user in selecting a visual starting template (e.g., "Fresh & Bright," "Steady & Powerful"). All templates originate from assets/template-gallery/index.html, preserving their responsive behavior and component logic.
• Custom Path: Ideal for users with a clear idea or reference screenshot. The user describes their need directly (e.g., "Create a workbench to manage clients and orders; the homepage should show today's follow-ups and pending shipments."). The Agent parses this, automatically selects the structure, visual style, and implementation approach, and only asks for minimal clarification on truly critical missing information (e.g., involving costs or authorization).

Regardless of the path, the skill follows an engineering-style build process:
1. Understand & Confirm: Translates user input into an internal workbench plan, summarizes it in natural language, and outlines necessary assumptions.
2. Inspect Environment & Design: Checks the existing project and technical environment. Internally defines page structure, data flow, state management, etc., using references/workbench-spec-template.md as the implementation blueprint.
3. Step-by-Step Build: Constructs the workbench shell, prioritizes the implementation of the highest-value module, adds remaining components, implements states (loading/empty/error), and finally tests responsiveness and basic accessibility.

Boundaries, Assumptions, and Caveats

The skill delivers a functional frontend application prototype, not a final deployed product. Key caveats include:
* Data & Connections: Demo data in the first version is clearly marked. Real connections (e.g., to Excel, Feishu) are only attempted after explicit user authorization and provision of valid data sources; the Agent never initiates connections to third-party services unilaterally.
* Functional Scope: The skill focuses on converting descriptions into usable UI and interaction flows. It does not feature-bloat, replicate brand content from reference images, and only exposes functionality that is verified, simulated, or requires user setup.
* Technical Decisions: The Agent automatically selects technical solutions (e.g., frameworks, storage) based on the current environment, avoiding such choices for the user. All visual specifics (e.g., color codes, fonts) are handled through template logic and automated design decisions.