sme-ui-ux-engine
Ce que ça fait
Use when editing or refactoring frontend React views, modifying container widths, adjusting dark-mode CSS variables, styling user chat input elements, and handling real-time data lookups.
L'installation ouvre cette fiche dans votre application AgentsRoom. Si l'application n'est pas encore installée, vous serez redirigé vers la page de téléchargement.
SKILL.md
---
name: sme-ui-ux-engine
description: Use when editing or refactoring frontend React views, modifying container widths, adjusting dark-mode CSS variables, styling user chat input elements, and handling real-time data lookups.
---
# SME Platform UI/UX Quality & Accessibility Guardrails
When compiling, refactoring, or evaluating structural visual nodes, you must strictly implement the following execution rules:
### 1. Unified Split-Screen Layout (Anti-Overlap Constraint)
- Do not position the floating AI Chat Widget absolutely over core data tables or hide metrics like "Last Sold Timestamp" or "Status."
- Refactor the primary layout container in `App.js` using Flexbox or Grid properties.
- When the chat panel state is CLOSED, the product data table must fill 100% of the viewport width.
- When the chat panel state is OPEN, the screen must smoothly split: the inventory data grid dynamically constraints to 65% width, and the AI Chat Panel slides out to anchor flush against the right-hand boundary taking exactly 35% width. They must never mask each other.
### 2. High-Contrast Accessibility (Form Input Polish)
- Mitigate dark-on-dark contrast failure violations across the application.
- The chat footer input field area background, boundary text box, and typography placeholders (such as "Ask about a product...") must maintain distinct, high-contrast values against the page background. Text must be easily legible.
### 3. Asynchronous State Synchronization
- Prevent data divergence anomalies (e.g., the data grid flashing a count of 50 while the AI chat outputs a database reading of 45).
- Centralize your fetch responses. When an interactive natural language query executes against `/api/analytics/forecast`, any retrieved inventory numbers must pipe back up into a shared parent state context or global provider hook. Both the table columns and chat bubble vectors must re-render matching metrics instantly.
### 4. String Hygiene & Markdown Parsing
- Do not let unrendered raw markdown strings leak directly into conversational bubbles or list elements.
- Implement an HTML text mapper method. Convert incoming text flags (e.g., `**Coffee Beans**` or `**4.6 days**`) into semantic bold elements (`<strong>` or `<b>`) and cleaner typography spacing.
### 5. Open Sidebar Utilization (CRUD Forms Layout)
- Eliminate extensive empty dark blocks or dead vertical space underneath left sidebar navigation tags.
- Reposition static notification components to the top dashboard navigation cluster. Use the empty left sidebar space to insert an inline, beautifully styled data ingestion wrapper containing standard fields ("Item Name", "SKU", "Initial stock") along with an active "+ Add New Product" button to support quick CRUD data triggers.
Tags
Pour aller plus loin
Claude Ads : le skill Claude Code qui audite tes comptes publicitaires
Claude Ads est un skill open source pour Claude Code : plus de 250 vérifications sur Google, Meta, LinkedIn, TikTok ou Amazon Ads, un score sur 100 et un plan d'action priorisé, en une dizaine de minutes. Installation, commandes, limites, et comment l'orchestrer dans AgentsRoom.
AGENTS.md : un seul fichier de contexte pour tous tes agents de code (Codex, Antigravity, Claude)
AGENTS.md, c'est le fichier d'instructions portable que tes agents de code lisent avant de toucher à ton code. Ce qu'il faut y mettre, en quoi il diffère de CLAUDE.md, et comment garder un seul contexte entre Codex, Antigravity et Claude.
Télécharger AgentsRoom
Lancez tous vos agents IA, sur tous vos projets, depuis une seule fenêtre.
App companion : suivez vos agents en déplacement
Utilisez Claude, Codex, Antigravity CLI ou un autre fournisseur IA.
Remontez bugs et demandes directement dans votre backlog public.