AGENT SKILL

flow-ai-chat

A linked AI assistant flow: empty chat with suggested prompts, a streaming reply, an error with retry, and a result the user acts on, as one collection.

License 许可
MIT
最近更新
2026-11-14
Stars
★ 1.1k
Forks
104

01 前言 Front Matter

od
{ "mode": "prototype", "craft": { "requires": [ "product-ui", "state-coverage", "action-states", "ux-writing" ] }, "preview": { "type": "html", "entry": "index.html" }, "surface": "web", "platform": "desktop", "scenario": "design", "design_system": { "requires": true }, "example_prompt": "The AI assistant inside a travel-planning app: empty chat with suggestions, a streaming answer, a network error with retry, and an itinerary card the user can save." }
name
flow-ai-chat
tags
[ "ai chat", "assistant", "chatbot", "copilot", "streaming", "AI 助手", "聊天" ]
emoji
💬
en_name
AI Chat Flow
zh_name
AI 对话流程
category
app
scenario
design
triggers
[ "ai chat", "chat screen", "assistant screen", "chatbot ui", "copilot ui" ]
description
A linked AI assistant flow: empty chat with suggested prompts, a streaming reply, an error with retry, and a result the user acts on, as one collection.
en_description
A linked AI assistant flow: empty chat with suggested prompts, a streaming reply, an error with retry, and a result the user acts on, as one collection.

02 资源文件 1

03 SKILL.md 正文

<!-- Kit skill (openspec add-app-screens). -->

AI Chat Flow

A flow pack: one collection of linked screens, each with its branches, made in one ask. Build every screen from the app screen patterns' shell and rules (below), one artifact per screen.

Screens

Screen id Title What it holds
chat-empty Ask anything what the assistant can do here in one line; three or four suggested prompts from the brief's domain; the composer
chat-streaming Answering the user message, the reply arriving with a visible "Stop" control and the composer locked
chat-error Something failed the failed reply with a plain cause and "Try again"; the user's message kept
chat-result Result the reply as a structured card the user can act on (save, share, edit); feedback controls; ends the flow

Main path: chat-empty → chat-streaming → chat-error → chat-result. Add the branches each screen lists as its own data-od-state frames or, when a branch is a whole screen, as a screen.

Workflow

  1. Ask once, at most three questions, together: which branches to include, which platform (web app, iPhone, Android), and the product name if the brief doesn't give it. Accept "you pick". Never more than eight screens.
  2. Plan the flow before the first screen. Pass flow to prepare_open_design_brief on every screen: the screens above (trimmed to the user's choice), the links between them (from, trigger label, to) and the shared navigation if the flow has one. Use the same collectionId and the screen id as screenRole.
  3. Build each screen with its own prepare_open_design_brief call, then register it with the same collectionId, screenRole and flow.
  4. Check the collection with check_open_design_artifact and collectionId, fix flow findings (broken links, dead ends, unfilled screens, navigation drift, store rules), then show the flow map.

Rules for this flow

  • Say what the assistant can't do and where its information comes from; show sources when it cites any.
  • Long replies keep the composer reachable; the newest message stays in view.
  • Mark the result screen with data-od-end.

Markup contract (shared by every app screen)

  • States. Each state of the screen is a sibling <section data-od-state="…">: default (populated), and the ones this pattern lists below from empty, loading, error, success, no-access, long-content. Only default is visible; the others carry hidden. The kit's check renders every state, and the preview shows them side by side.
  • Slots. Keep the seed's data-od-slot names (nav, toolbar, content, empty, loading) so states, links and comments line up.
  • Links between screens. When this screen is part of a planned flow, link to another screen with its entry path in href and its screen id in data-od-goto, and use the shared navigation the brief gives instead of drawing your own.
  • The app's own components. When the brief lists the app's components, build each part from one of them and mark it data-od-component="<Name>"; mark a part with no match data-od-component="new:<Name>".
  • Sample data. Records come from the data file the brief describes, never typed into the markup row by row: one template row or card with data-od-repeat="<list>" and data-od-bind=".<field>" inside it, single values with data-od-bind="<list>[n].<field>", and data-od-format for money, dates and numbers. Every screen that shows the same record binds to the same one.
  • Tokens only. Every colour, size, radius and shadow comes from the design system's :root variables. No raw hex outside :root.
  • Platform. Web app: the shell as given. iPhone, Android, tablet or desktop app: put data-od-device on <body> (iphone, android, tablet with data-od-orientation, or desktop-app with data-od-window="mac" or "windows"); the kit draws the device frame, status bar and home indicator around the screen, so never draw them yourself. iPhone or Android: a single column 390 px (iPhone) or 412 px (Android) wide, with a bottom tab bar instead of the side nav, a top bar with the title and a back button on inner screens, and 44 px (iPhone) or 48 px (Android) touch targets. Android screens use Material patterns (top app bar, floating action button where there is one main create action). Tablet: the sidebar collapses to an icon rail with labels in a tooltip, content in two columns where the pattern has a detail pane.
  • Density. Use the density the brief gives (compact, medium or airy) with the numbers from the product UI craft rules.

Seed

Start from this shell (one HTML file, inline CSS). Replace the comments; keep the structure.

<!-- od:app-shell:start -->

<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Screen title · Product</title>
<style>
/* Paste the active design system's :root block here, then use only var(--…). */
*,*::before,*::after{box-sizing:border-box}
body{margin:0;background:var(--bg);color:var(--fg);font:var(--text-base)/var(--leading-body) var(--font-body)}
.app{display:grid;grid-template-columns:1fr;min-height:100dvh}
.app-nav{display:flex;gap:var(--space-1);padding:var(--space-2) var(--space-4);border-bottom:1px solid var(--border);overflow-x:auto}
.app-nav a{display:flex;align-items:center;gap:var(--space-2);min-height:44px;padding:0 var(--space-3);border-radius:var(--radius-md);color:var(--muted);text-decoration:none;white-space:nowrap}
.app-nav a[aria-current="page"]{background:var(--surface);color:var(--fg);box-shadow:var(--elev-ring)}
.app-main{min-width:0;padding:var(--space-4)}
.app-toolbar{display:flex;flex-wrap:wrap;align-items:center;gap:var(--space-3);margin-bottom:var(--space-4)}
.app-toolbar h1{margin:0 auto 0 0;font:600 var(--text-xl)/var(--leading-tight) var(--font-display)}
[data-od-state][hidden]{display:none}
@media (min-width:64rem){
  .app{grid-template-columns:260px 1fr}
  .app-nav{flex-direction:column;border-bottom:0;border-right:1px solid var(--border);padding:var(--space-4)}
  .app-main{padding:var(--space-6) var(--space-8)}
}
</style>
</head>
<body>
<div class="app">
  <nav class="app-nav" aria-label="Main" data-od-slot="nav">
    <a href="#" aria-current="page">Current place</a>
    <a href="#">Another place</a>
  </nav>
  <main class="app-main">
    <header class="app-toolbar" data-od-slot="toolbar">
      <h1>Screen title</h1>
    </header>
    <section data-od-state="default" data-od-slot="content"><!-- the populated screen --></section>
    <section data-od-state="empty" data-od-slot="empty" hidden><!-- what will be here, why it is empty, the first action --></section>
    <section data-od-state="loading" data-od-slot="loading" hidden aria-busy="true"><!-- a skeleton shaped like the content --></section>
  </main>
</div>
</body>
</html>

<!-- od:app-shell:end -->

04 同仓库更多 Skills

feimacode
feimacode/open-design-agent-kit

youtube-clipper

YouTube clip generation and editing with automated workflows — pull source video, slice highlights, add captions, and export.

★ 1.1k
feimacode
feimacode/open-design-agent-kit

writing-guidelines

Review docs/prose for Writing Guidelines compliance. Use when asked to "review my docs", "check writing style", "audit prose", "review docs voice and tone", or "check this page against the writing handbook".

★ 1.1k
feimacode
feimacode/open-design-agent-kit

wpds

WordPress Design System. Apply WordPress's official design tokens, typography, and component patterns to themes and sites.

★ 1.1k
feimacode
feimacode/open-design-agent-kit

weread-year-in-review-video-template

WeRead-inspired HyperFrames video template for vertical annual reading reports, personal reading dashboards, book-note recaps, and shareable year-in-review stories. Use when users want a 9:16 HTML-to-MP4 reading report with warm paper texture, editorial Chinese typography, book-page metaphors, data highlights, and deterministic motion.

★ 1.1k
feimacode
feimacode/open-design-agent-kit

web-design-guidelines

Review UI code for Web Interface Guidelines compliance by the Vercel engineering team. Covers layout, typography, color, motion, and accessibility for product UI.

★ 1.1k
feimacode
feimacode/open-design-agent-kit

web-clone

网站复刻 / 克隆方法论。USE WHEN 用户说 复刻网站、克隆网站、clone website、抄个站、仿站、 照着这个站做一个、reproduce site、还原某个网页效果、把这个站搬下来改成我的、 复刻某个交互/WebGL/Canvas/Three.js 效果。提供「先拿真源码 → 判路径 → 逆向拆解 → 搭工程 → 替换内容」的可移植决策树,覆盖静态站 / React-Vue-Next 内容站 / WebGL-Canvas 重前端站三大分支,并强制核对任何 AI 二手分析里的可执行代码。

★ 1.1k