用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/curiositech/windags-skills --skill dag-visual-editor-design命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| license | BSL-1.1 |
| name | dag-visual-editor-design |
| version | 1.0.0 |
| description | Design modern, intuitive DAG/workflow visual editors that feel like LEGO, not LabView |
| category | Agent & Orchestration |
| tags | ["dag","visual-editor","design","ui","graph-editing"] |
| trigger_phrases | ["design dag editor","workflow builder ui","node graph ux","visual programming interface","make dag editor intuitive"] |
| allowed-tools | ["Read","Write","Edit","WebSearch","WebFetch"] |
Design modern, intuitive DAG and workflow visual editors following the LEGO philosophy: snap blocks together simply rather than wire complex ports.
Node Count < 20 AND Simple Flow?
├─ YES → Force-directed layout (React Flow default)
│ • Fast rendering
│ • Good for exploring connections
│
└─ NO → Node Count > 100?
├─ YES → Hierarchical (Dagre) + Virtualization
│ • Set viewport culling
│ • Lazy load node details
│
└─ NO → Branch Factor > 3 per node?
├─ YES → ELK Layered Algorithm
│ • Handles complex routing
│ • Minimizes edge crossings
│
└─ NO → Dagre LR (Left-Right)
• Standard choice
• rankdir: 'LR', ranksep: 80
Data Type Diversity?
├─ Single type (e.g., JSON) → Implicit connections
│ • No handles visible by default
│ • Snap zones on hover
│
├─ 2-3 types → Color-coded handles
│ • Red: Error streams
│ • Blue: Data streams
│ • Green: Success/completion
│
└─ 4+ types → Bundled connections
• Group related channels
• Label bundles clearly
• Consider type coercion nodes
User Skill Level?
├─ Beginner → Canvas + Sidebar
│ • Drag nodes from categorized list
│ • Template-based workflows
│
├─ Intermediate → Quick Add (Slash Commands)
│ • Type "/" for node search
│ • Context-aware suggestions
│
└─ Expert → Keyboard First
• Hotkeys for common nodes
• Text-based node creation
• Batch operations
Symptoms: Edges crossing everywhere, impossible to follow data flow, users getting lost Detection: If >30% of edges cross other edges, or users spend >20s tracing a path Fix:
Symptoms: Pan/zoom feels broken, users can't find their content, minimap unhelpful Detection: Users hitting zoom limits frequently, >5 seconds to locate nodes after navigation Fix:
Symptoms: Users connecting wrong ports, type errors, unexpected data flow Detection: >20% connection error rate, frequent undo of connections Fix:
Symptoms: Editor freezes with >50 nodes, stuttering during pan/zoom Detection: Frame rate drops below 30fps, render times >100ms Fix:
Symptoms: Users don't know if workflow is running, what failed, or why it stopped Detection: Users asking "is it working?" or clicking run button multiple times Fix:
Scenario: Design editor for CSV → Transform → Database pipeline
Step 1: Choose Layout
rankdir: 'LR', ranksep: 120 for readable spacingStep 2: Design Node Structure
const TransformNode = ({ data }) => (
<div className="w-64 border-2 border-gray-200 rounded-lg bg-white">
<div className="bg-blue-50 px-3 py-2 border-b">
<h3>🔄 Transform Data</h3>
</div>
<div className="p-3">
<div className="text-sm">Filter: {data.filter}</div>
<div className="text-sm">Sort: {data.sort}</div>
</div>
<Handle type="target" position={Position.Left} />
<Handle type="source" position={Position.Right} />
</div>
);
Step 3: Connection Logic
Novice Miss: Would add separate handles for each column Expert Catch: Keeps single connection, shows column mapping in node detail
Don't use DAG editors for:
Delegate to other skills: