Dwsy/zai-vision-skill

Última atividade de origem registrada: 23 de janeiro de 2026 às 01:24

Catálogo do SkillsMP atualizado: 9 de novembro de 2026 às 14:22

skills coletadas: 1

Estrelas no GitHub: 3
Forks no GitHub: 0

Instalar com um assistente de IA

Copie este prompt para o assistente de IA que você usa.

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/Dwsy/zai-vision-skill.

README do repositório

Arquivo de origem do README: README.md

Ler o README original

Expandir README completo

# zai-skill 智谱 AI 多模态视觉分析 Claude Skill，提供 UI 转代码、OCR、错误诊断、技术图表理解等 8 个强大的视觉分析工具。 ## Features - ✅ **UI 转代码**: 将 UI 截图转换为代码、提示词、设计规范 - ✅ **OCR 文字提取**: 从截图中提取代码、终端输出、文档文字 - ✅ **错误诊断**: 分析错误弹窗、堆栈和日志，提供修复建议 - ✅ **技术图表理解**: 解读架构图、流程图、UML、ER 图 - ✅ **数据可视化分析**: 分析仪表盘、统计图表，提炼趋势和异常 - ✅ **UI 差异对比**: 对比两张 UI 截图，识别视觉差异 - ✅ **通用图像分析**: 灵活理解任何视觉内容 - ✅ **视频分析**: 支持 MP4/MOV/M4V 格式（最大 8MB） ## Installation ```bash cd ~/.pi/agent/skills/zai-vision # 安装依赖 pip install mcp # 或者使用 uv（可选） uv sync ``` ## Configuration 复制示例配置并编辑： ```bash cp mcp-config.example.json mcp-config.json ``` 编辑 `mcp-config.json` 配置智谱 AI API Key： ```json { "name": "zai-vision", "transport": "stdio", "command": "npx", "args": ["@z_ai/mcp-server"], "env": { "Z_AI_API_KEY": "your-api-key", "Z_AI_MODE": "ZHIPU" } } ``` ## Usage ### List all tools ```bash cd ~/.pi/agent/skills/zai-vision python3 executor.py --list ``` ### Describe a tool ```bash python3 executor.py --describe ui_to_artifact ``` ### Call a tool ```bash python3 executor.py --call '{ "tool": "ui_to_artifact", "arguments": { "image_source": "/path/to/image.png", "output_type": "code", "prompt": "Generate React code with Tailwind CSS" } }' ``` ## Tools | Tool | Description | |------|-------------| | `ui_to_artifact` | UI 截图转代码/提示词/规范/描述 | | `extract_text_from_screenshot` | OCR 文字提取 | | `diagnose_error_screenshot` | 错误诊断和修复建议 | | `understand_technical_diagram` | 技术图表理解 | | `analyze_data_visualization` | 数据可视化分析 | | `ui_diff_check` | UI 差异对比 | | `analyze_image` | 通用图像分析 | | `analyze_video` | 视频分析（最大 8MB） | ## Examples ### Example 1: UI to Code ```bash python3 executor.py --call '{ "tool": "ui_to_artifact", "arguments": { "image_source": "/path/to/ui-screenshot.png", "output_type": "code", "prompt": "Generate React component with TypeScript and Tailwind CSS" } }' ``` ### Example 2: Extract Text ```bash python3 executor.py --call '{ "tool": "extract_text_from_screenshot", "arguments": { "image_source": "/path/to/code-screenshot.png" } }' ``` ### Example 3: Diagnose Error ```bash python3 executor.py --call '{ "tool": "diagnose_error_screenshot", "arguments": { "image_source": "/path/to/error-screenshot.png" } }' ``` ### Example 4: Analyze Diagram ```bash python3 executor.py --call '{ "tool": "understand_technical_diagram", "arguments": { "image_source": "/path/to/architecture-diagram.png" } }' ``` ### Example 5: Analyze Video ```bash python3 executor.py --call '{ "tool": "analyze_video", "arguments": { "video_source": "/path/to/video.mp4" } }' ``` ## Performance | Scenario | MCP (preload) | Skill (dynamic) | Savings | |----------|---------------|-----------------|---------| | Idle | 4000 tokens | 150 tokens | 96% | | Active | 4000 tokens | 5k tokens | - | | Executing | 4000 tokens | 0 tokens | 100% | ## Features - ✅ **渐进式加载**: 仅在需要时加载工具定义，节省 96% 上下文 - ✅ **简化执行**: 直接使用 Python3，无需额外依赖管理 - ✅ **可重现构建**: requirements.txt 确保依赖版本一致性 ## Requirements - Python 3.10+ - mcp package: `pip install mcp` - 智谱 AI API Key ## API Key 获取智谱 AI API Key: 1. 访问: https://open.bigmodel.cn/ 2. 注册账号 3. 创建 API Key 4. 在 `mcp-config.json` 中配置 ## Documentation - [智谱 AI 官方文档](https://docs.bigmodel.cn/cn/coding-plan/mcp/vision-mcp-server) - [MCP 规范](https://modelcontextprotocol.io) ## License MIT ## Credits Based on [@z_ai/mcp-server](https://www.npmjs.com/package/@z_ai/mcp-server) by Z.AI

Carregando o README do repositório…

Skills neste repositório

Desenvolvedores de software

1 categorias ocupacionais · 100% classificado

skill

ocupação

descrição

atualizado

zai-vision

Desenvolvedores de software

Dynamic access to zai-vision MCP server (8 tools, transport: stdio)

Idioma do texto original: inglês

2026-01-23

Mostrando 1 de 1 skills coletadas.