视觉理解-UI设计分析：为 Agent 接入截图理解与结构化输出
视觉理解-UI设计分析（@user_ae22680b/visual-ui-understander，v2.0.0）是 SkillHub 社区 design-media 类 Skill，面向 CodeBuddy/WorkBuddy/OpenClaw 等 Agent。通过 understand_image.py 调用腾讯混元，支持 UI 设计分析、OCR、无障碍检查、多图输入与 --json 结构化输出。文章说明 skillhub 安装到各客户端 skills 目录、TENCENTCLOUD 密钥与混元开通、典型 CLI 用法及多图与配额等注意事项。
Read MoreVisionForge: Visual Understanding and Image Generation Plugin for DeepSeek Harness
VisionForge is an open-source visual plugin for DeepSeek Harness (DSH), designed to provide a unified visual processing solution for model inference through the “everything is a plugin” philosophy. It offers “paste-to-read” visual understanding, supports multi-provider visual engines with failover, and integrates features such as image generation, editing, upscaled downloads, and local cache management. Users can install it via npm and, after configuring API keys, perform full-chain operations—including image analysis, generation, and editing—directly within conversations.
Read Moredsh-image-describer: Equipping DeepSeek Harness's Text-Only Models with Visual Understanding Capabilities
The `dsh-image-describer` plugin in DeepSeek Harness (DSH) intercepts the message stream and converts image content into structured text, enabling text-only LLMs to understand images. It supports dual-mode interaction, including pasting images into the chat box (attachment mode) and referencing local file paths in the conversation (Tool mode), and automatically performs OCR and content recognition. Before use, enable the image input channel for the text model in the configuration file and specify a multimodal vision model, such as MiniMax-M3, along with its API key.
Read Moredsh-vision-android: DeepSeek Harness Visual Understanding and Android UI Automation Plugin
dsh-vision-android is a plugin based on DeepSeek Harness, designed to provide agents with multimodal visual understanding and Android UI automation capabilities. The plugin integrates 11 tools, primarily addressing two main needs: sending local screenshots or images to vision models for analysis, and executing real-time click, swipe, input, and key operations on Android devices via ADB. Its core functions are divided into vision understanding (such as `vision_ask`, `mobile_look`, screenshots, etc.) and UI automation (such as `mobile_tap`, `mobile_swipe`, text input, APK installation, app launching, etc.). When used, agents can form a workflow of "screenshot analysis → locate element → execute operation →"
Read MoreDSH-LLM-Vision-Bridge: Integrating Image Understanding into DeepSeek Text Models in the DSH Web GUI
dsh-llm-vision-bridge is a DSH plugin licensed under MIT, designed to address the issue of text models being unable to directly process images. This plugin registers the `deepseek-vision` provider in DSH's `LlmAdapter`. When a request includes images, it automatically routes them to a configured vision model (such as pi-ai or a local llama.cpp) to generate text descriptions, which are then used by the DeepSeek text model to continue the conversation based on the descriptions. If no images are present, the request is directly passed through to a fallback provider for zero-overhead operation. Key features include an LRU cache to avoid duplicate descriptions, automatic retries for 503/429 errors and timeouts, and configurable failure policies (placeholder or error). After installation, the web service must be restarted, and configuration is required in the settings.
Read More