长期记忆 / 知识图谱 会思考的进化 AI：为 Agent 搭统一记忆底座
SkillHub 社区 Skill「长期记忆 / 知识图谱 会思考的进化 AI」（@user_1a470ba8/zwjh-skill，v2.6.0）面向 AI Agent，提供跨会话长期记忆、实体关系图谱、自动沉淀，以及语义与时间线检索、记忆健康度审计和本地备份（可接百度网盘）。标注 requires_api_key=false，强调纯本地与零密钥。文章说明其解决上下文断裂的痛点，列出官方核实功能，并给出基于 skillhub CLI 在 Cursor、Claude Code、Codex 等目录下的安装命令与 SkillHub 安装文档链接。
Read More音频转文本整理稿（本地部署版）：本地 ASR 与中文整理稿输出
介绍 SkillHub 社区技能「音频转文本整理稿（本地部署版）」（@user_695efa3d/transcript-organizer，v1.0.0）。技能接收音视频、播客链接、字幕或已有转写稿，在本机用 faster-whisper 或 mlx-whisper 完成 ASR，经清洗整理后输出分享式提纯版中文原文到本地 .txt，无需 API Key。文章说明核心能力、SkillHub CLI 安装路径（含 Cursor 等 skills 目录）及典型对话用法，并列出适用场景与本地部署注意点。
Read Moredsh-lessons: Recording AI Errors and Preventing Recurrence
Dsh-lessons is an open-source plugin for DeepSeek Harness (DSH), designed to address the issue of locally running models ignoring context or repeatedly making the same mistakes. The plugin captures user feedback—either automatically detected or manually logged—to generate a “lessons list,” which is dynamically injected into system prompts in subsequent conversations to prevent the model from repeating past errors. It supports visual dashboard management, session-level toggles, and configurable settings, helping developers continuously optimize local agent performance.
Read Moredsh-comfyui-image: Run ComfyUI Locally to Generate Multimodal Assets
The DeepSeek Harness plugin `dsh-comfyui-image` enables agents to directly control local ComfyUI, leveraging GPU resources to generate images, videos, audio, and 3D assets without cloud API costs. The plugin supports 116+ workflow templates across 67 model families, providing core features such as text generation, template browsing, status reporting, and skill decision-making. Installation requires Harness 0.2.0-rc.2 or later, and supports automatic detection of ComfyUI status or launching headless mode. Environment variables allow flexible configuration of the port and model paths. Models are downloaded from mirror sites by default, and local generation tasks can be invoked directly using natural language instructions.
Read Moredsh-local-hanaccount: Access Gateway Plugin for Local DSH Web
`dsh-local-hanaccount` is a single-operator access gateway plugin designed for local DeepSeek Harness (DSH) web profile configurations. It provides a unified login entry point, supports password, passkey, and SSH key authentication, and includes IP allowlist/blocklist, session management, and Nginx integration features. By default, the plugin protects all business routes, follows a fail-closed policy, and does not interfere with DSH’s native session management. It is suitable for scenarios requiring a controllable access lock for local web services.
Read MoreDeepSeek Harness Local Qwen Model Adapter (dsh-llm-qwen-local)
`dsh-llm-qwen-local` is a plugin-centric framework for deploying Qwen models locally and integrating them into the DSH ecosystem. The plugin provides an OpenAI-compatible interface through a vLLM backend, supports multimodal toggles, inference level configuration, and image projection, and features zero runtime dependencies along with a visual Web configuration interface. To use it, DSH version 0.1.2-rc.1 or later is required, and vLLM startup parameters must be configured correctly. The plugin is released under the MIT License and aims to simplify the process of integrating local models.
Read Moredsh-qwen38-local-qol: DeepSeek Harness Local Qwen3.8 Optimization Plugin
`dsh-qwen38-local-qol` is a QoL plugin for DeepSeek Harness designed to address the pain points of managing reasoning budgets and context compression when deploying the Qwen3.8 model locally. It supports four backends—llama.cpp, NInfer, TabbyAPI, and oMLX—and provides a unified configuration entry point along with the ability to adjust the reasoning budget on a per-request basis. It also improves compression strategies, supporting the pruning of prefill content when compression is triggered while preserving recent reasoning and image blocks nested within tool results. Users can install the plugin via the command line and configure the server, context window, reasoning budget, and compression strategy in real time through the settings panel.
Read Moredsh-connect-trae: Connect to Local Trae Models and Manage Usage
DSH Connect Trae is a standalone bundle plugin designed to integrate local Trae models with the DSH ecosystem, addressing unified account management and status querying. The plugin supports concurrent access to both domestic and international providers, with features including automatic multi-account detection, secure token isolation, and multi-session switching. Its model coverage includes mainstream models available through the Trae SOLO channel, and it supports tool calling and image recognition (limited to multimodal models). All queries made by the plugin are read-only, and only check-ins consume quota. DSH version 0.1.7 or later is required.
Read Moredsh-tea-store: Tea Storage Guide
`dsh-tea-store` is a DeepSeek Harness (DSH) plugin maintained by istone, focusing on providing tea storage guidance. It is implemented in pure Node.js, requires no network access, and depends only on `@deepseek-ai/cordis` and `@deepseek-ai/dsh-tools`. It can be installed using `dsh plugin add github:uckkk/dsh-tea-store`. Once installed, you can directly invoke the tools it provides within DSH sessions to obtain tea art solutions.
Read Moredsh-generate-image: Generate and Edit Images Using Local Codex CLI
The DSH plugin `dsh-generate-image` introduces local visual capabilities for DeepSeek models by leveraging a locally logged-in Codex CLI to perform text-to-image and image-to-image generation. The plugin can be invoked directly through the `generate_image` tool, without requiring an API key or a build step. Generated images are saved in the directory created by Codex. Installation requires a local Codex login and Node.js >= 18, and it can be completed with a single npx command.
Read MoreDeepSeek Harness Plugin: dsh-chromium (Chromium Element)
The DeepSeek Harness (DSH) plugin `dsh-chromium`, a pure Node.js implementation based on the Cordis architecture, is designed for local deployment or offline scenarios. It relies on the dependencies `@deepseek-ai/cordis` and `@deepseek-ai/dsh-tools` and supports running in environments without internet access. The plugin executes with the permissions of the current process and therefore carries permission-related risks; users should ensure its source is trustworthy. As an MIT-licensed localization extension tool, it provides localized execution capabilities for the DSH ecosystem.
Read MoreDSH LM Studio Router: A DSH Plugin to Route Token-Intensive Tasks to Local LM Studio
`dsh-lmstudio-router` is a DSH Web GUI plugin designed to intelligently route token-intensive remote tasks (such as text compression and title generation) to local LM Studio to reduce costs. Features include automatically starting LM Studio, context adaptation, and a forced routing marker `[local]`. Users must install it via the command line, specify the local API endpoint and model in the configuration file, and after restarting the service, use the `[local]` marker in messages to route specific requests to local processing.
Read MoreConnect local DeepSeek Harness to DingTalk
`@dingtalk-real-ai/dsh-dingtalk` is the DingTalk connector plugin for DeepSeek Harness (DSH), designed to connect local DSH agents to DingTalk. It utilizes Stream connection technology, eliminating the need for public network inbound endpoints while ensuring that sensitive credentials, such as QR codes and Client Secrets, are processed exclusively via a separate terminal to prevent them from entering machine-readable output or Checkpoints. The plugin supports commands such as `plan`, `apply`, and `resume`. Checkpoint permissions are restricted to 0600, and the storage of secret fields is strictly prohibited. During installation, version numbers must be pinned, and configuration plans should be generated and applied via JSON mode; sensitive steps require the manual execution of the Resume operation within an interactive terminal. This plugin requires Node.js.
Read MoreDSH Plugin plugin-switch: Enable or Disable Plugins Directly in the Plugin Manifest
plugin-switch is a local web plugin for DeepSeek Harness (DSH), designed to simplify plugin management. It allows users to enable or disable plugins in real time via a page slider without restarting the service, and the status is maintained after restart. Core features include real-time effect, group filtering, batch operations in a single transaction, automatic backup with one-click undo, strong confirmation for risky actions, and multi-tab synchronization. Installation requires Node ≥ 22 and a DSH web profile environment. It is recommended to use the `dsh plugin` command to install from GitHub, or alternatively via PowerShell scripts or Git clone, but care should be taken to avoid duplicate installations. Typical use cases include temporary enable/disable, batch adjustments, and reverting accidental operations. If the toggle page disappears, you can modify the local configuration file or use CLI tools.
Read MoreFastAPI + Uvicorn: Basic Configuration for Local Development and Deployment
This article introduces the web development and deployment process using FastAPI with Uvicorn. FastAPI is a high-performance Python framework supporting asynchronous operations and automatic API documentation. Uvicorn, as an ASGI server, is the recommended deployment tool for FastAPI, and their combination enables efficient development. **Environment Installation**: First, create a virtual environment (e.g., `python -m venv venv`), activate it, and then install dependencies with `pip install fastapi uvicorn`. **Development Configuration**: Write `main.py`, define routes (e.g., root route `/` and parameterized route `/items/{item_id}`), and start the server with `uvicorn main:app --reload` for auto-reloading in development mode. Verify the interface by accessing `http://127.0.0.1:8000`. **Production Deployment**: Use the basic command `uvicorn main:app --host 0.0.0.0 --port 8000`. Specify the number of worker processes with `--workers` for multi-processing. Ensure the deployment server opens the port and manage processes via `nohup` or `systemd`. **Common Issues**: For port conflicts, change the port. If the service is unreachable, confirm `--host 0.0.0.0` and firewall settings. If installation fails, update pip or check Python version compatibility.
Read More