根据 PRD 生成设计方案：把需求文档落到可开发的技术设计
SkillHub 社区 Skill「根据prd生成设计方案」（@user_df16bbeb/prd-to-tech-design，v1.0.7）面向 PRD 到技术设计的衔接：解读需求与边界、扫描仓库识别技术栈与既有接口表结构、生成符合项目规范的设计文档，并对模糊与冲突点标注「需人工确认」，生成后做一致性、完整性、可行性自检。无需 API Key。可通过 skillhub.cn 安装指引或 skillhub CLI 装入 Cursor 等 Agent 的 skills 目录；适合在提供 PRD 链接或片段、要求 tech design/接口方案时主动触发。
Read More项目文档生成专家：从代码仓库批量产出七类标准文档
项目文档生成专家是 SkillHub 社区 Skill（@user_dd91c74f/project-doc-generator，v1.0.3），无需 API Key。它引导 Agent 分析项目代码并生成需求规格、概要/详细设计、数据库设计、API 文档、测试计划、部署与用户手册等七套标准文档，并支持版本管理。用户可说「生成项目文档」或「形成项目文档」触发。文章说明安装方式：skillhub CLI 配合 --dir 安装到 Cursor 等 skills 目录，或让 Agent 按 skillhub.cn/install/skillhub.md 安装，并提示校对生成结果与正确安装路径。
Read Moredsh-serena: Injecting IDE-Level Symbolic Code Intelligence into the DeepSeek Harness Agent
The dsh-serena plugin from DeepSeek Harness (DSH) integrates the Serena MCP server to give agents IDE-level symbol navigation, including finding symbols and references, renaming, and code overviews. It runs fully locally with no external service dependencies, and requires the Serena CLI to be installed and the Hooks file configured. It is recommended to start the session with `--project-from-cwd`; the agent then automatically activates the project and performs detailed structured analysis. The plugin is compatible with DSH 0.2.0-rc.2. The first query may require downloading the corresponding language server, which can be slow.
Read Moredsh-gitnexus-plugin: DeepSeek Harness Code Intelligence Knowledge Graph Integration
The `dsh-gitnexus-plugin` is a plugin for DeepSeek Harness (DSH) designed to help agents understand codebase structure. It uses GitNexus to index a codebase as a knowledge graph and provides `mcp__gitnexus__*` tools supporting concept search, context retrieval, impact analysis, taint discovery, change detection, and renaming. Installation requires Node.js >= 20 and the GitNexus CLI. The plugin spawns an MCP server over stdio transport and supports configurable timeouts under the MIT license.
Read Moredsh-codegraph-plugin: Exposing codegraph as 8 Code Intelligence Tools
The dsh-codegraph-plugin for DeepSeek Harness wraps the codegraph CLI into eight tool interfaces, providing DSH with structured code context support. It depends on a host-installed codegraph CLI, maintains local indexes using a SQLite WAL mechanism, and offers capabilities such as symbol search, call relationships, impact analysis, and code reading. The plugin includes fault tolerance and supports configuration that guides the model to prioritize code graphs over grep, making it suitable for scenarios that require deep understanding of code structure and dependency relationships.
Read Moredsh-plugin-codegraph: Structured Code Intelligence for the DeepSeek Harness
`dsh-plugin-codegraph` is a plugin for DeepSeek Harness designed to address the need for structured queries about code—such as declaration locations, call relationships, and modification impacts—by constructing a local code symbol graph. The plugin provides two independent tools: `codegraph` (read-only queries) and `codegraph_index` (index construction), and supports multiple languages, including TypeScript, Python, and Go. Its core design separates indexing from querying to ensure rapid query responses without implicitly triggering time-consuming construction tasks, enabling agents with precise, non-speculative code understanding.
Read MoreIntroducing the DeepSeek Harness Plugin: dsh-lens
The DeepSeek Harness (DSH) plugin `dsh-lens` is designed to extend workflows with modular capabilities and provide a standardized analysis process. It offers five core features: project structure reconnaissance, orchestration pain point mapping, runtime policy diagnostics, sensitive information scanning, and pluginization potential assessment. As a guided tool, it generates reports in the `dsh-scan/` directory without modifying code or performing automatic deployment. Installation is completed via CLI commands, and the skill directory is mounted to `@deepseek-ai/dsh-tool-skill`.
Read Moredsh-repo-analyzer: Intelligent Local Repository Analysis
`dsh-repo-analyzer` is a local repository intelligence plugin designed to provide DeepSeek Harness agents with an understanding of codebase structure through static analysis. It requires no external services and runs entirely locally. The plugin offers three core functions: `repo_scan`, which scans repository inventory and identifies the tech stack and directory hierarchy; `repo_deps`, which parses and queries project dependencies; and `repo_refs`, which resolves import/require relationships and generates a module reference graph. Configuration is flexible, supporting custom traversal depth, file counts, and exclusion paths. With these tools, agents can efficiently retrieve stack information, dependency references, and module locations, quickly building a clear understanding of the code structure.
Read Moredsh-kgraph-plugin: Project Knowledge Graph Visualization Plugin
dsh-kgraph-plugin is a read-only bundle plugin for DeepSeek Harness, designed to help developers quickly understand the architecture, dependencies, and layered design of complex projects by automatically generating module-level and file-level knowledge graphs. It supports multiple languages (Java, TypeScript, Python, Go, etc.) and multi-module recognition, providing a visual dashboard and the `project_kgraph` tool to help users refactor and optimize code more efficiently. After installation, users can simply input a project path to scan it and view the module dependency graph and file-level layered structure, without modifying the source code at any stage.
Read Moredsh-understand-anything: A DeepSeek Harness Plugin for Generating Codebase Knowledge Graphs
`dsh-understand-anything` is the codebase analysis plugin for DeepSeek Harness. By generating structured knowledge graphs, it addresses the issues of traditional RAG consuming excessive amounts of tokens and easily losing architectural details. The plugin supports architecture understanding, change impact analysis, onboarding guides, and multi-domain extraction without requiring raw files to be stuffed into the context. Installation requires Node.js ≥ 22 and pnpm ≥ 10. Please note that, due to plugin compatibility limitations, Web applications, Claude Code–specific features, and the analysis engine have not been translated.
Read More