深度研究引擎：可执行的自适应深度研究 Agent
深度研究引擎（@user_00c9b356/deepresearchengine，v1.3.0）是 SkillHub 社区 ai-agent 类 Skill，无需 API Key 标签即可按文档安装。它将研究任务固化为拆子问题、多源并行检索、交叉验证、大纲先行与带证据标注和置信度的报告输出，并给出深度研究等触发词及简单问答、实时数据、主观偏好等排除范围。文章说明 skillhub CLI 安装命令、Cursor 与 Claude Code 等 skills 目录，以及典型提示用法与适用边界。
Read Moredsh-deep-research：在 DSH 里一次调用完成多源检索与证据报告
dsh-deep-research 是 houdinicc 维护的 DSH 深度研究插件，在一次 deep_research 调用或 DeepResearch 面板操作内完成多查询检索、URL 去重排序、正文取证、交叉印证与 Markdown 证据报告；联网经宿主 ctx.web，无独立 API Key。本文介绍定位、与 web_search 的差异、安装命令、参数与配置、面板与历史 API，以及适用场景与安全边界。
Read More深度研究：多视角检索与带引用的结构化调研报告
「深度研究」是 SkillHub 社区 Skill @user_94161289/deep-research-v3（v1.0.0），面向学术级深度调研：多视角网络搜索、交叉验证来源、综合结论并输出带 citations 的结构化报告。无需 API Key。本文说明定位、核心能力，以及通过 skillhub CLI 安装到 Cursor 等客户端 skills 目录的步骤与典型提示用法。
Read MoreDeepSeek Harness Plugin: Kestrel Deep Research Engine
Kestrel is a deep research engine designed for DeepSeek Harness, aiming to address credibility issues in traditional tools such as fabricated citations and non-independent sources. It enforces mechanical validation at the code level, requiring citations to include an exact original quote—quotes shorter than 24 characters are rejected—and uses hash matching to verify citation authenticity. Kestrel does not treat the number of sources as a measure of truth; instead, it uses algorithms such as MinHash to compute the number of independent origins, enabling it to identify reposts from the same source and assess source credibility. The plugin can be used directly in DSH or invoked as a standalone library for its validation and graph analysis features, making it suitable for developers who require high-confidence research.
Read Moredsh-socrates: DeepSeek Harness Deep Research Plugin
`dsh-socrates` is a deep research plugin for DeepSeek Harness, released under the MIT license. Its core philosophy is “Everything is a plugin.” By introducing Socratic clarification mechanisms and adaptive closed-loop research logic, it addresses the lack of verification in traditional pipelines. The plugin supports pre-research clarification, parallel multi-round research, claim-level cross-validation, programmatic citation auditing, and academic three-source retrieval. Through model-tier scheduling (Pro/Flash), it achieves high-precision deep research and automatically generates documents that include an evidence library and a citation report.
Read Moredsh-open-deep-research: DeepSeek Harness Deep Research Plugin
dsh-open-deep-research is a TypeScript framework based on DeepSeek Harness that addresses the challenges of building research agents for web search, web page reading, and generating cited reports. It offers planning, execution, and multimodal retrieval capabilities, and supports access via CLI, DSH Tool, and APIs. After installing Node.js, pnpm, and the DSH environment, users can run it through the `research` or `research-jina` profile; the former supports search only, while the latter integrates Jina to enable web page reading. The project is currently in Alpha preview and has limitations such as long-document truncation, but it is well suited to quickly building complex research workflows.
Read Moredsh-deep-research: Adaptive Deep Research Plugin Based on Official Workflow
Maintained by omdsh-dev, dsh-deep-research is a DeepSeek Harness workflow-type plugin with the npm package name @dsh-external/dsh-deep-research. It is built on the official workflow engine and integrates built-in web_search and web_fetch functionalities. Utilizing control theory and information theory, it implements planning, parallel research, marginal gain stopping, and optional adversarial review, registering a deep_research tool for conversational triggering. This article introduces its core mechanisms, installation commands, tool parameters, and typical usage, while noting that profiles must include a workflows provider.
Read MoreUse dsh-deep-research to complement the adaptive deep research closed-loop for DeepSeek Harness
`dsh-deep-research` is a DeepSeek Harness workflow plugin maintained by `omdsh-dev`, which turns deep research into an adaptive closed loop: planning first defines the answer space, research stops based on information gain, synthesis retains uncertainty, and review is optional. This article checks the installation commands, tool parameters and precautions against the community directory page, GitHub README and source code.
Read More