omdsh-dev/dsh-deep-research
122Last commit Aug 12, 2026
dsh-deep-research DSH plugin
This plugin implements a deep research workflow as a DSH plugin, using the official workflow engine. It applies cybernetic and information-theoretic principles to create a self-adaptive research loop that automatically decomposes topics, runs parallel sub-research, and iterates until marginal gain diminishes.
How to install the dsh-deep-research DSH plugin
dsh plugin --profile <profile> add git+https://github.com/dsh-external/dsh-deep-research.gitThis source command needs manual review. Copying does not run it.
dsh-deep-research DSH plugin data source
dsh-deep-research DSH plugin snapshot date: Aug 16, 2026
discovered
What the dsh-deep-research DSH plugin can do
- Automatic topic decomposition into sub-questions based on information dimensions
- Parallel sub-research with marginal gain verification and adaptive re-planning
- Optional adversarial review for citation checking, coverage audit, and contradiction labeling
- Model tiering: separate planner, researcher, synthesizer, and reviewer model configurations
- Depth control (1-3) and review toggle for precision/quality adjustment
Where the dsh-deep-research DSH plugin fits
- Deep research on a specific topic with multi-source synthesis
- Comparative analysis of several implementations or approaches
- Decision-support research where purpose is defined to narrow the answer space
- Literature collection and evidence-based report generation
Who the dsh-deep-research DSH plugin is for
- DSH users who need automated in-depth research capabilities
- Researchers and analysts seeking a structured, adaptive research workflow within DSH
dsh-deep-research DSH plugin limitations
- Requires a DSH profile that provides the official workflow engine (ctx.workflows) and built-in web tools (web_search/web_fetch)
- Planning/synthesis tasks require strong models for best results; researcher model can be cheaper but must be capable
- Plugin is early-stage; may have edge cases in complex research scenarios
dsh-deep-research DSH plugin: from the repository README
Quoted from the omdsh-dev/dsh-deep-research README, the upstream source of the dsh-deep-research DSH plugin. Copyright remains with the original authors.
把 deep-research 流程做成 **DSH 扩展插件**（plugin，与 skill 体系分开）， 基于 **DSH 官方 workflow 引擎**（`ctx.workflows` / `@deepseek-ai/dsh-workflow-workerthread`） 实现，按 **控制论 + 信息论** 设计——不是固定提示词流水线，而是**活的、自适应的研究闭环**。 ## 理论 → 机制 | 理论 | 插件里的落地 | | --------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------- | | 控制论：参考信号校准（闭环控制错误目标 = 白费） | 规划代理先定义**答案空间**（`scope`：研究支撑什么判断/决策）与每个子问题的**验收标准**（`acceptance`），再开始研究 | | Ashby 必要多样性定律（控制器多样性 < 系统多样性 ⇒ 必有盲区） | 规划代理枚举主题的**信息维度**，每个子问题映射一个维度，并输出**覆盖度自检** `coverage_gaps` | | 信息论：信息 = 不确定性的减少 | 研究子代理维护三态证据 `confirmed / uncertain / gaps`——条件熵的工程表达；报告保留置信度与矛盾，不掩盖不确定性 | | 信息论：边际信息增益（EIG）递减 ⇒ 无限搜索是错的 | 每个研究子代理：预测（针对哪个高熵点、预期新增什么）→
Read the full READMERepository license: MIT
dsh-deep-research DSH plugin questions
How do I install this plugin?
Run `dsh plugin --profile <profile> add git+https://github.com/dsh-external/dsh-deep-research.git` where `<profile>` is your DSH profile (e.g., tui, headless, web). Then restart DSH with `dsh --profile <profile>`. The plugin will be injected as a tool called `deep_research`.
What models should I use for the different roles?
The plugin allows model tiering: `plannerModel` and `synthesizerModel` should be strong models (e.g., GPT-4, Claude 3.5 Sonnet) for planning and synthesis. `researcherModel` can be a cheaper model (e.g., GPT-3.5, Claude 3 Haiku) to reduce cost. `reviewerModel` is optional. Configure these in your DSH profile settings.
Can I use this plugin with any DSH profile?
No, it requires a profile that provides the official workflow engine (`ctx.workflows`) and built-in web tools (`web_search`, `web_fetch`). The default tui and headless profiles include these. If your profile doesn't, the plugin will remain pending. You may need to register a workflows provider in DSH Hub or switch to a compatible profile.
How do I update or uninstall the plugin?
To update all plugins: `dsh plugin --profile <profile> update`. To uninstall specifically: `dsh plugin --profile <profile> remove @dsh-external/dsh-deep-research`. Alternatively, remove the dependency from your profile's package.json and run `dsh plugin --profile <profile> update`.
What does the 'depth' parameter do?
The `depth` parameter controls the precision/tolerance of the research. depth=1 is preliminary, depth=2 (default) is thorough, depth=3 is exhaustive. It determines the maximum number of research cycles (depth+1). Higher depth means more iterations but also more cost.