At a glance
What it does
A DeepSeek Harness research bundle for discovering, extracting, verifying, and auditing multi-platform sources.
Deepseek Harness Profile
Not declared in supplied evidence
Evidence-verified
Checked Sep 14, 2026, 2:03 PM UTC
Code-evidenced contributions
What it adds to DSH
Adds source discovery, deep research, extraction, cross-source verification, and credibility auditing tools.
Mechanism evidence ↗Guides agents through scout, extract, verify, and audit steps for multi-platform research.
Mechanism evidence ↗Before you choose it
dsh-harvest adds five research tools: candidate-source scouting, asynchronous deep research, article extraction, cross-source claim verification, and five-dimensional source credibility auditing. Its intended flow is scout → extract → verify → audit, with structured JSON and readable output for agents.
Best for
DeepSeek Harness users who need repeatable multi-source research rather than a single web-search result.
Common tasks
- Find candidate sources across web, GitHub, social, community, and video channels.
- Extract original text from selected URLs and label paywalled or unreachable sources.
- Cross-check important claims and audit sources before citing them in a research report.
Permissions and data
Uses network-backed research channels and invokes available local command-line tools; optional Tavily credentials may be supplied by DSH configuration or environment variables.
Permissions- Network access to configured search, source, and platform endpoints.
- Permission to run available local CLI tools such as gh, mcporter, opencli, yt-dlp, and bili.
- Research queries and requested URLs may be sent to enabled external channels.
- Tool outputs include source URLs and research results for agent consumption.
- GitHub, web/Exa, social platforms, LinkedIn, YouTube, Bilibili, V2EX, optional Tavily endpoints.
- Tavily API key is optional for configured Tavily search or research endpoints.
- LinkedIn access is documented as requiring a logged-in session.
Limitations
- Available channels depend on separately installed CLIs, browser bridging, login state, PATH configuration, and platform support.
- Failed channels are skipped; this reduces coverage and does not prove a topic has no relevant sources.
- Windows support for some npm-installed command shims is described as dependent on PowerShell handling; Xiaoyuzhou transcription is documented as unreachable on Windows.
- The README describes runtime behavior, but installation and runtime execution were not verified in the supplied evidence.
What DSHub checked
- The package manifest, bundle patch, and immutable Git source commit were captured and bundle structure passed verification.
- The manifest declares version 0.2.0 and MIT licensing.
- A harvest skill document and bundle patch are present.
What DSHub did not check
- Successful installation in a DeepSeek Harness profile was not demonstrated.
- Runtime behavior, channel availability, credential handling, and operating-system compatibility were not executed or independently tested.
- No npm registry package for the declared version was found in the supplied record.
Pinned install
Install dsh-harvest
This plugin bundle does not have a DSH Plugin install action. Use its source documentation for the delivery method.
Maintainer source
Project README
🌾 dsh-harvest
<p align="center"> <img src="https://img.shields.io/badge/license-MIT-green" alt="license"/> <img src="https://img.shields.io/badge/DeepSeek%20Harness-plugin-4b32c3" alt="DeepSeek Harness"/> <img src="https://img.shields.io/badge/version-0.1.0-blue" alt="version"/> <img src="https://img.shields.io/badge/zero--deps-self--contained-2ea44f" alt="zero deps"/> <img src="https://img.shields.io/badge/runtime-Node%20ESM-339933" alt="Node ESM"/> <img src="https://img.shields.io/badge/target%20platforms-Win%20%7C%20macOS%20%7C%20Linux-lightgrey" alt="target platforms"/> </p>DSH 原生多平台调研流水线插件 —— 把「发现 → 抓取 → 验证 → 审计」做成四个原生工具。 目标平台:Windows / macOS / Linux(安装与数据通道均已按平台分块说明)。
dsh-harvest 让 DeepSeek Harness 里的智能体像开联合收割机一样做调研:一次撒网多个平台、穿透反爬抓原文、跨源交叉验证、最后给来源打分。任一数据通道挂了也不会卡住整条流水线。
✨ 五个原生工具
| 工具 | 职责 | 失败策略 |
|---|---|---|
harvest_scout |
多平台并行发现候选来源(Web/GitHub/社媒等) | 单通道失败记 [SKIP],不阻塞 |
harvest_deep_research |
异步端到端深度调研(生成长篇综合研报与信源) | 超时/异常安全回退,不卡死主会话 |
harvest_extract |
逐条抓取正文 | 直抓 → Jina 升级 → 标记 paywall/unreachable |
harvest_verify |
跨源交叉验证断言 | 一致 / 弱 / 未证实 三态 |
harvest_audit |
来源可信度五维审计 | 🟢 可信 / 🟡 谨慎 / 🔴 弃用 |
数据流:scout → extract → verify → audit,每个工具输出结构化 JSON + 折叠文本,供 agent 直接读取。
🔌 数据通道
9 条数据通道,任一通道失败(CLI 未装 / 浏览器桥接未连 / 解析失败)记 [SKIP],不阻塞整条流水线。
| 通道 | 底层 | 类型 | macOS / Linux 可用性 |
|---|---|---|---|
| GitHub | gh search repos |
代码 | ✅ 三平台原生可用,未装即 [SKIP] |
| Web / Exa | mcporter call exa.web_search_exa |
网页搜索 | ✅ mac/linux 可直接调用(修复后形态) |
| Twitter / Reddit / 小红书 | opencli(浏览器桥接) |
社交 | ✅ mac/linux 可直接调用(修复后形态) |
mcporter linkedin-scraper(需登录态) |
招聘 | ✅ mac/linux 可直接调用(修复后形态) | |
| YouTube | yt-dlp ytsearchN: |
视频 | ✅ 三平台原生可用,未装即 [SKIP] |
| B站 | bili search(bili-cli,第三方) |
视频 | ✅ 三平台原生可用,未装即 [SKIP] |
| V2EX | curl v2ex.com/api/topics/hot.json |
社区 | ✅ 三平台可用(HTTP 直连 + curl 兜底) |
修复说明:
mcporter/opencli在 Windows 上以 npm 全局安装生成的.ps1/.cmd垫片存在(NodeexecFile无法直接执行,需经 PowerShell 调用);在 macOS/Linux 上它们是可执行脚本/二进制,可直接 argv 调用。本次跨平台修复(并行进行中,见 DESIGN.md 数据通道矩阵)即按该目标形态实现,文档按修复后的形态描述。macOS/Linux 上未装对应 CLI 时通道记[SKIP],与 Windows 行为一致。
另:
harvest_extract自动识别 RSS feed(feedparser)与小宇宙播客 URL(Whisper 转写),把 agent-reach 剩余 3 个平台(小宇宙/RSS/网页阅读)也纳入 —— 合计覆盖全部 13 平台。 平台依赖:RSS 解析走 Python(python3/python,需 feedparser 包);小宇宙转写走 bash(~/.agent-reach/tools/xiaoyuzhou/transcribe.sh,需 groq key)。Windows 默认无 bash,小宇宙通道在 Windows 上恒unreachable(优雅降级,不报错)。
数据通道 CLI 获取方式(按平台)
未安装的通道不会阻塞流水线(自动记 [SKIP]),按需安装即可:
| CLI | 用途 | Windows | macOS | Linux |
|---|---|---|---|---|
gh |
GitHub 搜索 | winget / 官方安装包 | brew install gh |
官方 apt 仓库 / 二进制 |
mcporter |
Web/Exa · LinkedIn | npm i -g mcporter(经 PowerShell 垫片) |
npm i -g mcporter(可执行脚本,直调) |
同 macOS |
opencli |
Twitter · Reddit · 小红书 | npm i -g opencli |
npm i -g opencli |
同 macOS |
yt-dlp |
YouTube | 官方 yt-dlp.exe / winget |
brew install yt-dlp |
pip install yt-dlp / 发行版包 |
bili-cli |
B站 | GitHub 发布页 bili.exe |
brew / 源码构建 | 源码构建 |
bili-cli 重要提示:
bili是第三方非官方 bilibili 客户端(Go 单二进制),非哔哩哔哩官方发布;只从可信来源(如项目自己的 GitHub 发布页)获取并自行核验,注意其登录方式与合规边界。 Windows npm 全局:mcporter/opencli经 npm 全局安装,垫片生成在 npm 全局 bin 目录(npm prefix -g可查);若该目录非常规位置,设环境变量DSH_HARVEST_BIN指向它以覆盖默认查找。 macOS 提示:若dsh web从 Finder / LaunchAgent 启动,进程 PATH 通常只有系统目录,/opt/homebrew/bin下的gh/yt-dlp取不到(即便已装也会[SKIP]);从终端启动,或在启动脚本里显式把用户 bin 目录加进 PATH(如export PATH="/opt/homebrew/bin:$PATH")。
📦 安装(DSH 原生)
Windows(PowerShell)
# 1. 克隆到 DSH 插件目录
git clone https://github.com/<you>/dsh-harvest.git $env:USERPROFILE\.dsh\plugins\dsh-harvest
# 2. 在 web profile 的 package.json 里加 link 依赖 + bundles 条目
# "dependencies": { "dsh-harvest": "link:C:/Users/<you>/.dsh/plugins/dsh-harvest" },
# "dsh.profile.bundles": [ ..., "dsh-harvest" ]
# (link 路径用正斜杠,把 <you> 换成你的用户名)
# 3. 链接
cd $env:USERPROFILE\.dsh\profiles\web
pnpm install
# 4. 重启
dsh web
macOS / Linux(bash / zsh)
# 1. 克隆到 DSH 插件目录
git clone https://github.com/<you>/dsh-harvest.git ~/.dsh/plugins/dsh-harvest
# 2. 在 web profile 的 package.json 里加 link 依赖 + bundles 条目
# macOS: "dependencies": { "dsh-harvest": "link:/Users/<you>/.dsh/plugins/dsh-harvest" }
# Linux: "dependencies": { "dsh-harvest": "link:/home/<you>/.dsh/plugins/dsh-harvest" }
# "dsh.profile.bundles": [ ..., "dsh-harvest" ]
# 3. 链接
cd ~/.dsh/profiles/web
pnpm install
# 4. 重启
dsh web
两平台差异只在根目录:Windows 用
$env:USERPROFILE(PowerShell),macOS/Linux 用~。link:依赖需写绝对路径(JSON 不做变量展开),<you>换成你的用户名即可复制执行。
🚀 用法
harvest_scout(query="最新 AI 项目", limit=8)
harvest_deep_research(input="深入剖析大模型推理与架构演进")
harvest_extract(urls=["https://…", "https://…"])
harvest_verify(claims=["DeepSeek 出了视觉模型"], sources=[…])
harvest_audit(sources=[{title, url, type}])
⚙️ 配置与安全(支持标准凭据管理)
插件支持零代码侵入的多级配置机制,任选一种即可生效:
方式 1:DSH 原生标准配置(推荐)
在 ~/.dsh/settings.yaml 中配置:
harvest:
tavilyApiKeyEnv: TAVILY_API_KEY
tavilyEndpoint: https://search.cliproxyapi.xyz/search # 可选,默认官方标准端点
tavilyResearchEndpoint: https://search.cliproxyapi.xyz/research # 可选,默认官方标准端点
然后在 ~/.dsh/.credentials.yaml 中保存密钥:
refs:
TAVILY_API_KEY: 你的Tavily或中转Key
方式 2:系统环境变量
TAVILY_API_KEY: 配置密钥。TAVILY_ENDPOINT: 默认为https://api.tavily.com/search。TAVILY_RESEARCH_ENDPOINT: 默认为https://api.tavily.com/research。
🏗️ 架构
lib/
index.js # 插件主体:注册 4 个工具
backends.js # 9 通道封装 + 优雅跳过
extract.js # 抓取路由(直抓 → r.jina.ai;RSS / 小宇宙)
http.js # HTTP 层(fetch + 平台化兜底)
零依赖:只用 Node 内建(node:child_process / 全局 fetch),不 import 任何第三方包。
📄 归属与血统
方法论(Scout → Extract → Verify → Audit)源于作者自研的 omni-scope 工作流,按 DSH 原生插件形态从零重写。
运行时仅以子进程调用 gh / mcporter / opencli / yt-dlp / bili 作为数据通道,不包含其源码。详见 ATTRIBUTION.md。
📝 License
MIT © 2026 dsh-harvest contributors
Operate deliberately
Install and manage
Prerequisites and target Profile
Target: Deepseek Harness Profile
Delivery: Dsh Bundle Git — toustifer/dsh-harvest#ab40151dbb123655689777f1630c0ecdac610d54。
Verify, update, and remove
Show lifecycle commands
dsh plugin --profile deepseek-harness listCompatibility and access
DSH bundle; target platforms are documented: Not declared in supplied evidence。
Review compatibility evidence ↗
Risk facts
Optional Tavily API credentials can be configured through DSH credentials or environment variables.
Evidence ↗May access web, social, code, video, and third-party search channels through HTTP and local CLI tools.
Evidence ↗Evidence and editorial reviewManifest, Bundle patch, distribution and freshness
Immutable evidence
Review status and source activity
Use the pinned Git bundle route. Treat its verification and audit labels as workflow outputs, not independent certification.
AI reviewed Sep 14, 2026, 2:04 PM UTC。GitHub facts last checked Sep 14, 2026, 2:04 PM UTC。
No material source change has been recorded since this evidence baseline.