DeepSeek Harness Plugin

taxueseek/argo#dsh-plugin

Stars ★ 178 Category Tools & Capabilities Added 2026-08-14

Unified search with evidence verification for agents: 120+ engines with multilingual routing, vertical domains and structured modality cards, 10 MCP tools plus wide_research orchestration, and a fetch-verify loop that backfills evidence scores.

Install

# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)

dsh plugin --profile web add github:taxueseek/argo#path:/packages/dsh-plugin

Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).

README

This plugin publishes its README in Chinese only.

DeepSeek Harness 插件:一行同时挂上 Argo 搜索 MCP 和 wide_research 并行研究编排。

安装

dsh plugin --profile web add "github:taxueseek/argo#main&path:packages/dsh-plugin"

重启 dsh web 后生效。本 bundle 提供 三形态接入——同一套 argo 搜索能力(同引擎、同压缩、同守卫),三种调用通道互为冗余,按宿主能力自动降级:

  • 原生一等工具(默认开):argo_search / argo_fetch(ctx.tools.register),走 CLI 单发(--call)复用 MCP 同一套引擎与守卫,不依赖 MCP 连接、零常驻 token 开销;nativeTools: [] 关闭
  • 原生 web seam(默认开):注册 argo 搜索 provider,宿主内置 web_search 路由到 argo 引擎链;searchProviderEnabled: false 时注册但不可选
  • MCP 形态(按需开):19 个 mcp__argo__* 完整工具面(argo_search / argo_local_search / argo_local_read / argo_recompute / argo_research / argo_evidence / argo_clarify / argo_fetch / argo_crawl / argo_screenshot / argo_pdf / argo_social_search / argo_article / argo_job / argo_extract / argo_preflight / argo_answer / argo_watch / argo_cite,与 CLI 能力面对齐)。工具定义每轮注入上下文且压缩清不掉(实测 19 工具 ≈3.7K token/轮),而搜索/抓取高频路径已由原生工具与 web seam 覆盖,故默认不挂载;需要长尾取证能力(screenshot / pdf / social / article / job / evidence / clarify / crawl / local_*)时在 profile patch 中取消 mcp-argo 行注释

另有 1 个 wide_research:规划互补轨道(depends_on 依赖分阶段,默认并行)→ 有界并发子代理取证 → 来源账本(仅 http(s) URL 入账)→ 综合报告;输出自带 quality_gate_results(conclusion_cap:low/medium/high),passed=false 时降级表述、先核验账本来源再下结论。worker 取证工具白名单默认双写 MCP 与原生工具名(allow 语义)——MCP 挂载与否两态自洽:MCP 在时用全量面,缺席时自动落到 argo_search / argo_fetch 与宿主 web_search / web_fetch。

日常一问用 argo_search;单代理取证用 argo_research;多视角对比、要引用账本用 wide_research。worker 默认只能调 argo 取证工具,不会再调 argo_research(硬保护,用户层放行也无效)。

搜索引擎侧,Firecrawl 等匿名可用引擎 keyless 免配置即用(免费层额度),配置了密钥的引擎自动升级为认证请求;缺密钥不阻断路由。

依赖

  • Node.js ≥ 18(npx 拉取 argo 包)
  • Python 3.9+,pip install pyyaml
  • DSH 子代理 provider 需支持 outputSchema、toolFilter、depthLimit(标准 spawn 即可)

自定义(可选)

两行配置 id 分别为 mcp-argo 和 wide-research。用户层 cordis.patch.yml 同 id + name 才是替换:

- id: mcp-argo
  name: '@deepseek-ai/dsh-mcp-client'
  config:
    serverName: argo
    transport: stdio
    command: /usr/bin/python3
    args:
      - /path/to/argo/scripts/mcp_server.py

- id: wide-research
  name: '@taxueseek/argo-dsh'
  config:
    defaultWorkers: 6
    maxWorkers: 9
    maxTracks: 9
    # 原生一等工具(可选调整):默认只注册 argo_search / argo_fetch(高频路径,
    # 零常驻 token 开销)。全部长尾工具(article / job / social_search / pdf /
    # crawl / evidence / clarify / screenshot / local_* / recompute / local_read /
    # local_search,共 13 个,schema 自动派生自 MCP 真源)都可按需追加,例如:
    # nativeTools: ['argo_search', 'argo_fetch', 'argo_pdf', 'argo_article']
    # nativeTools: []                               # 空数组关闭
    # nativeTimeoutMs: 60000                        # 单次进程超时(ms)
    # nativeMaxChars: 24000                         # 返回文本上限(字符)
    # searchProviderEnabled: false                  # 关闭原生 web seam
    childToolAllow:
      - mcp__argo__argo_search
      - mcp__argo__argo_evidence
      - mcp__argo__argo_fetch
      - web_search
      - web_fetch

注意:argo_research 不提供原生单发形态——它是分钟级编排,默认 60s 进程超时会中断研究;深度研究请用 wide_research(本 bundle 已注册)或挂载 MCP 形态。

卸载

在 DSH 的 Settings → Plugins 管理面板中移除本 bundle,或从 profile package.json 的 dsh.profile.bundles 中删除 @taxueseek/argo-dsh 后重启。

Content from the project README on GitHub ↗

Links

More in this category

View the whole category →

Community comments

Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.