Jungod1121/dsh-anchored-standard 预览 preview

Jungod1121/dsh-anchored-standard

两阶段 DeepSeek Harness 预设:最小对齐引导(bash+read),然后在首次工具调用或回复后启用完整标准工具。

Project Overview项目介绍

This is a native two-phase agent preset built exclusively for DeepSeek Harness (DSH), designed to improve the reasoning trajectory of DeepSeek V4 Pro. It follows a tested anchoring approach first documented by the DSH community: the first request only includes a task-matched persona and 2-3 core tools, then unlocks all 25 standard DSH tools after the first durable tool call from the model. You can install it via the DSH CLI with one command: dsh plugin --profile web add github:Jungod1121/dsh-anchored-standard, then restart DSH to activate the preset.

The preset automatically classifies the user’s initial task into three clear categories: bug fixing and maintenance, new feature build from scratch, and open-ended ambiguous tasks. Each category gets a tailored initial persona and matching tool set, keeping the first turn context as clean as possible to properly anchor the model’s reasoning. The anchored state persists across session refreshes and restores, so you don’t have to restart the session if you reload DSH mid-task. It targets developers using DeepSeek V4 Pro with max reasoning effort for full-stack development work.

This preset has been verified and tested only against DSH version 0.1.0-rc.6, and it is explicitly built for DeepSeek V4 Pro with maximum reasoning effort enabled. V4 Flash does not benefit from this preset, as it generalizes well across different prompt styles without trajectory anchoring. It is released under the open-source MIT license, and is an independent community project not affiliated with or endorsed by DeepSeek. The reported 98/99 ability score comes from a specific test task, and is not a universal guarantee across all workloads.

这是一个专为DeepSeek Harness(DSH)开发的两阶段智能体预设插件,基于DeepSeek V4 Pro模型的推理特性设计,核心思路是通过锚定首回合请求的推理轨迹来提升模型整体能力。它采用分阶段启动方案,首回合仅提供匹配任务类型的极简提示词和少量工具,在模型完成首次稳定工具调用后,再暴露完整的标准工具目录,结合了极简预设的推理优势和标准预设的全能力覆盖。

它会根据用户首次发送的任务自动分类,对应修复调试、从零构建、模糊任务三种场景,分别提供不同的首回合工具和角色设定。首回合保持最干净的上下文环境,会话记录首次工具调用后自动切换为全工具模式,锚定状态会持久保存在会话事件中,刷新或恢复会话都能保留状态。它面向使用DeepSeek V4 Pro配合DSH进行全栈开发的开发者,帮助获得比原生预设更高的任务完成率。

该预设仅针对DeepSeek Harness 0.1.0-rc.6版本验证适配,专门面向开启最大推理强度的V4 Pro模型,V4 Flash不需要该预设。安装可通过DSH插件命令一键添加,也可手动复制预设目录到DSH本地,卸载也支持命令移除。它采用MIT许可开源,属于社区项目,不隶属于DeepSeek官方,能力评分仅针对特定测试任务,不保证通用场景效果。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 1 note1 项提示
  • 26 stars - an early-stage project星标 26,属于早期项目
DSH walks through these 9 checksDSH 会逐条核对这 9 项

Compatibility兼容性

  • DSH, Node, OS and profile requirementsDSH 版本 / Node 版本 / 操作系统 / profile 是否满足要求
  • External dependencies and runtimes (Electron / Python / Docker, ...)外部依赖与运行时(Electron / Python / Docker 等)是否齐备
  • Conflicts with installed plugins: command names, skill / tool names, ports, duplicate MCP registration与已装插件是否冲突:命令名、skill / tool 重名、端口占用、重复 MCP 注册

Security安全性

  • Repo matches the facts registered here; archived or abandoned?仓库是否与页面登记一致,是否归档或长期停更
  • Safety of preinstall / install / postinstall and install.sh / setup.ps1preinstall / install / postinstall 与 install.sh、setup.ps1 是否安全
  • curl|bash, download-then-execute, obfuscation, unrelated domains → stop immediatelycurl|bash、下载即执行、混淆代码、无关域名 → 立刻停止
  • Typosquatting or unmaintained packages among the new dependencies新增依赖里有没有 typosquatting 或无人维护的包
  • Requested permissions vs. what the feature actually needs申请了哪些权限、是否超出功能所需(filesystem / network / shell / clipboard)
  • Any sudo / admin requirement, plus uninstall and rollback是否要求 sudo / 管理员权限,以及卸载与回滚方式

Anything uncertain must be marked unknown with a note on how to confirm it. This site's signal screen is a static snapshot, not a security audit.拿不准的必须标「未知」并说明要我怎么确认。本站的信号筛查是静态快照,不能替代安全审计。

Or use CLI install (for developers)或使用命令行安装(适合开发者)

CLI Install命令行安装

dsh plugin --profile web add github:Jungod1121/dsh-anchored-standard

把 Jungod1121/dsh-anchored-standard 加入你的 DSH 配置(web profile)即可启用。

READMEREADME

dsh-anchored-standard

Anchor the first request. Unlock the full stack.

A two-phase agent preset for DeepSeek Harness: bootstrap DeepSeek V4 Pro with the Minimal-aligned prompt and two tools, then expose the complete Standard catalog after the first durable tool call or reply.

harness license stars dsh-plugin

中文说明 · Landing page · Design origin


The one-line idea

V4 Pro's ability ceiling is high — but which reasoning strategy it uses is decided by what the first API request shows it. The community evaluation at xiaobright/modeltest measured:

Preset First request sees Full task sees Project2 V4.1b (max)
Standard 25 tools + long prompt 25 tools 91 / 92
Minimal 1-line prompt + 2 tools 2 tools only 99 / 96 — but too narrow for real work
Anchored Standard 1-line prompt + 2 tools all 25 tools 98 / 99

The gain comes from anchoring the first-turn trajectory — not from keeping the tool surface small forever. This preset keeps Minimal's brain and gives it Standard's hands.

How it works

sequenceDiagram
    participant U as User
    participant A as Agent loop
    participant M as DeepSeek V4 Pro

    U->>A: first message (new session)
    A->>M: request #1<br/>task-matched anchor persona<br/>tools: bash + read (± edit/write)
    Note over M: reasoning chain anchors on the task-matched scaffold
    M->>A: first tool/call
    Note over A: promotion signal (durable session event)
    A->>M: request #2<br/>same persona · all 25 Standard tools
    Note over M: full-capability work continues on the anchored trajectory

Showing the opening section of the README — the full document lives in the repository以上为 README 开头摘要,完整文档在仓库内 · View the full README on GitHub →在 GitHub 查看完整 README →

← 上一个 Prev dsh-office-tools 下一个 Next dsh-codex-sync →