a1141171521/dsh-thinking-effort

Declare Codex-style reasoning effort levels (Off / Low / Medium / High) for OpenAI-compatible third-party models in DeepSeek Harness, so the built-in model selector shows its Effort row for those models — and pin the route-level default effort so switching away and back keeps your choice.

catalog descriptioncatalog 简介 / catalog description:dsh 工作区可添加多文件夹/第三方厂商模型可开启推理等级插件

Project Overview项目介绍

This is a native plugin built exclusively for DeepSeek Harness, designed to add reasoning effort level support for third-party OpenAI-compatible models added to DSH. By default, DSH only shows the effort selection row for models that explicitly declare reasoning levels, which most custom third-party models do not include. This plugin automatically adds the required reasoning effort declaration to all qualifying OpenAI-compatible models during activation, and pins a default effort level at the route level to preserve user choices across model switches. It also includes an optional additional module for adding whitelisted reference directories to DSH sessions.

The plugin is built with a modular structure, so new features can be added without changing the core installation flow. To install the plugin, you can run the command dsh plugin --profile <your-profile> add github:a1141171521/dsh-thinking-effort in your terminal, or install it from a local directory or packaged tarball if you prefer. After installation, you must restart your DSH process for the plugin to load correctly, as a browser refresh alone will not trigger a rescan of available plugins. Once loaded, you can select your third-party model in the DSH model selector and find the new Effort row to adjust your reasoning level.

The plugin has no external third-party dependencies, as it is written entirely in plain JavaScript and requires no build step to run after installation. It works with both openai-completions and openai-responses API providers, and all changes it makes are idempotent, so it will not overwrite existing user configurations for models that already have reasoning effort declared. Uninstalling the plugin does not automatically roll back the changes it has made to your DSH settings, so you will need to manually remove the added declarations if you no longer want them. Local user effort selections are stored in the browser’s localStorage, which persists even after the plugin is uninstalled.

这是一款专为DeepSeek Harness开发的原生插件,核心功能是为DSH中接入的第三方OpenAI兼容模型补全推理档位声明,让原本不支持档位选择的第三方模型也能在DSH内置模型选择器中显示推理档位(Effort)选择栏,同时支持固定路由级默认推理档位,切换模型后切回依然能保留用户选择的档位,插件采用模块化设计,还额外附加了实用的引用目录管理功能模块。

该插件面向需要在DSH中使用第三方OpenAI兼容推理模型的用户,安装插件后必须重启DSH进程才能生效,插件会自动扫描llm-pi-ai配置下所有符合条件的模型,补全缺失的档位声明并自动钉住默认档位。用户可直接在模型选择器的下拉菜单中找到Effort档位选项,自由切换Off、Low、Medium、High等不同档位,还能通过配置自定义全局默认档位。

本插件是纯JavaScript实现,无第三方依赖,无需构建步骤,安装后即可直接运行,支持同时处理openai-completions和openai-responses两种类型的API提供商。所有配置修改操作都是幂等的,不会重复覆盖用户已有的自定义配置,卸载插件不会自动回滚已写入的配置,若厂商不支持reasoning_effort参数,需要用户手动删除配置中的对应声明。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 2 warnings2 项注意
  • No license declared - all rights reserved by default; ask the author before commercial use or redistribution未声明开源许可证 —— 默认「保留所有权利」,商用或再分发前先问作者
  • Only 4 stars - very few users, little community feedback星标只有 4,几乎没人在用,遇到问题缺少社区反馈
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 demo add github:a1141171521/dsh-thinking-effort

把 a1141171521/dsh-thinking-effort 加入你的 DSH 配置(web profile)即可启用。

READMEREADME

dsh-thinking-effort

Declare Codex-style reasoning effort levels (Off / Low / Medium / High) for OpenAI-compatible third-party models in DeepSeek Harness, so the built-in model selector shows its Effort row for those models — and pin the route-level default effort so switching away and back keeps your choice.

💬 DeepSeek Harness 插件交流群

欢迎加入 DeepSeek Harness 插件交流 QQ 群——插件开发、模型接入、推理档位、 第三方厂商配置等话题都可以在这里交流,本插件的问题与建议也欢迎直接进群反馈:

🐧 QQ 群号:1105449571

打开 QQ → 搜索群号 1105449571 即可加入(进群请备注「dsh 插件」)

群内可交流:插件开发经验、模型接入与档位配置、DSH 使用技巧、新功能需求投票等。

多功能插件(v0.5.0+)

本插件按模块组合多个扩展功能,安装一次全部生效,后续功能持续加入:

模块 功能
modules/thinking-effort.js 推理档位:为第三方模型声明档位并钉住默认档位
modules/refdir.js 引用目录(附加文件夹):为会话添加可读写的白名单文件夹,对话可用 refdir_list/read/write/edit/grep 工具操作,类似 Claude Desktop 的附加文件夹/副工作区

新增功能 = 在 modules/ 下加一个模块文件并在 index.js 挂上,安装方式不变。

引用目录(refdir 模块)

入口

  • 点击输入框左侧 + 按钮 → 命令菜单选择 refdir 引用目录 → 「+ 添加引用目录」→ 系统文件夹选择器选中即添加(再点已添加的目录行 = 移除)
  • 输入框工具行新增 📁 芯片(显示引用目录数量)、会话标题栏新增 「📁 引用目录」按钮 → 打开管理面板:显示主工作区路径、目录列表(可一键移除)、添加按钮

目录选择器(v0.5.6+):选择文件夹优先走宿主命令 /refdir-pick (Host 的 ctx.directoryPicker),不再直接依赖客户端服务名——客户端该服务 随版本改过名(旧版 workspaces.pickDirectory,当前版 uiWorkspace.pickDirectory),旧写法在当前版本会报「目录选择器不可用」。 客户端服务仍作为回退保留。

对话获得的能力(添加引用目录后)

  • 5 个专属动态工具:refdir_list(列文件)、refdir_read(读文件)、refdir_write(写文件)、refdir_edit(编辑)、refdir_grep(搜索)
  • 工具只能访问你显式添加的引用目录内的路径(白名单校验,fs.contains 围栏),引用目录之外的路径一律拒绝
  • 每个会话一份独立列表;持久化到该会话日志目录旁的 refdirs.json(重启 DSH 后重新加载插件即可恢复)
  • 浏览器 UI 通过标准命令通道(Host 注册 refdir-list/add/remove 命令,客户端 ctx.remote.commands.execute 调用)读写列表,不依赖自定义 RPC;插件本身保持纯 JS 零第三方依赖

这是什么

DSH 的 llm 服务只为适配器声明了档位的模型接受 reasoningEffort。第三方 (自定义)模型通常没有声明,导致模型选择器隐藏 Effort 行,显式选择档位时 还会报 UNSUPPORTED_REASONING_EFFORT。本插件在激活时(以及 llm-pi-ai 配置每次变化时)自动为 llm-pi-ai 下所有 OpenAI 兼容模型补上 档位声明:

reasoningEfforts:
  off: null      # 支持关闭思考,不发送参数
  low: low       # 发送 reasoning_effort: "low"
  medium: medium
  high: high
  xhigh: xhigh
  max: max

档位集合与 DeepSeek 协议网关一致(该协议只接受 low / medium / high / xhigh / max,拒绝 minimal)。已声明的模型会补齐缺失档位,已有值 (包括手动钉为 null 的)保持不动;reasoningEfforts: false 的模型 不触碰。

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-automation 下一个 Next dsh-startup-animation →