FuJackGao/dsh-collab 预览 preview

FuJackGao/dsh-collab

Bidirectional coding collaboration between Codex and DeepSeek Harness: Codex leads, DeepSeek writes code, Codex reviews. Includes a local marketplace plugin, an MCP server (stdio + Streamable HTTP), and a host-composition gateway for the harness.

Project Overview项目介绍

DSH Collab is a bidirectional coding collaboration package built for DeepSeek Harness. It lets ChatGPT Codex act as a project manager that dispatches coding tasks to local DeepSeek Harness coding sub-agents, receives streamed progress, and then runs an independent reviewer that actually executes builds and tests before issuing a verdict. The bundle ships three layers: a Codex-side local marketplace plugin under codex-plugin/ carrying the dsh-collab skill, a zero-dependency Node MCP server at mcp/dsh-mcp.mjs exposing five tools (dsh_task, dsh_task_status, dsh_task_cancel, dsh_review, dsh_read_file) over both stdio and HTTP transports, and the harness/dsh-bridge.mjs gateway that must be mounted into the host cordis profile and exposes /api/dsh-bridge/* plus WebSocket push, native tools, and per-workspace coding and reviewer sessions.

A typical session begins with a Codex prompt that invokes dsh_task against a --cwd directory that is auto-created if missing, optionally tagged with --lane and a --model fast|pro choice (resolved dynamically to deepseek-v4-flash or deepseek-v4-pro). The coding sub-agent writes to the shared workspace, streams progress back to Codex, and a separate reviewer sub-agent then reads real files on disk, performs static review, and actually runs the build and tests before dsh_read_file is used for acceptance; users can also enable --commit to auto-stage and commit changes, or query lifecycle state through --list, --status <id>, and --cancel <id> commands. The intended audience is individual developers and small teams who want Codex to handle orchestration and acceptance while DeepSeek Harness performs the heavier coding and reviewer work.

Installation requires copying harness/dsh-bridge.mjs into a DeepSeek Harness profile directory such as $DSH_HOME/profiles/web/ and adding a cordis.patch.yml entry that inserts dsh-bridge so the gateway restarts on http://127.0.0.1:3080. On the Codex side, users either register the MCP server in ~/.codex/config.toml under [mcp_servers.dsh] or install codex-plugin/ via the personal marketplace manifest and codex plugin add dsh@personal; ChatGPT web additionally needs the Secure MCP Tunnel client plus an OpenAI organization-scoped Runtime API key, and write-capable tools are subject to OpenAI product restrictions on Plus accounts. Documented caveats include a Windows Codex Desktop 26.803 bug that breaks local-marketplace skill injection, a fallback to one-shot execution when AbortSignal is unavailable in dynamic plugin hosts, and MIT licensing.

DSH Collab 是一个面向 DeepSeek Harness 的双向编码协作套件,让 Codex(ChatGPT)以项目经理身份把编码任务派给本机的 DeepSeek Harness 编码子代理,再由独立评审员真实跑通构建与测试后出具报告。架构分为三段:Codex 端的 MCP 工具与本地市场插件、位于 mcp/dsh-mcp.mjs 的零依赖 Node MCP 服务器(stdio 与 Streamable HTTP 双传输,暴露 dsh_task、dsh_review、dsh_read_file 等五个工具),以及 harness/dsh-bridge.mjs 提供的 HTTP/WS 网关与编码/评审子代理会话管理。

典型流程是用户在 Codex 会话中用自然语言发起 dsh_task,指定 --cwd 与可选的 --lane/--model fast|pro,编码子代理在共享工作目录里并行执行并实时回传,完成后由评审员静态审查并实际运行构建和测试,再通过 dsh_read_file 验收;任务支持 --commit 自动 Git 提交、持久化会话与跨任务记忆注入,以及 --list、--status <id>、--cancel <id> 等管理命令。它适合需要让 Codex 承担调度与验收、由本地 DeepSeek Harness 承担高吞吐编码与评审的个人开发者与小团队。

依赖方面:必须把 dsh-bridge.mjs 注入宿主组合并通过 cordis.patch.yml 注册,重启后默认监听 http://127.0.0.1:3080/api/dsh-bridge/status;Codex 端二选一配置 MCP 或本地市场插件,ChatGPT 普通聊天区需配合 Secure MCP Tunnel,且写入型工具对 Plus 账户存在产品级限制。已知限制包括 Codex Desktop 26.803 在 Windows 上插件技能注入 bug、动态插件环境无 AbortSignal 时冷恢复降级为一次性执行;许可证为 MIT。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 2 warnings2 项注意
  • Only 2 stars - very few users, little community feedback星标只有 2,几乎没人在用,遇到问题缺少社区反馈
  • No DSH plugin manifest detected - it may only carry the dsh-plugin topic, so the install method must be confirmed on the spot未检测到 DSH 插件清单:可能只是打了 dsh-plugin 话题,安装方式要现场确认
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:FuJackGao/dsh-collab

把 FuJackGao/dsh-collab 加入你的 DSH 配置(web profile)即可启用。

READMEREADME

DSH Collab — Codex × DeepSeek Harness 双向编码协作

Codex 主导,DeepSeek 执行,Codex 验收。 一个插件体系,让 ChatGPT(Codex)像项目经理一样把编码任务派给本机 DeepSeek Harness 的编码子代理,实时回传结果,读文件验收,并由独立评审员实际跑通构建/测试后出具评审报告。

Codex 工作区 (ChatGPT)
   │  派活 / 评审 / 验收
   ▼
DeepSeek Harness 网关 ── 编码子代理(多工作区、多 lane、fast/pro 模型)
   │                     独立评审子代理(必须跑通构建/测试)
   ▼
共享工作目录(任意目录,--cwd 指定)
flowchart LR
    subgraph Codex["Codex (ChatGPT) — 项目经理 + 验收员"]
        A1[dsh_task 派活]
        A2[dsh_review 评审]
        A3[dsh_read_file 验收]
    end
    subgraph DSH["DeepSeek Harness"]
        B1[HTTP 网关 /api/dsh-bridge]
        B2[编码子代理<br/>多工作区 · 多 lane · fast/pro]
        B3[独立评审子代理<br/>必须跑通构建/测试]
    end
    subgraph FS["共享工作目录"]
        C1[任意项目目录]
    end
    A1 --> B1
    A2 --> B1
    A3 --> C1
    B1 --> B2 --> C1
    B1 --> B3 --> C1
    B2 -.实时回传汇报.-> B1
    B1 -.任务结果.-> A1

组件

目录 内容 说明
codex-plugin/ Codex 本地市场插件包 标准 marketplace 结构(.agents/plugins/marketplace.json + 插件目录),含协作技能 dsh-collab
mcp/dsh-mcp.mjs MCP 服务器(零依赖 Node) stdio 传输(Codex [mcp_servers] 直连)+ Streamable HTTP 传输(--http,供 Secure MCP Tunnel / cloudflared 桥接 ChatGPT);5 个工具:dsh_task / dsh_task_status / dsh_task_cancel / dsh_review / dsh_read_file
harness/dsh-bridge.mjs DeepSeek Harness 宿主组合网关插件 提供 HTTP 网关(/api/dsh-bridge/*)、WS 推送、原生工具,管理编码/评审子代理会话。必须装进 harness 宿主组合(见下文)

功能

  • 派活:dsh_task / task.mjs,任意工作目录(--cwd,不存在自动创建并注册为工作区)
  • 并行分工:同目录多 lane(--lane)+ 跨目录天然并行
  • 模型切换:fast(deepseek-v4-flash)或 pro(deepseek-v4-pro),动态解析不写死
  • 双向评审:独立评审子代理与编码员分离,读磁盘真实文件、静态审查、实际运行构建/测试验证可跑通,输出结构化报告
  • Git 自动提交:--commit,任务完成后在 cwd 自动 git add -A 并提交
  • 跨任务记忆:持久子代理会话 + 历史摘要注入(冷恢复失败自动降级为一次性执行,功能不中断)
  • 任务管理:--list / --status <id> / --cancel <id>
  • 结果回传:任务汇报直接回 Codex 对话(--wait 默认)

安装

1. DeepSeek Harness 网关(必需,后端)

把 harness/dsh-bridge.mjs 复制到你的 dsh profile 目录(例如 $DSH_HOME/profiles/web/),并在该目录的 cordis.patch.yml 中追加:

- insert:
    - id: dsh-bridge
      name: './dsh-bridge.mjs'

重启 harness。默认共享工作目录为 D:\Harness,可用环境变量 DSH_BRIDGE_WORKSPACE 覆盖。验证:

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-rules 下一个 Next dsh-messenger-gateway →