february2015/dsh-taskswarm

TaskSwarm (蜂群) — 基于DeepSeek Harness的多智能体任务编排插件:支持waves/lanes并行执行、git worktree隔离、任务包及跨模型评审、崩溃恢复 | 多智能体任务编排,适用于DeepSeek Harness:具备波浪/车道并行、git工作树隔离、任务分组、跨模型审查及崩溃恢复功能

Project Overview项目介绍

dsh-taskswarm is a native multi-agent task orchestration plugin built exclusively for DeepSeek Harness (DSH). It is ported from the upstream TaskPlane project, and enables users to organize batch tasks into dependency-ordered waves, run multiple worker agents in parallel isolated git worktrees, and automatically review and merge output after all tasks complete. It supports parallel execution of tasks within the same wave, which significantly speeds up the delivery of batches of dependent development work. Between August 17 and 18, 2026, users need to install it directly from the GitHub repository because the npm package was unpublished for cleanup of old versions, and after August 18, you can install it via the standard dsh plugin add dsh-taskswarm command.

This plugin is designed for developers who need to process batches of dependent development tasks, and adds a project-level orchestration layer on top of DSH's native subagent capabilities. A typical workflow starts with users writing task packets in advance, each consisting of a PROMPT.md for task description and a STATUS.md for progress tracking. A conversational supervisor agent plans the waves, each task runs in parallel in an independent git worktree, and after completion, the process automatically moves to review, semantic conflict resolution and result merging. The plugin supports crash recovery, pause and restart, handles unresolvable conflicts by pausing batches for intervention, and can also work with the dsh-dingo plugin to display real-time task status.

TaskSwarm is released under the MIT license, so it is free to use, modify, and redistribute for any purpose, including commercial projects. It relies on a working DSH environment and git tooling to run, and ships with a local zero-dependency real-time web dashboard that automatically starts when a batch begins and prints the access link in the chat. The dashboard automatically handles port negotiation, so users do not need to manually configure ports. It also saves token usage by optimizing notification design. First-time users are recommended to read the official runbook before use, to learn the standard operating procedures for cleanup, error recovery, and work salvage, and follow the correct installation steps based on the current date.

dsh-taskswarm 是 DeepSeek Harness (DSH) 的原生多智能体任务编排插件,基于上游 TaskPlane 架构移植而来,支持将批量任务按依赖关系组织为有序波次,通过 git worktree 实现多个工作任务并行隔离运行,最终自动完成结果审查和合并。2026年8月17日至18日因 npm 包需下架清理旧版本,用户需从 GitHub 仓库安装,8月19日后可直接通过 dsh plugin add 命令安装官方 npm 包。

该插件面向需要批量处理多依赖开发任务的开发者,是 DSH 原生子代理能力之上新增的项目级编排层。典型工作流为:用户提前编写标准化任务包(包含任务说明PROMPT.md和进度记录STATUS.md),由监督智能体规划波次,每个任务在独立 git worktree 中并行执行,完成后自动进入审查、冲突语义解决和结果合并,支持崩溃后恢复、暂停重启,也可配合 dsh-dingo 插件显示任务状态。

该插件采用 MIT 许可证,免费可商用修改和分发,运行依赖 DSH 环境和 git 工具,自带本地零依赖实时网页看板,会自动启动并打印访问链接,无需额外配置端口。用户首次使用前建议阅读官方运行手册,了解清理、错误恢复和工作挽救的标准操作流程,同时注意安装的临时时间限制,按说明选择正确的安装方式。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 1 warning1 项注意
  • Only 3 stars - very few users, little community feedback星标只有 3,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 web add dsh-taskswarm

把 february2015/dsh-taskswarm 加入你的 DSH 配置(web profile)即可启用。

READMEREADME

TaskSwarm(蜂群)

⚠️ Temporary install notice (Aug 17–18, 2026): the npm package dsh-taskswarm was fully unpublished to clean up old versions and is subject to a 24-hour cool-down (npm's anti-squatting protection). For Aug 17 and Aug 18, install from GitHub: dsh plugin add https://github.com/february2015/dsh-taskswarm.git Starting Aug 19, the npm package is back: dsh plugin add dsh-taskswarm. This notice will be removed then.

Multi-agent task orchestration for DeepSeek Harness (DSH).

TaskSwarm arranges a batch of tasks into dependency-ordered waves, runs multiple AI workers in parallel lanes isolated by git worktrees, then automatically reviews and merges their output.

蜂群 (fēngqún) is a swarm of bees: the queen directs, the workers each buzz on their own task in parallel — exactly what this project does: the supervisor plans the waves, then every worker advances in its own lane.

  • License: MIT
  • Upstream: TaskPlane(Pi ecosystem multi-agent orchestration)— this project is a native port
  • 中文文档: README.zh-CN.md

Features

  • Waves / Lanes parallel orchestration — tasks are topologically layered by their dependency DAG into waves; tasks within a wave run concurrently
  • Git worktree isolation — every task (lane) works in its own worktree; results are merged into the taskswarm/orch integration branch
  • Task packets — each task is a PROMPT.md (mission / steps / constraints) + STATUS.md (progress), giving workers durable memory across context resets
  • Checkpoint discipline — automatic git commits at step boundaries; a crashed worker never loses committed work
  • Cross-model review — an independent reviewer scores each task per its Review Level; PASS merges, REVISE sends it back for revision
  • File mailbox — workers and supervisor communicate asynchronously (notify / escalate / request) without shared context
  • Conversational supervisor — shares your session: reports wave completion, lane failures, batch completion; takes verbal commands (start / pause / abort / integrate / open dashboard); notifications and the prompt are bilingual (中文 / English) — say "use English" to switch, auto-detected from your session language, persisted to .taskswarm/config.json across restarts
  • Web Dashboard — local realtime dashboard, zero-dependency node:http + SSE, multiple instances with automatic port negotiation. Auto-starts when a batch starts and prints the link in chat — one dashboard per workspace, ever (an already-running instance is reused, never duplicated)
  • Crash-recoverable — durable disk state + checkpoints + retained lane branches; after a kill/restart you can salvage work, clean up residue, and re-run
  • Orch-based lane baselines — every lane worktree is created from the taskswarm/orch integration HEAD, so it inherits all previously merged output instead of re-inventing shared code
  • Task-packet validation — /tswarm-check + npm run check:tasks surface malformed packets (bad IDs, missing steps/criteria) with actionable causes instead of silently skipping them
  • LLM merge agent — when a lane merge into taskswarm/orch conflicts, an independent merger agent resolves the conflict semantically inside the orch worktree; unresolvable conflicts land the lane in a conflict state and pause the batch for supervisor intervention

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-cursor-codex 下一个 Next dsh-context-milvus →