TaskSwarm：DeepSeek Harness 上的多智能体任务编排插件
TaskSwarm 是 DeepSeek Harness (DSH) 上的多智能体任务编排插件，旨在自动化处理依赖任务。其核心机制是将任务按依赖关系分层为 Waves，并在同层内通过独立 Lanes 并行执行。每个 Lane 使用 Git Worktree 隔离，合并至 `taskswarm/orch` 集成分支，支持崩溃恢复与语义冲突解决。插件提供 Supervisor 协调、独立 Reviewer 审查及 LLM Merger 合并功能，并通过文件邮箱实现异步通信。内置基于 SSE 的本地 Web Dashboard 实时展示状态。安装需通过 `dsh plugin` 命令添加 npm 包
Read Moredsh-my-go：基于 DSH 的 Sisyphus 多智能体编排插件
dsh-my-go 是一款构建在 DeepSeek Harness (DSH) 之上的智能体编排系统，采用 MIT 许可证。该系统通过 Sisyphus 调度器统一管理 7 个专业子智能体（如 Hermes、Oracle 等），采用星型拓扑结构，确保同一时段仅有一个子智能体运行。其核心能力包括按工种灵活配置 provider、model 和 reasoningEffort，支持基于北京时间的错峰路由以优化资源使用，并提供 go_work 等通信工具及 Sisyphus 质检机制。模型配置在派发前经校验，若不存在则回落至父会话路由。系统提供 WebUI 设置页与树状图面板，便于可视化管理。安装需
Read Moredsh-crew：在 dsh 中用 PM 和角色代理完成变更
dsh-crew 是 DeepSeek Harness（dsh）的一个 MIT 许可证插件，旨在将 dsh 会话转化为产品经理（PM），通过调度一组角色代理（如架构师、工程师、QA、评审员等）来协作完成智能体工作流。各角色间通过磁盘文件交换信息，用户仅需与 PM 交互。核心功能分为两条路径：“Ask lane”用于提问和查询状态，不产生代码变更；“Team lane”用于执行变更，遵循“一变更一任务”原则，流程包括一轮 QA、各类评审及最终 commit。所有决策记录在仓库文档中，而 push 和 publish 操作必须由用户每次手动确认。插件要求 Node.js 18 及以上环境，支持通过
Read Moredsh-virtual-product-team: Driving DSH's Product Pipeline Through Dialogue
`songoao25/dsh-virtual-product-team` is a conversational mode preset for DeepSeek Harness (DSH), designed to automate the full product lifecycle management through a multi-agent workflow. The tool decomposes tasks into 12 stages across 8 phases, including idea validation, requirement definition, technical design, development QA, release, marketing, and iteration. Its core mechanism employs "stage gating," where progress is reported to the user at the end of each phase, and advancement to the next stage occurs only after user approval. Users are only responsible for decision-making and approval, requiring neither technical expertise nor coding skills. For installation, it supports a full installation via `./install.sh` or installing only the skill pack under a specified profile. When in use, users need to start a new conversation and enable Product in the mode selector.
Read MoreUse dsh-agent-teams to orchestrate DeepSeek Harness sessions into an intelligent agent team
`dsh-agent-teams` is a DeepSeek Harness community plugin maintained by NanmiCoder. It turns the current session into a team leader, allows spawning resumable sub-Agents by role, enables multi-agent collaboration with dependency-bound tasks, direct email access and a shared scheduler, and provides an activity panel on the web interface. This article verifies the installation command, ten tools, status directory and known limitations based on the directory page, GitHub README and usage.md. It is licensed under MIT, and you need to check the source code before installation; the plugin runs with the permissions of the current dsh process.
Read MoreY Combinator Open-Sources QM: Company-Grade Multi-Person Collaborative Agent Workspace
On July 31, 2026, Y Combinator open-sourced its internal multi-agent framework QM (Quartermaster) under the MIT license, and it quickly gained over 10,000 stars on GitHub. QM is a work-agent infrastructure for entire companies, supporting both Slack and Web terminals. Each employee and channel has isolated memory, sandboxes and permissions, and can switch Harness among Pi, OpenCode, Codex and Claude Code. This article sorts out its architecture, Strict/Auto/Dangerous security policies and qm init self-hosted deployment method, for teams evaluating enterprise-level agent operating systems for reference.
Read More