智能体编排

DeepSeek Harness 是一个基于插件架构的智能体运行与编排框架，并提供可通过命令行启动的本地 Web 界面。

iPolloWork 是一个用于协调多种智能体运行时、项目任务、插件与可编辑创作成果的桌面工作台。

一个用于发现和调度本地 Skills、分解复杂任务、编排执行流程并检查交付结果的智能体工作流插件。

HelloAGENTS 是一个面向 AI 编程 CLI 的工作流编排层，用于协调规划、执行、项目知识、验证检查和可恢复会话。

Provides agents with a unified catalog for calling APIs and CLIs, sharing skills, and managing team credentials.

SandBase Harness 是一个本地优先的 AI 智能体编排运行时，提供持久会话、沙箱工具执行、MCP 集成、记忆、凭据管理以及审计回放功能。

用于在 DeepSeek Harness 会话中创建和协调多个子智能体，管理依赖任务、团队消息、持久化状态并展示实时活动面板。

用于自托管和编排兼容 Claude API 的智能体，提供会话、沙箱、工具、记忆、持久化状态及外部服务集成。

This is a DeepSeek Harness Web plugin for creating reusable AI agent teams, orchestrating multi-model tasks in conversation workflows, and reviewing execution outcomes in a run center.

DSH Crew dispatches tasks from Claude Code or Codex to DSH agents with independent sessions, configurable tiers, and progress tracking.

Cetus 是一款 macOS 桌面智能体编排应用，可统一启动和管理多个运行时、安排后台任务，并提供屏幕上下文与工作结果审阅功能。

Adds capabilities for creating, running, governing, observing, pausing, resuming, and reusing multi-agent workflows in DSH.

面向开发工作的命令行智能体，可通过终端工具、脚本、子智能体、计划与上下文管理来编排多步骤任务。

Token Bank 是一个本地 AI 网关，用于统一接入多个智能体和模型提供商，并提供请求路由、协议转换、配置接入与用量追踪。

ForkProbe 会用同一任务并排试跑多个 AI Skill 或产物流水线，生成本地报告供用户比较并选择后续执行路径。

Axern 是一个面向 AI 智能体的编排平台，用于管理隔离代码执行、持久服务、资源生命周期和任务证据。

为 DSH 提供可生成、保存、治理、观察、暂停恢复和重跑的多智能体工作流能力。

Morning Star 是一个面向软件交付的智能体工作流插件，用于通过计划、派发、迭代、评审、验收和代码库审计流程协调多智能体协作。

为 DeepSeek Harness 提供可复用的多模型智能体团队，用于规划并执行有界工作流以及查看运行状态和令牌用量。

这是一个将 OpenBiliClaw 个性化内容推荐、内容库、对话和画像功能集成到 DeepSeek Harness，并向智能体提供桥接工具的客户端插件。

在终端中运行 DeepSeek Harness 或兼容 ACP 智能体，提供流式会话、工具调用展示、持久化会话、图片提示和可扩展界面。

All in Luna 将复杂目标拆分为具备独立上下文和依赖关系的顶层任务，协调并行执行、子代理工作、结果汇总与运行恢复。

An agent orchestration preset for DeepSeek Harness that starts sessions with a task-matched prompt and limited tools, then exposes the full Standard tool catalog after the first durable tool call or reply.

Organizes brainstorming in DeepSeek Harness into isolated branch sessions coordinated through a visual thinking tree and explicit handoffs.

一个基于 Electron 的本地 AI 工作桌面，用于管理 DeepSeek Harness 会话、智能体、项目、文件、网页研究、Git Worktree 和 Office 产物。

天枢是一个基于插件组合的编码智能体运行框架，用于协调代码任务、子智能体、工作流、终端操作、项目检索与跨会话记忆。

为 DeepSeek Harness 的工具审批链增加只读的第二模型审查器，根据配置返回允许、拒绝或委托结果并记录审查事件。

通过 Docker、Compose 或 Helm 运行 DeepSeek Harness，提供本地 Web UI、无头 CLI、持久化状态、工作区工具和可选的内嵌 Chromium 浏览器桌面。

A DSH creative preset that coordinates specialized agents for novel worldbuilding, plot planning, character management, drafting, and review.

Adds a selectable DeepSeek Harness smart-routing mode that guides tasks toward inspection-first or direct-execution workflows based on the initial request.

通过终端提供 DeepSeek Harness 智能体的交互式聊天界面，支持流式输出、工具调用审批、计划审查和会话恢复。

Adds a MuseAI roleplay experience to DSH with worldbook and character-card management, model-backed chat, interactive adventures, relationship views, and persistent sessions.

Provides dsh root agents and subagents with automatic LLM provider and model retries and fallback switching based on roles, time slots, and failure state.

A DeepSeek Harness multi-agent framework for exploring, proving, cross-validating, and tracking mathematical problem-solving projects.

IllusionAgent 是一个支持终端与浏览器交互、多模型接入、工具扩展和计划任务的多智能体编排平台。

通过角色、权限和会话管理，将任务委派给可持续运行的 Codex、Claude Code 及 ACP 子智能体。

A curated DeepSeek Harness distribution for interactive and headless coding agents with code intelligence, project editing, MCP integration, and optional debugging.

Adds an independently configured reviewer model that observes each primary agent turn and injects severity-ranked advisory notes into the session.

dsh-Kingdom is a DeepSeek Harness plugin that provides session-based multi-agent governance workflows for role setup, territory management, task planning, execution control, and review.

A DeepSeek Harness agent-orchestration plugin that manages task phases, verification debt, benchmark evidence, tool restrictions, and completion conditions from runtime events.

面向 DeepSeek Harness 的具身机器人研究插件套件，用于机器人资产检查、MuJoCo 仿真、故障诊断、实验编排、数据处理和证据报告生成。

Connects ChatGPT Web to DeepSeek Harness over MCP to create, inspect, continue, and supervise agent sessions and goal execution.

Distributes and synchronizes a router-flash agent preset for DeepSeek Harness to adjust prompting and the default preset used with DeepSeek V4 Flash.

Discover, inspect, enable, disable, and monitor local Agent Skills from the DeepSeek Harness settings page.

A DSH coordinator preset that routes tasks to domain-specific expert subagents and aggregates their results.

Create isolated one-shot or persistent side sessions without interrupting the main conversation, and view their messages, context, and tool activity in a side panel.

A DeepSeek Harness plugin suite that improves agent execution through failure handling, task planning, memory deduplication, skill loading, and file and environment tools.

一个通过可配置 profile 管理和编排前台或后台子代理，并在 Web 界面展示进度与子代理树的插件。

为 dsh 会话配置独立的审查模型，观察主模型对话并以分级建议形式注入反馈。

An agent preset for DeepSeek Harness that uses model-aware routing, staged tool exposure, context gating, and epoch-based session management to stabilize long sessions.

This plugin provides the Miraculous Standard agent preset for DeepSeek Harness, using a minimal first-request setup and staged policy to stabilize startup behavior and manage tool catalogs across model and session epochs.

一个支持智能体动态创建、持久化、热加载和卸载工具及行为钩子的自进化编排插件。

Configures DeepSeek Harness subagent model routing, role templates, delegation behavior, and continuable background execution.

A long-form web-novel writing plugin that coordinates outlining, drafting, parallel generation, lore management, research, consistency review, and local library workflows.

该插件为 DeepSeek Harness 智能体的部分高风险工具调用增加模型审查、拒绝或升级流程，并提供外部模型路由与按请求计费功能。

通过 ACP v1 将 dsh 会话连接到 BitFun，使模型能够调用可配置的 subagent_bitfun 工具委托任务执行。

通过 /btw 和 /side 创建与管理独立的侧会话，让用户在不中断主会话的情况下进行一次性提问或持续对话。

将 DeepSeek Harness 的智能体会话、工具、计划、命令和持久化能力通过 Agent Client Protocol 提供给 Zed 等客户端。

通过 Docker 镜像和 Compose 部署 DeepSeek Harness，提供 Web 与无头任务执行，并允许智能体操作挂载的工作区。

Provides a DeepSeek Harness Minecraft agent with game-data analysis, visual and keyboard-mouse control, plus version installation, authentication, and game launching.

Routes DeepSeek Harness steps between strong and cheap model tiers while providing subagent dispatch, review, failure escalation, and high-impact action gating.

通过类似 Claude Code 的终端界面编排 DeepSeek Harness 智能体、会话、命令、工具、审批和子智能体。

Provides DeepSeek Harness with registration, selection, binding, and routing for multiple agent loops and drivers.

在 VS Code 中提供 DeepSeek Harness 的智能体聊天、运行时连接、Trace 查看、会话管理以及工作区和文件操作集成。

通过飞书或 Lark 会话启动、监控、控制和审批 DeepSeek Harness 任务，并支持项目映射及文件附件交互。

为 dsh 智能体和子智能体提供基于角色的 LLM 提供商与模型备用链、重试切换、时段路由及诊断功能。

Provides DeepSeek Harness with model-failure fallback, configurable multi-agent orchestration modes, run monitoring, and resumable execution.

An installable skill for coordinating Hermes and DeepSeek Harness agents with dispatch specifications, model routing, execution boundaries, and verification gates.

A CLI and MCP server for governing AI workflows with code-model fine-tuning, inference routing, evaluations, traces, approvals, and orchestration framework integrations.

通过 MCP 将外部客户端连接到 Harness 智能体，以委派、排队并获取可执行的编码任务结果。

Configures dedicated auxiliary model routes and tools for vision, context compaction, approval reviews, subagents, session titles, and image generation in DeepSeek Harness.

在 Codex 或 Claude Code 中通过 MCP 将任务委派给 DeepSeek Harness，并支持观察、继续、取消会话及处理人工交互。

Provides native DeepSeek Harness Rooms for connecting independent Sessions and provider-backed AI members with messaging, broadcasting, and membership management.

用于让 Codex 或 Claude 与 DeepSeek Harness 页面智能体双向通信、分派任务并执行验收闭环的协作编排工具集。

MCP Lens helps DeepSeek Harness manage large MCP catalogs by searching for relevant tools, exposing selected schemas, and routing permitted calls to configured MCP servers.

为 DeepSeek Harness 提供基于独立评审模型的工具权限审批流程，并支持规则路由、拒绝熔断和人工升级。

用于协调并行 DSH 会话的文件声明、写入保护、待处理编辑和 Git 三方合并的主机插件。

通过 ACP 将 DeepSeek Harness 智能体接入 Paseo，提供模型自动发现、规划与执行模式、思考强度、斜杠命令、实时事件流和子代理支持。

面向 AI 智能体的模块化工程框架，用于编排规划、项目分析、逆向工程、编程、知识管理和验证工作流。

将其他 AI 编程工具的配置、技能、记忆和指令桥接到 DeepSeek Harness 中，便于统一使用相关上下文。

Connects DeepSeek Harness to the official OpenCode terminal UI for interacting with agents, sessions, models, permissions, and tools.

Integrates a configurable bundle of MCP servers with DeepSeek Harness for filesystem operations, browser automation, GitHub access, memory graphs, and remote HTTP tools.

为 DeepSeek Harness 的第三方模型提供推理强度设置，并支持为本地子 Agent 调用选择提供商、模型和推理级别。

面向 DeepSeek Harness 的智能体工作流插件，提供规划与审查文档流程、持久记忆、工作区检查点、分支、暂存差异、媒体处理、数据迁移和通知等功能。

A DeepSeek Harness permission gateway for reviewing, allowing, or denying AI tool calls involving files, commands, sub-agents, and sandbox access.

将 DeepSeek Harness 智能体聊天、工具调用、规划、子代理和工作区会话管理集成到 VS Code 中。

Adds a configurable We Need prompting mode to DeepSeek Harness through session tools, commands, and an optional persistent agent preset.

This plugin adapts prompts, tools, planning, search, and WebUI presentation to a Codex-style interface for selected model routes using existing DSH capabilities.

用于在 DeepSeek Harness 中启动和管理可持续运行的后台子智能体，并通过进度面板和持久化团队房间进行协作。

This is a DeepSeek Harness loop plugin that connects sessions to Kimi Code via ACP, running conversations and tool-driven execution flow within the DSH interface.

A DeepSeek Harness virtual product team mode that guides users from idea validation and requirements through planning, development, testing, release preparation, and iteration.

Provides DSH with GitHub workflow routing and orchestration for review feedback, CI diagnosis, and approval-gated publishing of local changes.

为 DeepSeek Harness 提供基于请求复杂度的模型路由、快速回答、故障降级以及会话用量和成本面板。

为 DeepSeek Harness 提供 LLM provider 回退链，在符合条件的请求失败时按配置切换到备用 provider 和模型。

Provides a remote data channel for a writing panel to access project and library content, search data, and host-side synchronization or project initialization workflows.

该插件将图片附件转换为带附件标识符的文本占位符，并委托会话中的视觉子代理读取图片内容。

为 DeepSeek Harness 提供多智能体编排能力，可派生配置独立模型与思考强度的 worker，并支持智能体间直接通信和广播。

在 DSH 侧边面板中运行独立的并行智能体会话，并将会话摘要选择性带回主对话。

A DeepSeek Harness plugin for delegating tasks to subagents with per-call model or provider selection and monitoring their live activity in a multi-panel web interface.

为 DeepSeek Harness 中的智能体工具调用提供策略检查、审批、结果扫描和决策记录。

A runtime safety guard for DeepSeek Harness agents that checks tool calls and file access for hazardous commands, sensitive data, network targets, and prompt injection before allowing, logging, blocking, or requesting approval.

为 DeepSeek Harness 选择、检查并管理内置能力，根据任务使用后台命令、符号导航、会话搜索、子智能体和工作流等原生工具。

在 VS Code 中提供连接 DSH 实例的智能体聊天窗口，用于管理工作区会话并调用文件、终端、搜索和子任务等能力。

This is a DeepSeek Harness web client plugin that adds three-mode input steering (now/next/later) plus queue management with freeze and resume controls to help users organize how pending messages are processed.

根据智能体最近的工具调用负载，自动调整后续请求的推理强度，以减少简单工具链的等待时间。

用于在同一 DeepSeek Harness 进程中的独立 Agent 会话之间查找目标、发送消息并查询投递状态的协作编排插件。

为 DeepSeek Harness 中相互独立的智能体会话提供跨会话消息、资源声明、验证、决策记录和死锁检测能力。

Creates temporary child-agent sessions in a draggable floating window, isolated from the parent conversation and cleaned up on close.

通过匹配工具调用并注入自定义规则来引导智能体行为，也可首次拒绝匹配的工具调用。

Manage DeepSeek Harness agent sessions by listing, inspecting, reading, recovering, and messaging them.

Loads and manages MCP servers per session workspace, registering project-specific tools for the corresponding agent with configuration reloads and reconnection support.

一个将 Qwen-MM 能力以 DSH 配置包形式接入的插件，用于获取能力源、启动 MCP 服务并向模型提供对应工具与技能。

CloserAI 是一个桌面端 AI 工作台，用于运行智能体并协调聊天、文档处理和代码工作流。

Routes model requests for default, planner, and subagent roles, with web settings for model selection and reasoning effort.

为 DeepSeek Harness Code Mode 提供持久 TypeScript Realm，并支持子 Agent、后台任务、工具调用与会话状态编排。

Generates clickable next-message suggestions above the chat input and inserts a selected suggestion into the draft without sending it automatically.

为 DeepSeek Harness 提供由 Orchestrator 调度多个专职子代理完成调查、决策、执行与验证的多智能体编排模式。

在 DeepSeek Harness 中让智能体向其他实时会话发起有上下文的咨询，并通过分叉会话、共享任务板和异步回复协同工作。

通过 MCP 搜索、检查并可选安装 DeepSeek Harness 插件，并在启用运行时后调用其工具。

PromptWall is a local DeepSeek Harness policy plugin that checks tool outputs for suspected prompt injection and controls credential-like data sent to network-capable tools.

通过斜杠命令将 DeepSeek Harness 接入 Claude Code，用于代码审查、批评、任务委派和可恢复会话。

An agent orchestration plugin for DeepSeek Harness that detects repetitive reasoning, retries selected failures, and continues turns that appear unfinished.

为 DeepSeek Harness 中的内置、第三方和 MCP 工具调用提供可配置的允许、审批或拒绝策略。

A web-integrated orchestration plugin for managing skill, MCP, and plugin registries, monitoring tool metrics, and controlling toolkit installation, lifecycle, rollback, selection, and mutation.

根据当前会话的近期文本和可用技能目录，为用户返回按相关性排序的技能加载建议。

为 DSH 提供基于 Underseal 适配器的智能体委派工具与执行守卫，用于协调受约束的代理工作流程。

为 DeepSeek Harness 提供一次集中提问多个问题的交互式问卷工具，并以结构化答案返回用户选择。

这是一个面向 DeepSeek Harness 的智能体编排插件，可动态生成并持久化任务工作流计划、修订记录和完成结果。

Orchestrates multi-agent tasks with GPT planning, DeepSeek worker execution, independent review, incremental Git checks, adaptive parallel scheduling, and image input.

提供可复用的结构化智能体技能模块，用于指导编码工作流、工具使用、上下文管理和智能体可靠性评估。

用于 DeepSeek Harness 的模型故障转移插件，可在模型或提供商连续失败时启用熔断、探测恢复并将后续请求路由到配置的备用模型。

为 DeepSeek Harness 提供支持工作目录、模型、提供商、角色、预设和工具过滤器按调用配置的前台与后台子代理委派工具。

AgentFuse 是一个面向 DeepSeek Harness 的智能体工具调用策略插件，可在执行前按规则允许、阻止或转交审批，并记录决策证据。

A minimal DeepSeek Harness coding-agent preset with Bash, local file editing, workspace instruction loading, and web search.

Adds a DeepSeek Harness subagent_model tool that lets each delegated call select the subagent provider, model, and output token limit.

用于通过受管理的运行时子进程和 JSON-RPC 协议编排 DeepSeek Harness 编码智能体回合的 TypeScript/Node.js SDK。

为 DeepSeek Harness 提供建模编程与论文写作两套岗位 Agent 预设，并配套团队协作、证据核验和视觉审查工作流。

通过命令行启动和编排 DeepSeek Harness，支持提示词追加、会话恢复、命名会话、结构化 JSON 输出及模型参数覆盖。

Provides isolated Git worktrees for DeepSeek Harness subagents with support for parallel task review, validation, delivery, archiving, and recovery.

通过本地兼容 OpenAI 的网关为 DeepSeek Harness 提供模型路由、子代理调度以及技能搜索与加载功能。

This is a DeepSeek Harness plugin for coordinating multi-agent long tasks where a lead plans dependencies, dispatches and reviews node work, and tracks progress through persisted task logs plus board/status views.

This plugin provides turn-level control for DeepSeek Harness chats, letting you fork and rerun conversation turns, retry or regenerate assistant messages, and navigate branch history through a version-tree interface.

Remotely assign DeepSeek Harness tasks through Feishu, WeCom, Telegram, or WeChat, receive results, and handle operation approvals.

用于浏览和分发 DeepSeek Harness 插件及 Agent 配置，并通过 xlings 安装插件、组合 Agent 配置和注册运行命令。

在 DeepSeek Harness 中集中管理来自多个任务源的任务，并编排隔离的 Agent 执行、工作区、生命周期钩子和运行状态。

Create custom AI roles, invoke them with @ mentions, and coordinate multi-agent discussions that summarize consensus, disagreements, and action suggestions.

dsh-legion is a DeepSeek Harness plugin that enables configurable semantic Profiles, Teams, and Strategies to route tasks to different subagent backends and models, providing reusable AI agent orchestration.

A DeepSeek Harness slash-command plugin that switches blank sessions between built-in agent presets and shows the current preset.

为 DeepSeek Harness 提供多种强制使用中文进行思考与输出的智能体预设，并包含中文上下文压缩引擎。

Orchestrates specialized agents in DeepSeek Harness for parallel task execution, development workflows, and runtime model presets.

用于 DeepSeek Harness 的子代理编排插件，可选择模型派发子代理并以交互式卡片实时查看其思考、任务清单和最新发言。

用于在 DeepSeek Harness 中监督多智能体与工作流，并提供图片图库、视觉模型调用和开发状态面板。

这是一个通过本地 MCP 消息桥在 Codex 与 DeepSeek Harness 之间双向委派任务、传递进度并协调协作的工具。

通过多次采样、内部投票以及计划、执行和审查阶段来编排 dsh 智能体工作流。

为 DSH Web 审批卡片增加“允许本次任务”选项，在当前任务内授权同类高权限请求并在任务结束后失效。

面向 AI Agent 的外部 Skill 发现、评估、比较与谨慎采用工作流插件。

根据请求复杂度将 DeepSeek Harness 的 LLM 请求自动分级，并通过配置的提供商链路进行路由。

为 DeepSeek Harness 提供支持按次选择模型、依赖门控、角色人设和运行花名册的子代理委派工具。

DeepJIT 通过分析智能体执行轨迹，将重复的多工具工作流编译为可复用的技能和流程模板并加载回运行中的 harness。

通过注入系统提示词，引导 DSH 智能体将图像识别任务路由到配置的 mimo-v2.5 子智能体，并在不可用时回退到 read_image。

Bridges WeChat iLink messages to DeepSeek Harness agent sessions with per-peer session management, slash commands, and an optional browser settings panel.

将本地安装的 DeepSeek Harness 集成到 VS Code 原生智能体聊天会话中，并提供会话管理、流式交互、附件、命令和模型选项。

Binds DeepSeek Harness conversations to isolated Git worktrees and branches for parallel agent collaboration and worktree management.

将 DeepSeek Harness 智能体运行时封装为 AI SDK LanguageModel，以支持文本生成、流式响应、会话管理及运行时工具调用。

Loads project-defined MCP servers from .dsh/mcp.json and dynamically exposes their tools to agent sessions with live configuration reload.

A canvas-based DeepSeek Harness plugin for configuring agent graphs, spawning agent instances, routing messages, and saving or loading cluster configurations.

通过冻结基准评测、生成候选配置并在严格比较后接受或回滚，帮助 DeepSeek Harness 迭代 Agent Profile。

用于在 DSH 中统一观测多个本机 Agent 的会话与状态，并发送指令、跟踪进度和编排任务。

为 DeepSeek Harness 的子代理委派工具增加按调用配置模型、提供方、角色和工具过滤器的能力，并支持从本地预设引用角色。

Preprocesses chat image attachments with a configured vision model and passes the resulting descriptions to a text-only DeepSeek agent.

通过技能目录搜索并按临时、项目或全局作用域安装和管理可供 dsh 智能体使用的技能。

将 DeepSeek Harness 集成为 Paseo 的 ACP provider，支持从 Paseo 客户端启动、管理和并行运行多个 dsh agent。

用于将智能体技能、规则和 MCP 配置打包，并安装或移除到多种 AI 编程运行环境中。

面向 DSH CLI 的终端智能体界面，用于管理对话、智能体预设、子智能体、会话及工具调用。

一款在终端中提供类 Claude Code 交互体验的智能体工具，支持流式模型响应、工具调用展示、会话恢复、兼容 OpenAI 的自定义模型提供商和权限控制。

Provides DeepSeek Harness with deterministic keyword-based model routing and an optional isolated image-generation channel.

A Feishu private-chat interface for creating, binding, monitoring, and approving DeepSeek Harness local-agent sessions while synchronizing task results.

为 DeepSeek Harness 提供带人工审批门控、代理交接、条件路由、Shell 步骤和持久化状态的声明式工作流引擎。

为智能体会话及其子代理树提供可持久化的令牌预算、截止时间、用量核算和请求准入控制。

为每个智能体提供按需工具发现和渐进式工具架构披露，并在不放宽现有限制的情况下加载选定的全局工具。

面向 DeepSeek Harness 的产品交付编排插件，可协调多角色智能体完成从研究、规划和实现到评审、测试及发布审计的端到端工作流。

用于在 dsh 中管理和测试 MCP 服务器、安装或迁移 skill，并导入 Claude Code 与 Codex 全局记忆的命令行及 Web 插件。

A coding-coach skill bundle that guides non-developers through structured product planning, design, implementation, testing, review, and deployment.

A DeepSeek Harness plugin for coordinating specialized agents, parallel tasks, and software-development workflow skills.

这是一个为 DeepSeek Harness Coding Agent 提供任务授权、生命周期状态、受控修改和独立审查流程的治理工作流插件。

WhaleKit 是一个面向 DeepSeek Harness 的智能体编排技能框架，通过可组合技能支持分级任务路由、规划执行、审查测试、调试和子智能体协作。

This plugin tracks observed subagent runs and requires their terminal outputs to be acknowledged before allowing a goal to complete.

A DSH orchestration plugin that preflights visible tools and injects ranked, task-relevant plugin suggestions into the agent runtime context.

Adds a dsh engineering workflow agent preset that organizes tasks through requirements clarification, plan approval, TDD, parallel subagent execution, and verified finishing.

将任务委派给 Qoder Cloud Agents，并根据提示自动路由到研究或仓库分析代理后返回执行结果。

Adds an Auto model option to DeepSeek Harness that classifies each turn and routes it to a configured fast or high-capability model.

A DeepSeek Harness orchestration suite for plugin installation and runtime loading, subagent teamwork, model routing, skill management, and cross-session communication.

为 DeepSeek Harness 提供可视化智能体工作流编辑与运行，以及针对多个模型提供商的基准评测和结果查看功能。

面向 Claude Code 和 Codex 的 UI 设计工作流编排插件，用于生成设计规范、执行审查并通过可选的浏览器预览和运行时证据完成验收闭环。

通过官方 ACP 将 DeepSeek Harness 智能体配置接入 OpenClaw，并把已配置消息渠道中的对话转发到 Harness。

DeepSeek Harness 是一个基于插件架构的智能体工作台，可通过命令行启动本地 Web 界面来运行智能体工作流。

Orchestrates router, parallel advisor, and aggregator LLM calls with streaming output, provider failover, privacy filtering, and cost controls.

DeepSee orchestrates DeepSeek coding workflows with configurable model routing, image understanding, and visual checks for interface work.

一套为 DeepSeek Harness 扩展智能体编排能力的插件集合，提供提示词与语言控制、跨智能体记忆、视觉回退、执行守卫、生图工具及技能命令处理。

Connects Zed and other IDEs to DeepSeek Harness agents over ACP, with session management, tool calls, terminal output, code diffs, and plan updates.

Provides DeepSeek Harness subagents and users with isolated Git worktree management for creation, restoration, listing, protected cleanup, and optional forced removal.

Connects the dsh web agent to local Herdr-managed agents for discovery, startup, prompting, observation, pane command execution, and workspace cleanup.

Inferglow 是一个用 Go 构建的 AI Agent 基础设施框架，用于集成大模型、注册工具并编排工作流与会话。

dsh-crew is a dsh plugin that displays live terminal panes for AI coding agents in the web UI and provides tools to start, message, inspect, and dismiss those sessions.

Connects DeepSeek to Zed through ACP for in-editor conversations, session management, and approval-based project file and command operations.

在每个顶层回合结束前启动只读验证子代理，并将未通过的缺口反馈给主代理以推动后续处理。

TaskSwarm is a multi-agent task orchestration plugin that plans dependency-ordered waves and runs, reviews, and merges tasks in parallel isolated Git worktrees.

通过回环 RPC 桥让 AI agent 控制 Godot 4.x 编辑器，执行场景操作、项目检查、调试、代码 lint 和导出等开发任务。

Adds an advisor workflow for DeepSeek Harness agents that automatically consults a configured stronger model on complex, ambiguous, or security-sensitive decisions.

通过路由提示和可按调用指定提供商与模型的工具，帮助 DeepSeek Harness 将任务委派给子代理。

为 DeepSeek Harness 提供基于任务判断的双层模型路由、模型链故障转移和受限的子智能体任务编排。

Connects DeepSeek Harness to Zed through ACP with streamed responses, model and permission controls, session management, file and terminal tools, and MCP support.

APEX is a DeepSeek Harness orchestration preset that coordinates a primary model with bounded worker sessions for code implementation, web validation, and read-only visual review.

MOLT 是一个命令行编码智能体工作流工具，可执行开发任务、生成并测试可复用技能，并将其保存到技能库中。

Automatically loads user- and project-level Claude skills, rules, instructions, and MCP configurations into DSH with workspace-aware registration and a status dashboard.

Adds persistent agent teammates to DeepSeek Harness with member messaging, shared tasks and workspaces, and a live team collaboration view.

Connects Beav’s social-media research, knowledge, copywriting, and multimedia content workflows to DeepSeek Harness.

This is a DeepSeek Harness orchestration plugin that stages tool actions in a batch inbox, lets users review full call details, and submits an approved batch for execution.

Provides a persistent Chat/Agent switch in DeepSeek Harness to control whether the current session uses its existing tools and autonomous execution flow.

Provides agent lifecycle primitives for interruptible sleep and wake-up, status awareness, hourly time updates, and immediate response to user events.

根据任务历史结果和错误记录，在配置的模型档位之间为智能体选择更合适的模型，并可选地自动覆盖模型路由。

dsh-launch is a DSH plugin for running and managing long-lived services in a detached broker process, with a sidebar and model tools for starting, stopping, restarting, listing, and viewing service logs.

Provides a terminal-based interactive chat profile for DeepSeek Harness with multi-turn sessions, streaming responses, file references, and tool-call approvals.

DSH AI-PM is a DeepSeek Harness plugin for PRD workflow orchestration, providing requirement pipeline operations and a web board flow with manual human confirmation gates before finalizing work items.

An Ink-based terminal interface for interactively coordinating multiple local agents while managing sessions, tasks, plans, and history.

An agent orchestration plugin for running multi-model coding matches in isolated Git worktrees, validating and scoring changes, reviewing diffs, and explicitly applying a selected winner.

JuneAgent is an installable AI assistant plugin that provides June-M and June-S agent presets with companion skills, and deploys them into local configuration locations for use in Claude/DSH sessions.

A project-agnostic skill that coordinates multi-agent teams with role definitions, loop guards, structured handoffs, risk-based review routing, and delivery workflows.

A configurable structured review workflow that coordinates independent panelist scoring, disagreement aggregation, critic review, and final decisions.

Provides configurable multi-expert teams for DeepSeek Harness, including team validation, routable skill registration, and lead-to-specialist child-agent orchestration.

Packages personas, models, permissions, and plugin selections into reusable scenarios that can be switched from a settings interface.

A DSH plugin for selecting different models across workflow stages such as the main conversation, subagents, and plan execution, including model selection during plan approval.

DeepSeek Harness 是一个采用插件架构、通过命令行运行并提供本地 Web 界面的智能体编排工具。

Provides each agent with progressive search and unlocking of installed tools through a per-agent tool palette.

Delegates standalone tasks in the current workspace to the locally installed Cursor CLI and provides model configuration and diagnostics.

这是一个面向 DeepSeek Harness 的终端智能体客户端，提供类 Claude Code 对话、类 Vim 模态输入、会话管理、流式回复以及工具调用交互。

这是一个增强 DSH Web 工具审批流程的插件，支持附加拒绝说明，并根据用户决定中止或继续当前智能体回合。

This plugin manages a searchable catalog of discovered skills and plugins, and when requested by the user, orchestrates existing installers to update related capability documentation and install status.

A DeepSeek Harness plugin for managing and invoking specialist agents through searchable expert catalogs, settings, and one-shot subagent delegation.

将文本处理、图像分析、OCR 和图像比较任务委托给本地 KoboldCpp 服务的 DeepSeek Harness 工具插件。

Dispatches tasks to multiple external coding-agent CLIs and returns their command output to the DeepSeek Harness conversation.

This is a DeepSeek Harness pre-step skill router that matches user input against configured rules and injects matching skill invocations before a step proceeds.

skill-bartender is an agent meta-skill that routes tasks to the minimal suitable skill workflow using a configurable policy, and supports an approval-gated flow for bringing in new skills.

A management kernel for defining, deploying, monitoring, and advancing multi-agent task pipelines with task ledgers, role sessions, automatic routing, and a web control panel.

为 DeepSeek Harness 智能体回合提供可配置的步骤数、工具调用次数和提供商令牌用量限制。

Maps DeepSeek Harness workspaces to Discord channels for inspecting session activity, monitoring subagents, and optionally driving agents remotely.

This plugin bridges supported plugins, skills, and MCP-style components from Codex, Claude Code, and Pi into DSH, and exposes /plugin-bridge commands to discover, import, and install them.

为 DeepSeek Harness 提供可按需加载的开发技能、GitHub 与浏览器工具、受信任运行时扩展及 Token 消耗监控。

DeepSeek Harness 是一个采用插件化架构并可启动本地 Web UI 的开源智能体运行框架。

将 Codex CLI 集成为 DeepSeek Harness 中可管理的智能体工具，支持并行会话、线程续接、进程控制以及在对话面板中查看活动和结果。

将 Kimi CLI 集成到 DeepSeek Harness 中，用于调用、继续、取消并在会话界面中查看外部编码智能体任务。

为 DeepSeek Harness 提供强调计划、执行、验证及并行子代理或工作流编排的智能体工作模式预设。

用于通过磁盘状态、机械校验、结构化调度、路由和审计来编排可恢复的长流程智能体任务。

DeepSeek Harness 是一个采用插件化架构、通过命令行启动并提供本地 Web UI 的智能体编排框架。

A task-orchestration plugin that routes heterogeneous tasks to suitable language models and executes them through parallel subagents with feedback rankings and usage statistics.

为 DeepSeek Harness Agent 对受支持的结构化文件修改设置每轮文件数、调用数和 UTF-8 字节数预算，并在超限前拒绝调用。

Orchestrates long-running tasks by coordinating planning, delegated subtasks, scoring and retries, model-based review, and persisted task state.

Provides a DeepSeek Harness preset compatible with opencode and oh-my-openagent, supporting multi-role orchestration, model routing, filesystem and terminal tools, search, and project-context injection.

This is a DeepSeek Harness plugin that provides a persistent Design Studio experience with live preview, enabling design-file management and element-based edits through coordinated AI flows with DeepSeek and a design agent.

A dsh plugin for orchestrating multi-agent software development workflows and deploying bundled skills and specialist agent prompts.

A read-only compatibility bridge that exposes Claude Code rules, skills, agent definitions, and command hooks to DeepSeek Harness with optional project-specific adaptations.

为 DeepSeek Harness 提供按类别整理的 1300 多个技能和 94 个智能体定义，并附带将技能安装到指定目录的工具与说明。

Provides a minimal-prompt PTC coding-agent mode with multi-step orchestration, file operations, shell execution, planning, subagents, workflows, and Windows Git Bash and persistent PowerShell support.

Uses configured external AI models for second-opinion and adversarial code reviews, task delegation, and background job management.

通过 TapPass 策略服务评估智能体的每次工具调用，并支持放行、拒绝或转交人工审批。

This is a DeepSeek Harness plugin that provides a `bridge` agent skill to migrate from a locked session to a target preset by generating a fixed-schema handoff summary and starting a new session from it.

Provides DeepSeek Harness with a bounded, checkpointed, and cancellable autonomous agent loop, plus interfaces for GitHub plugin discovery, AI-assisted installation, and workspace registration.

Manage subagent delegation through named roles and user-configured model routes, with tools to inspect roles and start continuable child agents.

将自然语言需求组装为经过会话验证的单个或多智能体预设，并整合目录中的 MCP 能力与交付产物。

为 DeepSeek Harness 的子代理调用提供逐次模型与推理强度选择，并支持子会话编排、前后台执行和模型发现。

A DSH plugin that orchestrates a nine-player Werewolf game with AI roles, stateful actions, cross-game experience accumulation, and human participation through a local browser panel.

sai 是一款面向 Android 的本地优先编码智能体，可在工作区中协调模型会话、文件与终端操作、Git 及浏览器工具，并通过审批执行任务。

Enables live DeepSeek Harness sessions in the same process to discover, inspect, message, steer, and observe one another.

Provides DeepSeek Harness tools to search a pinned catalog of agent skills and install selected skills into the current project.

Provides agents with progressive tool discovery, schema disclosure, and runtime call routing while keeping the model-visible tool surface stable.

Bridges projects configured for Claude Code, CodeBuddy Code, OpenCode, Codex, Pi, Gemini CLI, or Cursor into DeepSeek Harness while preserving supported skills, commands, memory, and related configuration.

Routes DSH top-level sessions between DeepSeek and registered pluggable loop engines according to presets, while managing preset installation and restoration.

Delegates coding and file-related tasks to the local Google Antigravity CLI agent with follow-up conversations, background execution, task management, and login support.

A Windows-native DeepSeek Harness agent preset that adapts task routing and selectively enables file tools, PowerShell, Git Bash, workflows, background jobs, and subagents.

基于插件架构运行智能体并通过本地 Web 界面或命令行进行交互的开发者预览工具。

Provides configurable allow, deny, or ask policies for agent tool calls, structured audit logging, and token quota management.

Helps users discover, expand, and combine local plugin recipes into ordered environment setup command sequences.

Dispatches tasks to multiple external coding-agent CLIs in headless mode and brings their results back into a DeepSeek Harness session.

Routes tasks to the smallest suitable set of skills, provides editable routing and skill audits, and requires quarantine, scanning, and explicit human approval before installing missing skills.

根据任务类型自动在配置的 DeepSeek 模型之间切换，并尊重手动模型选择。

Adds an auto-review permission mode for DeepSeek Harness that approves selected low-risk sandbox escalations, routes ambiguous requests for confirmation, and rejects configured destructive actions.

A DeepSeek Harness plugin for validating and executing bounded declarative task DAGs to coordinate multiple subagent tasks through the existing workflow engine.

Adds a smart_subagent tool that lets each call select a model and reasoning effort while supporting foreground and continuable background child agents.

DeepSeek Harness 是一个采用插件化架构并提供本地 Web UI 的智能体编排框架。

DeepSeek Harness 是一个采用插件化架构并通过本地 Web UI 使用的智能体编排框架。

dsh-proactive is a DSH/Cordis proactive scheduling plugin that provides task sensing, decision, execution, and reflection workflows with 18 runtime and memory-oriented tools for automated multi-model orchestration and governance.

A DeepSeek Harness companion plugin for conversation export and search, context handoff, cost management, prompt and multi-model evaluation, task orchestration, and auditing.

An enterprise DeepSeek Harness plugin for multi-step task orchestration, automated execution, team collaboration, knowledge management, security auditing, and conversation search and export.

dsh-proactive is a DSH plugin that provides task sensing, decision, and execution orchestration for multi-model workflows, with memory, reflection, and strategy tuning to improve scheduling over time.

This plugin provides DeepSeek Harness with runtime orchestration for tool calls by combining pre-checks, feedback-based control, and status checks to moderate repetitive or unstable execution loops.

Adds a user choice gate for agent skill triggers, presenting described candidates and falling back to the system default after a timeout.

Configures ordered provider chains for same-named models and automatically fails over to the next provider after terminal errors or exhausted retries.

从 Claude Code、Kiro、OpenCode、Codex 及 Harness 配置文件中发现并合并 MCP 服务器，并将其同步挂载到 DeepSeek Harness。

DeepSeek Harness 是一个采用插件化架构并提供本地 Web UI 的智能体编排框架。

A DeepSeek Harness data-loss-prevention plugin that intercepts selected credential and secret-related tool calls, redacts results and telemetry, and records audit decisions.

Connects local or remote MCP servers to DeepSeek Harness and automatically registers their tools for agent use.

An agent-orchestration preset for DeepSeek Harness that plans, delegates, and synthesizes tasks while gating tool execution in the root session.

This is a DeepSeek Harness plugin for agent orchestration that splits user tasks into subtasks, runs sub-agents in parallel, and converges their outputs.

This is a DeepSeek Harness plugin that injects methodology guidance into each new agent session, exposes retrieval and analysis tools, and provides a settings interface for enabling and configuring its behavior.

为 DeepSeek Harness 提供可安装和注册的多角色专家 Skill 套件，并支持生成及编排基于 Skill 的智能体工作流。

Generates minimal-mode agent presets for DeepSeek Harness on Windows with pwsh, Git Bash, and WSL2 support, and restores the configuration on uninstall.

A two-phase DeepSeek Harness agent preset that starts with minimal tools, then promotes the session to standard tools and delayed workspace instructions with Windows shell support.

Adds real dice rolls, check resolution, lore and character-card context, tabletop session state, commands, and DiceFrame save interoperability to DSH roleplaying sessions.

This plugin lets dsh users call custom subagents in chat with @mentions, using Markdown-defined prompts and per-agent model routing for delegated tasks.

根据请求复杂度在已配置的模型档位之间自动选择，并跟踪会话和每日费用预算。

根据请求上下文中的图片内容，自动将需要视觉理解的智能体请求路由到配置的多模态模型。

This is a DeepSeek Harness ACP server plugin that lets ACP-compatible clients connect via stdio or HTTP/SSE to run and manage dsh agent sessions with interactive prompts and tool-driven workflow events.

Enables DSH agents to dynamically create, update, remove, and persist callable tools through managed TypeScript modules.

This is a DeepSeek Harness plugin bundle that registers Claw Orchestrator as an ACP subagent provider in dsh, enabling delegation to single-engine or multi-engine collaboration modes for tasks.

A local-first DeepSeek Harness plugin that orchestrates issue workflows with workspace-scoped boards, per-issue agent sessions, automatic scheduling, and human approval gates.

Polaris is a DSH orchestration plugin that aggregates discovery, audit, execution, verification, and retrieval capabilities behind a single interface and routes requests to suitable implementations for automated development workflows.

dsh-kanban adds a web-based task board inside DSH with card management tools and model actions, and can dispatch ready cards to a local DSH headless worker flow for execution.

This is a dsh-TUI plugin that lets users configure subagents to use separate model/provider and reasoning-effort settings, and check active routing via /model-switch.

Provides session orchestration tools for creating, listing, and sending queued or interrupt prompts to ordinary DeepSeek Harness sessions.

This plugin provides a plug-and-play AI role library for DeepSeek Harness, enabling conversation-based loading of different roles on demand to assist with tasks.

Provides dsh with an on-demand catalog of Claude Code skill templates covering development, research, document processing, automation, and other workflows.

This plugin adds an `/octo` command in DSH to run multi-model agent workflows for tasks like research, development, review, and debate, with results coordinated through a shared workflow process.

A comprehensive DeepSeek Harness skill pack offering callable agent skills for engineering, research, business, and productivity workflows.

This is a DeepSeek Harness plugin that loads 20 Chinese AI coding skills (such as TDD, debugging, code review, and planning) and lets the agent match and call them automatically by scenario.

A DeepSeek Harness engine for running multi-agent shows and games with scenario-defined state, evidence, machine adjudication, replayable logs, and a web interface.

dsh-evolve-modes is a DeepSeek Harness web plugin that provides composable controls for task execution, reasoning strategy, quality checks, and human-reviewed self-evolution rules to configure and review AI agent workflows.

为 DeepSeek Harness 提供可独立配置的辅助模型路由，支持视觉理解、上下文压缩、审批审查、子代理、会话标题和图像生成等任务。

Adds a customizable interrupt button to DeepSeek Harness WebUI that pauses agent work, stops related tasks, and prompts a progress summary before awaiting new requirements.

This plugin provides a DSH agent preset that starts main and subagents in a minimal bootstrap persona/tool set, then promotes them to the full tool catalog after the first durable interaction.

This is a DSH plugin for cross-session communication, providing role-based message routing with persistent delivery and support for coordinating coding-review collaboration workflows.

This plugin adds local-first model tools to DSH so text and image tasks can be handled locally first, with fallback to a cloud model when local execution is not suitable.

Provides configurable advisor, reviewer, and designer roles that invoke the Claude Code CLI for bounded external expert consultations and parallel panels.

A DeepSeek Harness plugin that decomposes tasks for parallel subagent execution, aggregates results, and displays live progress in chat.

This plugin adds GROX Network integration to DeepSeek Harness, letting users find and hire agents, check hire status, and delegate tasks to a GROX subagent provider with returned outcomes.

A DeepSeek Harness bridge for discovering and verifying RoleHub roles, starting independent role-scoped Sessions, and optionally attaching them to Agent Team Rooms.

Runs Codex app-server tasks as one-shot subagents and projects their results, command executions, and file changes into child sessions in the Web UI.

This is a DeepSeek Harness host-side Cordis plugin that reorders each Agent step so completed subagent callback responses are processed before ordinary context messages.

This plugin adds a session-header mode switcher in DeepSeek Harness so users can change agent presets mid-conversation, including Standard, Code, Minimal, Creator, and custom presets.

在发送前调用提示词工程智能体补全、消歧、扩写并迭代优化草稿，支持版本回溯和直接应用到输入框。

Provides a DeepSeek Harness profile bundle of agent skills for engineering and productivity tasks.

This is a DeepSeek Harness plugin that creates and manages multi-agent teams via `teams_meta_*` tools, delegates and tracks tasks, and shows team and routing state in a read-only sidebar view.

面向个人开发者的本地自托管编码代理工作台，用任务会话、评论审批、项目知识库和上下文装配来管理代理执行。

This is a DeepSeek Harness web plugin that adds a Cyber Company workspace tab, showing staff and office views while routing mentions and coordinating multi-agent collaboration in session chat.

Merges multiple queued text follow-ups into one synthesized prompt for the next DeepSeek Harness turn, with a choice between merged and individual queue processing.

AutoEvo orchestrates DSH agent workflows for discovering, reviewing, reusing, installing, verifying, upgrading, and creating capabilities under controlled approvals.

This plugin lets Codex work with DeepSeek Harness sessions as a local bridge, allowing task dispatch and session-level collaboration from Codex tools.

KnowzCode is a DeepSeek Harness plugin that provides TDD and quality-gate workflow skills such as start-work, fix, regroup, and continue to help coordinate development execution.

Orchestrates multi-round pro and con subagent debates through a /debate command, with optional neutral judging and web card results.

通过实时语音控制 DeepSeek Harness，并在多个会话之间创建、调度和管理任务。

面向 DeepSeek Harness 编码代理的运行时治理插件，用于进度监控、证据校验、完成声明约束、能力路由以及 Linux/WSL 执行集成。

一个已归档的 DeepSeek Harness 运维工具包，用于证据驱动的智能体编排、受限本地记忆检索、仓库审计、基准测试准备和插件发布规划。

Imports SillyTavern character cards and lorebooks, creates roleplay workspaces and sessions, and manages prompts, context, variables, and compatible scripts.

This is a DeepSeek Harness plugin that mounts a Chinese `lion-skills` workflow skill bundle on `ctx.skills`, providing developer support skills such as code review, debugging, testing, naming, spec writing, and task breakdown.

为 DeepSeek Harness 工作流提供 Git worktree 隔离适配器，为每个任务创建并管理独立的智能体工作目录。

This plugin helps DeepSeek Harness users turn a paper into a planned academic talk by using paper retrieval, venue-aware timing rules, and slide generation steps to produce slide content with speaker-note scripts.

This DeepSeek Harness plugin adds a `debate` tool that runs multiple sub-agents in parallel rounds to gather evidence, cross-debate, and converge on conclusions, with background execution and resumable progress.

This DeepSeek Harness plugin provides orchestration tools to list live sessions and send task messages into another session, waking it to work independently.

oh-my-dsh is a DeepSeek Harness plugin that adds think/build model-tier routing with /omd controls, plus vision delegation and image-generation flows to switch model behavior per session workflow.

为 DeepSeek Harness 提供 ACP 会话控制、编码配置选择，并将 ACP 提供的 MCP 服务器挂载到智能体作用域中。

在 VS Code 中连接 DeepSeek Harness，发起智能体会话、查看流式工具活动，并以原生差异编辑器查看文件变更。