Commander One goal in.
One coherent delivery out. 给出一个目标,
交付一个完整结果。
Commander understands the outcome, constraints, inputs, and success criteria; chooses the strongest owner across agents, skills, connectors, tools, or direct work; then keeps execution moving, checks every result, and preserves progress through interruptions. 指挥官理解你需要的结果、约束、输入和成功标准,在 Agent、技能、连接器、工具和直接执行之间选择最合适的负责人,持续推进并核对每个结果,即使中断也会保留进度。
- Plans parallel and dependent work 编排并行与依赖任务
- Verifies before delivery 交付前核对结果
- Preserves context and progress 保留上下文与进度
Prepare the launch of our new AI meeting assistant: research the market, refine the positioning, design the landing page, and create launch assets. 准备发布新的 AI 会议助手:调研市场、明确定位、设计落地页并制作发布素材。
Ship ten official-Agent pages without managing five separate conversations.一次交付 10 个官方 Agent 页面,不必分别管理 5 段对话。
In this example, Commander keeps one outcome visible while specialists own research, copy, design, implementation, and verification.在这个示例中,指挥官保持一个总目标,并让专才分别负责研究、文案、设计、实现和验证。
“Add dedicated landing pages for Commander and the nine official Agents. Use their real workflows, support Chinese and English, show a realistic usage example, and keep one clear Orkas download action.”“为指挥官和 9 个官方 Agent 增加专属落地页。结合真实工作流、适配中英文、展示可信的使用示例,并保留一个清晰的 Orkas 下载入口。”
Coordination is a delivery responsibility, not a forwarding step. 调度不是简单转发,而是对最终交付负责。
Goal and success criteria 目标与成功标准
Turns a vague request into an explicit outcome, inputs, constraints, and delivery format. 把模糊请求整理为明确结果、输入、约束和交付形式。
- scope and non-goals
- available context
- completion evidence
Execution shape 执行方式
Runs independent tasks in parallel, dependent work in order, and tightly coupled work with one owner. 独立任务并行、依赖任务串行,紧密耦合的工作保持同一负责人。
- specialist handoff
- direct work
- progress continuity
Verification and recovery 验证与恢复
Reads every result, checks gaps and conflicts, reroutes when needed, and never packages blocked work as complete. 完整读取结果,核对缺口与冲突,需要时改派,不把受阻工作包装成完成。
- evidence review
- reroute or retry
- smallest resume condition
Understand, assemble, execute, verify. 理解、组队、执行、核验。
Understand the outcome 理解结果
Clarify what should exist when the work is done. 明确工作完成时应该得到什么。
Gather context 读取上下文
Use conversation, files, Library, instructions, memory, and backlog without asking twice. 使用对话、文件、资料库、指令、记忆和任务队列,不重复询问。
Coordinate execution 协调执行
Choose the right owners and sequence work by real dependencies. 选择合适负责人,并按真实依赖关系推进。
Verify and deliver 核验并交付
Resolve gaps and return one directly usable answer, file, decision, or next action. 处理缺口,交付一份可直接使用的答复、文件、决策或下一步。
You keep control of tools, data paths, and consequential actions. 工具、数据路径和重要操作仍由你掌控。
Project files, conversations, Agent settings, and generated outputs stay on your machine by default. 项目文件、对话、Agent 设置和生成产物默认保存在本机。
Commander uses a stronger specialist when one fits, and handles general work directly only when appropriate. 有合适专才时优先交给专才,没有更合适能力时才直接处理通用工作。
Missing permission, input, dependency, or verification remains visible with the smallest actionable recovery step. 缺少权限、输入、依赖或验证时会明确说明,并给出最小可执行恢复步骤。
Commander questions 关于指挥官
Do I need to choose the right Agent myself? 我需要自己判断该用哪个 Agent 吗?
No. You can name a specialist when you want one, or describe the outcome to Commander. Commander respects your choice and otherwise selects the strongest owner for each part. 不需要。你可以指定专才,也可以只向指挥官描述结果。指挥官会尊重你的指定,否则会为每部分选择最合适的负责人。
Can Commander do work directly? 指挥官可以直接执行工作吗?
Yes. When no specialist is a better fit, it can handle general research, analysis, writing, file work, scripting, and automation within the permissions you grant. 可以。没有更合适专才时,它能在你的授权范围内直接完成通用研究、分析、写作、文件处理、脚本和自动化。
What happens if work is interrupted? 任务中断后会怎样?
Commander preserves the original goal, completed work, and resume condition so the task can continue without silently discarding useful progress. 指挥官会保留原目标、已完成工作和恢复条件,继续时不会悄悄丢掉已有进度。
Give Commander the next real goal. 把下一个真实目标交给指挥官。
Download Orkas and coordinate research, creation, development, office delivery, and follow-up from one conversation. 下载 Orkas,在一段对话里协调研究、创作、研发、办公交付和后续跟进。
macOS and Windows · Managed official models or your own provider 支持 macOS 与 Windows · 官方模型或自有供应商