magi-ex

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

MAGI EX

Multi-model brainstorming panel. Spawns Scientist, Mother, and Woman teammates backed by Claude Opus, OpenAI Codex, and Google Gemini to explore a question or topic in parallel, then consolidates their proposals for the user.

多模型头脑风暴小组。会创建分别由Claude Opus、OpenAI Codex和Google Gemini驱动的科学家、母亲、女性三个协作角色，并行探索某个问题或主题，之后为用户整合它们的提案。

Checklist

操作清单

Follow these steps in order.

Clarify the question
Set up agent team
Parallel exploration
Consolidate and present options
Wait for user decision (write a plan, debate, or done)
Tear down agent team

请按顺序执行以下步骤：

明确问题
搭建Agent团队
并行探索
整合并展示选项
等待用户决策（编写方案、辩论或结束）
销毁Agent团队

Workflow

工作流

dot

digraph magi_ex {
    rankdir=TB;
    node [shape=box, style=rounded];

    clarify [label="Lead clarifies question\n(AskUserQuestion if underspecified)"];
    setup [label="Setup 3-agent team\n(Claude, Codex, Gemini)"];
    explore [label="Teammates explore in parallel\n(each delegates to its model)"];
    consolidate [label="Lead consolidates proposals\n+ attributes model source"];
    present [label="Present options to user\n(AskUserQuestion: Write a plan / Debate / Done)"];
    decide [label="User decision" shape=diamond];
    debate [label="Debate round\n(lead broadcasts, each critiques via its model)"];
    teardown [label="Shut down agent team"];
    handoff [label="Handoff to writing-plans"];

    clarify -> setup;
    setup -> explore;
    explore -> consolidate;
    consolidate -> present;
    present -> decide;
    decide -> debate [label="debate"];
    decide -> teardown [label="done"];
    decide -> teardown [label="write a plan"];
    teardown -> handoff [label="write a plan"];
    debate -> present;
}

dot

digraph magi_ex {
    rankdir=TB;
    node [shape=box, style=rounded];

    clarify [label="Lead clarifies question\n(AskUserQuestion if underspecified)"];
    setup [label="Setup 3-agent team\n(Claude, Codex, Gemini)"];
    explore [label="Teammates explore in parallel\n(each delegates to its model)"];
    consolidate [label="Lead consolidates proposals\n+ attributes model source"];
    present [label="Present options to user\n(AskUserQuestion: Write a plan / Debate / Done)"];
    decide [label="User decision" shape=diamond];
    debate [label="Debate round\n(lead broadcasts, each critiques via its model)"];
    teardown [label="Shut down agent team"];
    handoff [label="Handoff to writing-plans"];

    clarify -> setup;
    setup -> explore;
    explore -> consolidate;
    consolidate -> present;
    present -> decide;
    decide -> debate [label="debate"];
    decide -> teardown [label="done"];
    decide -> teardown [label="write a plan"];
    teardown -> handoff [label="write a plan"];
    debate -> present;
}

The Process

流程

User Question

用户问题

<user_question> $ARGUMENTS </user_question>

Clarify

明确问题

Before spawning teammates, the lead asks via

AskUserQuestion

to understand the idea or topic:

Ask questions one at a time to refine the idea
Prefer multiple choice questions when possible, but open-ended is fine too
Only one question per message
Focus on understanding: purpose, constraints, success criteria

Skip if the question is already clear and actionable. Include all clarified context in teammate spawn prompts.

在创建协作角色之前，负责人会通过

AskUserQuestion

提问，以充分理解需求或主题：

每次仅提一个问题来细化需求
优先使用选择题，开放问题也可接受
每条消息仅包含一个问题
聚焦于理解：目标、约束、成功标准

如果问题已经清晰可执行则跳过此步骤。将所有明确后的上下文都加入协作角色的初始化提示词中。

Setup

搭建团队

Read all 3 template files: MAGI-1.md, MAGI-2.md, MAGI-3.md
Read both reference files: codex.md, gemini.md

TeamCreate

with a descriptive team name

magi-{topic}

(e.g.,

magi-auth-strategy

)

Spawn all 3 teammates in a single message (3 parallel
```
Task
```
calls):

Teammate	`name`	`subagent_type`	Prompt includes
Scientist	`scientist`	`general-purpose`	MAGI-1.md template + user question + clarified context
Mother	`mother`	`general-purpose`	MAGI-2.md template + codex.md reference + user question + clarified context
Woman	`woman`	`general-purpose`	MAGI-3.md template + gemini.md reference + user question + clarified context

Set
```
team_name
```
on each
```
Task
```
call to the team name from above
Teammates don't inherit the lead's conversation history -- include all context in the spawn prompt

读取全部3个模板文件：MAGI-1.md、MAGI-2.md、MAGI-3.md
读取两个参考文件：codex.md、gemini.md
调用
```
TeamCreate
```
创建描述性的团队名
```
magi-{topic}
```
（例如
```
magi-auth-strategy
```
）
在单条消息中创建全部3个协作角色（3个并行的
```
Task
```
调用）：

协作角色	`name`	`subagent_type`	提示词包含内容
科学家	`scientist`	`general-purpose`	MAGI-1.md模板 + 用户问题 + 明确后的上下文
母亲	`mother`	`general-purpose`	MAGI-2.md模板 + codex.md参考 + 用户问题 + 明确后的上下文
女性	`woman`	`general-purpose`	MAGI-3.md模板 + gemini.md参考 + 用户问题 + 明确后的上下文

在每个
```
Task
```
调用中设置
```
team_name
```
为上述创建的团队名
协作角色不会继承负责人的对话历史，所有上下文都需要包含在初始化提示词中

Explore in Parallel

并行探索

Teammates begin working immediately upon spawning. The lead's role is coordination only:

Wait for teammates to send their proposals via
```
SendMessage
```
Forward teammate clarifying questions to the user via
```
AskUserQuestion
```
-- note which teammate (and which model) asked
Your role is to wait and coordinate -- teammates produce all proposals

Each teammate follows their own model-specific checklist defined in their template.

协作角色创建后会立即开始工作。负责人仅承担协调角色：

等待协作角色通过
```
SendMessage
```
提交它们的提案
将协作角色的澄清问题通过
```
AskUserQuestion
```
转发给用户，注明是哪个协作角色（以及对应哪个模型）提出的问题
你的职责是等待和协调，所有提案都由协作角色产出

每个协作角色会遵循其模板中定义的特定模型检查清单执行任务。

Consolidate + Present

整合与展示

Lead collects all proposals from the 3 teammates, then:

Deduplicates similar proposals (attributing to all teammates/models that proposed it)
Groups by theme if there are many proposals
Presents each option with:
- Which teammate(s) and model(s) proposed it (e.g., "Scientist [Opus]", "Mother [Codex]")
- Trade-off analysis from each perspective
- Who tagged it as their top pick and why
Asks the user via
```
AskUserQuestion
```
to select an option (one option per choice)
Asks the user via
```
AskUserQuestion
```
what to do next:
- Write a plan -- triggers teardown + handoff to
```
writing-plans
```
  with the selected option
- Debate -- teammates critique each other's proposals (triggers debate round below)
- Done -- shut down the agent team, no further action

负责人收集3个协作角色的所有提案后，执行以下操作：

去重相似提案（标注所有提出该方案的协作角色/模型）
如果提案数量多，按主题分组
展示每个选项时附带：
- 提出该选项的协作角色和对应模型（例如"科学家 [Opus]"、"母亲 [Codex]"）
- 各个视角的权衡分析
- 谁将其标注为首选方案及理由
通过
```
AskUserQuestion
```
请用户选择一个选项（每个选项为独立选择项）
通过
```
AskUserQuestion
```
请用户选择后续操作：
- 编写方案 -- 触发团队销毁，携带选中的选项交接给
```
writing-plans
```
- 辩论 -- 协作角色互相批评对方的提案（触发下方的辩论环节）
- 结束 -- 关闭Agent团队，不执行后续操作

Optional Debate (user-triggered)

可选辩论（用户触发）

Only runs if the user requests it. When triggered:

Lead broadcasts the consolidated option list to all 3 teammates via
```
SendMessage
```
Each teammate sends the proposals to their model for critique:
- Scientist (Opus): reasons directly about other proposals
- Mother (Codex): calls
```
mcp__codex__codex-reply
```
  with saved
```
threadId
```
- Woman (Gemini): pipes full context + proposals to
```
gemini -p -
```
Each teammate sends their model's critique back to the lead via
```
SendMessage
```
Lead collects updated stances and re-presents to the user

One round per debate request. The user can trigger multiple sequential debates.

仅在用户请求时运行。触发后执行：

负责人通过
```
SendMessage
```
将整合后的选项列表广播给所有3个协作角色
每个协作角色将提案发送给对应的模型进行批评：
- 科学家（Opus）：直接对其他提案进行推理分析
- 母亲（Codex）：使用保存的
```
threadId
```
  调用
```
mcp__codex__codex-reply
```
- 女性（Gemini）：将完整上下文+提案传递给
```
gemini -p -
```
每个协作角色将其模型的批评意见通过
```
SendMessage
```
发回给负责人
负责人收集更新后的立场，重新展示给用户

每次辩论请求仅执行一轮，用户可以触发多轮连续辩论。

Teardown

销毁团队

When to tear down

销毁时机

User selects Write a plan
User selects Done

用户选择编写方案
用户选择结束

Keep teammates alive during the debate loop

辩论循环期间保持协作角色运行

Tear down only after the user selects Write a plan or Done.

仅在用户选择编写方案或结束后再执行销毁。

Shutdown sequence

关闭流程

```
shutdown_request
```
to each teammate (Scientist, Mother, Woman)
Wait for all shutdown approvals
```
TeamDelete
```
(fails if teammates are still active)

向每个协作角色（科学家、母亲、女性）发送
```
shutdown_request
```
等待所有关闭确认
调用
```
TeamDelete
```
（如果协作角色仍在运行会调用失败）

Handoff (write a plan path only)

交接（仅编写方案路径）

After teardown, invoke

writing-plans

skill with the chosen option(s) as context.

销毁团队后，以选中的选项为上下文调用

writing-plans

skill。