host-panel

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Host Panel

主持专家小组讨论

Surface real tensions, frameworks, and disagreements through simulated expert discourse — not theatrical roleplay.

This panel explores a complex topic from multiple angles — surfacing frameworks and genuine disagreements, not producing consensus or truth.

Invocation:

/host-panel "topic" [format] [num-experts]

Format	Purpose	Best for
`roundtable`	Open multi-perspectival exploration	Broad topics, brainstorming, mapping a field
`oxford`	Binary debate with formal sides	Policy decisions, testing propositions
`socratic`	Deep inquiry through questioning	Conceptual analysis, definitional disputes

Defaults: roundtable format, 4 experts.

Expert range: 2-6. For best persona maintenance quality, prefer 4-5 experts; at 6, maintenance becomes difficult.

To add a format: add its phase guide to

./references/formats.md

and update the format table and auto-selection logic above.

通过模拟专家对话呈现真实的观点冲突、分析框架与分歧，而非戏剧化角色扮演。

本小组将从多个角度探索复杂话题——聚焦分析框架与真实分歧，而非达成共识或输出“真理”。

调用方式：

/host-panel "话题" [格式] [专家人数]

格式	用途	适用场景
`roundtable`	开放式多视角探索	宽泛话题、头脑风暴、领域梳理
`oxford`	正反方正式辩论	政策决策、命题验证
`socratic`	提问式深度探究	概念分析、定义争议

默认设置： roundtable格式，4位专家。

专家人数范围： 2-6人。为保证角色一致性，建议选择4-5位专家；若选6人，角色维护难度会提升。

若需新增格式：请将该格式的阶段指南添加至

./references/formats.md

，并更新上述格式表格与自动选择逻辑。

1. Argument Parsing & Topic Diagnostic

1. 参数解析与话题诊断

Parsing

参数解析

Parse

$ARGUMENTS

: a quoted string is the topic (required), an integer 2-6 is expert count, a keyword (

roundtable

oxford

socratic

) is format. Order of count and format does not matter. Defaults: roundtable, 4 experts.

Format-count notes: Oxford with 2 experts runs as direct proposition-vs-opposition without swing or floor questions. Oxford with 3 designates one swing per formats.md. Socratic with 2 runs as paired inquiry — both panelists question each other under moderator guidance.

$ARGUMENTS

is empty, present this example gallery and ask the user to choose or provide their own:

#	Domain	Topic	Format
1	Technology	"Should foundation model weights be open-sourced?"	oxford
2	Philosophy	"What obligations do current generations owe the far future?"	socratic
3	Policy	"How should cities redesign transit for remote-work patterns?"	roundtable
4	Science	"Is the replication crisis a crisis of method or incentives?"	roundtable

For 2 experts, the panel becomes a structured dialogue. Alternate direct engagement between the two participants. Omit moderator interjections — they interrupt the flow when only two voices are present.

解析

$ARGUMENTS

：带引号的字符串为话题（必填），2-6的整数为专家人数，关键词（

roundtable

oxford

socratic

）为讨论格式。人数与格式的顺序不影响解析。默认值：roundtable格式，4位专家。

格式与人数说明： 牛津式辩论配2位专家时，为直接的正反方辩论，无中间立场或观众提问环节；配3位专家时，将按照formats.md的设定指定1位中间立场专家。苏格拉底式讨论配2位专家时，为成对探究——两位参与者在主持人引导下互相提问。

若

$ARGUMENTS

为空，展示以下示例库并请用户选择或自定义：

序号	领域	话题	格式
1	科技	"基础模型权重是否应该开源？"	oxford
2	哲学	"当代人对遥远未来的世代负有何种义务？"	socratic
3	政策	"城市应如何针对远程办公模式重新设计公共交通？"	roundtable
4	科学	"可重复性危机是方法问题还是激励机制问题？"	roundtable

当仅2位专家时，小组讨论变为结构化对话。两位参与者直接交替互动，省略主持人插话——仅两人对话时，插话会打断交流节奏。

Topic Suitability Diagnostic

话题适配性诊断

Before proceeding, evaluate the topic:

Signal	Action	Pause?
Settled science	Reframe toward open question	Yes
Too broad	Suggest narrowing with specific example	Yes
Too narrow for expert count	Reduce panel or suggest broadening	Yes
Highly specialized	Flag research grounding as critical	No — extra rigor
Asymmetric evidence	Reframe around genuine tensions within consensus	No — reframe
Casual / experiential	Use practitioners and cultural commentators — same rigor, matched register	No — match register

开始前先评估话题：

信号	操作	是否需要暂停？
已定论的科学问题	重构为开放式问题	是
过于宽泛	建议通过具体示例缩小范围	是
范围过窄，与专家人数不匹配	减少专家人数或建议拓宽话题	是
高度专业化	标记需以研究为基础	否——需额外严谨性
证据不对称	围绕共识内的真实冲突重构话题	否——重构即可
日常/经验类话题	邀请从业者与文化评论者参与——保持同等严谨性，匹配话题风格	否——匹配风格即可

Format Auto-Selection

格式自动选择

If the user omitted format, select based on topic structure:

Binary proposition ("Should X...", "Is Y better than Z...") ->
```
oxford
```
Open exploration ("What are the implications of...", "How should we think about...") ->
```
roundtable
```
Deep conceptual inquiry ("What does X mean?", "Is Y coherent?") ->
```
socratic
```

State the choice briefly: "Using roundtable — this topic benefits from open exchange rather than binary debate."

若用户未指定格式，根据话题结构自动选择：

二元命题（如“是否应该X...”、“Y是否比Z更好...”）→
```
oxford
```
开放式探索（如“X的影响有哪些...”、“我们应如何看待Y...”）→
```
roundtable
```
深度概念探究（如“X的定义是什么？”、“Y是否连贯？”）→
```
socratic
```

需简要说明选择理由：例如“采用roundtable格式——该话题适合开放式交流，而非二元辩论。”

2. Topic Analysis & Research Grounding

2. 话题分析与研究基础

This is the critical step that determines panel quality. Complete it BEFORE generating any personas. Rushed or skipped research grounding produces shallow panels.

这是决定小组讨论质量的关键步骤。必须先完成此步骤，再生成任何角色。仓促或跳过研究基础会导致讨论流于表面。

Terrain Mapping

领域梳理

Identify:

Core disciplines this topic spans (e.g., economics, ethics, computer science, public health)
Key tensions: technical vs. ethical, theory vs. practice, empirical vs. normative, short-term vs. long-term, individual vs. systemic, efficiency vs. equity
Intellectual traditions with substantive positions on this topic — not generic "perspectives" but actual schools of thought with methodological commitments (e.g., capabilities approach vs. revealed preference theory, not "some people think X and others think Y")
What is specifically contested: which evidence is disputed, which frameworks are in tension, which assumptions are not shared across traditions

明确以下内容：

核心学科：话题涉及的学科领域（如经济学、伦理学、计算机科学、公共卫生）
关键冲突：技术vs伦理、理论vs实践、实证vs规范、短期vs长期、个体vs系统、效率vs公平
学术流派：对该话题有明确立场的学术流派——不是泛泛的“视角”，而是有明确方法论的思想体系（如可行能力方法vs显示偏好理论，而非“有人支持X，有人支持Y”）
具体争议点：哪些证据存在争议、哪些分析框架相互冲突、哪些假设是不同流派不共有的

Research Grounding

研究基础

Use WebSearch to find 3-5 recent, relevant sources. Prioritize:

Academic papers (.edu, arxiv.org)
Substantive analyses from established publications
Real debates between named scholars
Meta-analyses or literature reviews that map the field

If WebSearch is unavailable or returns thin results, draw on training knowledge and flag this explicitly: "Based on training knowledge — not verified against current literature."

If the topic has a live academic debate, identify actual participants and positions. Real names, real works, real disagreements.

Citation integrity rules:

Cite specific works when confident: "As Sen argues in Development as Freedom (1999)..."
When uncertain about specifics, reference the tradition or framework: "drawing on the capabilities approach"
NEVER fabricate titles, authors, years, or journal names. If unsure, say "a study in this tradition found..." rather than inventing a citation

通过网络搜索找到3-5篇近期相关资料，优先选择：

学术论文（.edu、arxiv.org）
权威出版物的深度分析
知名学者之间的真实辩论
梳理领域现状的元分析或文献综述

若无法使用网络搜索或搜索结果不足，可基于训练知识展开，并明确标记：“基于训练知识——未结合最新文献验证。”

若话题存在活跃的学术辩论，需明确真实参与者及其立场。使用真实姓名、著作与分歧点。

引用诚信规则：

确定信息准确时，引用具体著作：“正如Sen在《Development as Freedom》（1999）中提出的...”
不确定具体细节时，引用流派或框架：“基于可行能力方法的观点...”
禁止编造书名、作者、年份或期刊名称。若不确定，可表述为“该流派的一项研究发现...”，而非虚构引用。

Outputs (Show Before Proceeding)

前置输出（开始前展示）

Present to the user:

Topic map: key tensions, disciplines involved, the core question being addressed
Research brief: key works found, active debates, real scholarly positions
Suggested panel composition (brief): the intellectual traditions that should be represented based on the tensions identified

By default, produce the complete panel in a single response (topic map through synthesis). Pause for user input only when the topic diagnostic flagged an issue (too broad, too narrow, settled science) or when the topic is ambiguous enough that reframing is likely. The panel should teach the user something they did not already know.

向用户展示以下内容：

话题图谱：关键冲突、涉及学科、核心待解决问题
研究简报：找到的关键著作、活跃辩论、真实学术立场
建议小组构成（简要）：基于已识别的冲突，应代表的学术流派

默认情况下，一次性输出完整的小组讨论内容（从话题图谱到成果合成）。仅当话题诊断标记问题（过于宽泛、过于狭窄、已定论科学问题）或话题模糊需要重构时，才暂停等待用户输入。小组讨论应能为用户带来新的认知。

3. Persona Generation

3. 角色生成

Build personas that maximally cover the tensions identified in the topic map. Every major tension should have at least one vocal advocate on each side.

构建能最大程度覆盖话题图谱中冲突的角色。每个主要冲突至少有一位明确的支持者和反对者。

Required Attributes Per Panelist

每位参与者的必填属性

For each panelist, specify:

Name and credentials: institutional affiliation, career stage
Domain expertise — specific, not generic. "Computational neuroscientist studying emergent properties in artificial neural networks" NOT "AI researcher." "Labor economist specializing in automation displacement in manufacturing" NOT "economist."
Intellectual tradition — operationalized: how does this tradition shape their reasoning? What counts as evidence for them? What counts as a good explanation? What are their methodological commitments?
Argumentative style: data-driven, theoretical, historical, pragmatic, dialectical, narrative
Known blind spots — specific: "tends to underweight distributional effects when analyzing aggregate productivity gains" NOT "has biases"

为每位参与者明确：

姓名与资质：机构 affiliation、职业阶段
领域专长——具体而非泛泛。例如“研究人工神经网络涌现特性的计算神经科学家”而非“AI研究者”；“专注制造业自动化替代的劳动经济学家”而非“经济学家”。
学术流派——可落地：该流派如何影响其推理逻辑？什么是其认可的证据？什么是合理的解释？其方法论承诺是什么？
论证风格：数据驱动型、理论型、历史型、实用主义型、辩证型、叙事型
已知盲区——具体：“分析总生产率提升时，往往低估分配效应”而非“存在偏见”

Diversity Requirements

多样性要求

Full requirements (4+ experts):

No two panelists from the same intellectual tradition
At least one contrarian — someone whose position will be genuinely uncomfortable for the room, not merely mildly skeptical
At least one bridge figure who connects two disciplines (e.g., a bioethicist bridges biology and philosophy; a computational linguist bridges CS and linguistics)
Mix of career stages: emeritus professor, mid-career, early-career researcher. Different career stages produce different risk tolerances and different relationships to established wisdom

Scaled for smaller panels:

2 experts: ensure distinct traditions; prefer at least one bridge figure when the topic spans multiple disciplines (not required for single-discipline topics)
3 experts: ensure distinct traditions, at least one contrarian or bridge figure, at least two different career stages

完整要求（4位及以上专家）：

任意两位参与者不得来自同一学术流派
至少有一位持相反意见者——其立场应让其他参与者感到不适，而非仅仅轻微质疑
至少有一位跨领域衔接者（如生物伦理学家衔接生物学与哲学；计算语言学家衔接计算机科学与语言学）
职业阶段多样化：荣誉退休教授、中年研究者、青年研究者。不同职业阶段的风险容忍度与对既有共识的态度不同

针对小组成员较少的调整：

2位专家：确保来自不同流派；若话题跨多学科，优先选择至少一位跨领域衔接者（单学科话题无需此要求）
3位专家：确保来自不同流派，至少有一位持相反意见者或跨领域衔接者，至少涵盖两种不同职业阶段

Anti-Clustering Check

避免同质化检查

If two panelists share the same intellectual tradition, methodology, AND likely conclusions on the core tensions — replace one. Panels with clusters produce the illusion of diversity without the substance.

Consult

./references/archetypes.md

if the panel requires personas from 2+ distinct domains or if the topic falls outside well-known fields. Adapt archetypes to the specific topic rather than copying them verbatim.

When the topic has active scholarly debates, model panelists on real researchers' published positions (not their personal lives). Use composites when needed: "a researcher in the tradition of Amartya Sen's capabilities approach" is more grounded than an invented persona with no intellectual anchor.

若两位参与者的学术流派、方法论及对核心冲突的结论均一致——替换其中一位。同质化的小组会制造多样性假象，却无实质内容。

若小组需覆盖2个及以上不同领域或话题属于陌生领域，可参考

./references/archetypes.md

。需根据具体话题调整角色原型，而非直接复制。

若话题存在活跃的学术辩论，可基于真实研究者的公开立场塑造角色（而非其个人生活）。必要时可使用复合角色：“遵循Amartya Sen可行能力方法的研究者”比无学术根基的虚构角色更具可信度。

Announcement

角色介绍

Announce panelists with full credentials at the start of the panel. Give the user a clear sense of who is in the room and why each voice was selected.

在小组讨论开始前，完整介绍所有参与者的资质。让用户清晰了解参与人员及其入选理由。

Quality Calibration Example

质量校准示例

Target this level of specificity and intellectual depth:

**Dr. Amara Osei** (Development Economics, Oxford — capabilities approach):
Your proposal to use GDP growth as the primary metric repeats the same error
Rostow made with modernization theory. Sen demonstrated in *Development as
Freedom* that capability deprivation persists in high-growth economies. The
question isn't whether AI increases output — it's whether it expands substantive
freedoms for the least advantaged.

*[Moderator]: Dr. Osei raises a fundamental measurement question. Dr. Chen,
how do you respond to the claim that GDP masks distributional effects?*

Every panelist must speak at this level — citing specific works, engaging specific claims, reasoning from their stated tradition.

目标达到以下具体程度与学术深度：

**Dr. Amara Osei**（牛津大学发展经济学——可行能力方法）：
你提出将GDP增长作为核心衡量指标的建议，重复了Rostow在现代化理论中犯过的错误。Sen在《Development as Freedom》中已证明，高增长经济体中仍存在可行能力剥夺问题。问题不在于AI是否能提升产出——而在于它是否能为最弱势群体拓展实质自由。

*[主持人]：Osei博士提出了一个根本性的衡量标准问题。Chen博士，你如何回应GDP掩盖分配效应这一观点？*

所有参与者的发言必须达到此水平——引用具体著作、回应具体主张、基于其所属流派推理。

4. Moderator Standing Orders

4. 主持人规则

These behaviors apply continuously throughout all discussion phases. Claude acts as the moderator.

这些规则适用于所有讨论阶段。Claude将担任主持人角色。

Persona Integrity (before each panelist speaks)

角色一致性（每位参与者发言前）

Before each panelist speaks, execute this internal reasoning pipeline (silent — do not display any of these steps):

Recall: What are this panelist's core commitments and what have they argued so far?
Analyze: What have other panelists actually claimed? Consider arguments by substance, not by who said them — this forces engagement with ideas, not social dynamics.
Evaluate: Which claims would this panelist's tradition challenge, and on what grounds?
Respond: Formulate a response grounded in this tradition's vocabulary and reasoning patterns. When citing specific works, only cite works mentioned in the research grounding (Section 2) or well-known foundational texts. For less certain references, use tradition-level attribution.

Each panelist's vocabulary, reasoning structure, and evidence standards must match their intellectual tradition. See

./references/archetypes.md

for domain-specific patterns. A pragmatist and a theorist must sound different because they think differently.

每位参与者发言前，需执行以下内部推理流程（无需展示）：

回忆：该参与者的核心承诺及此前的论点是什么？
分析：其他参与者实际提出了哪些主张？需关注论点实质，而非发言者——这能促使参与者针对观点互动，而非陷入社交互动。
评估：该参与者所属流派会质疑哪些主张？依据是什么？
回应：基于该流派的词汇与推理模式构建回应。引用具体著作时，仅引用研究基础（第2节）中提及的作品或知名基础文献。若对引用细节不确定，可基于流派层面归因。

每位参与者的词汇、推理结构与证据标准必须与其所属学术流派匹配。可参考

./references/archetypes.md

中的领域特定模式。实用主义者与理论家的发言风格必须不同，因为他们的思维方式不同。

Turn Management (continuous)

发言管理（全程）

Call on panelists by name
Allow direct responses between panelists — real panels are conversations, not sequential monologues
Enforce roughly balanced airtime across all panelists (guidelines, not hard limits)

2-expert panels: Standing orders adapt for structured dialogue:

Moderator intervenes at phase transitions only, not mid-exchange.
Convergence detection deferred to synthesis.
Devil's Advocate uses format-specific phase names, not "Challenge Round."
Output: omit
```
> **Moderator:**
```
within phases; moderator voice in Phase 0, between-phase summaries, and Synthesis only.

称呼参与者姓名
允许参与者直接回应彼此——真实的小组讨论是对话，而非依次独白
确保所有参与者的发言时间大致均衡（为指导原则，非硬性限制）

2位专家的调整： 规则调整为结构化对话：

主持人仅在阶段转换时介入，而非对话中途。
共识判断推迟至成果合成阶段。
反方角色使用格式特定的阶段名称，而非“挑战环节”。
输出：阶段内省略
```
> **主持人:**
```
；主持人仅在第0阶段、阶段间总结与成果合成环节发言。

Provocation Triggers (reactive, during any phase)

触发干预的场景（讨论中，被动触发）

(For 2-expert panels, these fire at phase transitions only — see Turn Management.)

Intervene when any of these occur:

Convergence: 2+ panelists agree without challenge. "Dr. X, you seem to be agreeing with Dr. Y, but your tradition of [Z] typically takes a different view on this. What am I missing?" If consensus is genuine (different well-grounded reasons), acknowledge it and pivot toward marginal disagreements — implementation details, second-order effects, boundary conditions.
Vagueness: a panelist makes an abstract claim without grounding. "Can you give a specific example or cite specific evidence?"
Comfort zone: the discussion stays safe and polite. "What does this position imply that most people would find unacceptable?"
Stagnation: the same arguments are being recycled without progress. Introduce a new angle, a real-world case, or advance to the next phase.

As moderator, do not favor the emerging consensus. If 3+ panelists converge on a conclusion, explicitly steelman the strongest absent counterposition from a real intellectual tradition before allowing synthesis.

（对于2位专家的小组，仅在阶段转换时触发——见发言管理规则。）

出现以下情况时需介入：

共识形成：2位及以上参与者无异议达成一致。例如：“X博士，你似乎同意Y博士的观点，但你所属的[Z]流派通常对此有不同看法。我是否遗漏了什么？”若共识是真实的（基于不同的合理理由），需认可共识，并转向边缘分歧——如实施细节、二阶效应、边界条件。
模糊表述：参与者提出抽象主张但未提供依据。例如：“能否给出具体示例或引用具体证据？”
舒适区讨论：讨论停留在安全、礼貌的层面。例如：“该立场隐含的、大多数人难以接受的结论是什么？”
讨论停滞：重复相同论点，无进展。引入新视角、真实案例或进入下一阶段。

作为主持人，不得偏向正在形成的共识。若3位及以上参与者达成结论，需明确提出来自真实学术流派的最强反对立场，再进入成果合成阶段。

Devil's Advocate Rotation (during challenge-focused phases)

反方角色轮换（聚焦挑战的阶段）

During the format's challenge phase (Deep Dive for roundtable, Direct Rebuttal for Oxford, Deconstruct for Socratic), rotate devil's advocate assignments among panelists. Each assigned panelist steel-mans the position they most disagree with. Prioritize panelists whose positions are furthest from the discussion's mainstream.

在格式对应的挑战阶段（roundtable的深度探究环节、oxford的直接反驳环节、socratic的解构环节），轮换反方角色。每位被选中的参与者需强化其最反对的立场。优先选择与讨论主流立场差异最大的参与者。

Uncomfortable Implications (at least once per panel, MANDATORY)

揭示不适结论（每轮讨论至少一次，强制要求）

(For 2-expert panels, ask at a phase transition — see Turn Management.)

At least once per panel, ask 2-3 panelists (scale with panel size):

"What is the strongest case against your own position?"
"What uncomfortable implication does your view have that you would rather not discuss?"

Do not let panelists deflect. Press for specifics.

（对于2位专家的小组，仅在阶段转换时提问——见发言管理规则。）

每轮讨论至少一次，向2-3位参与者提问（根据小组人数调整）：

“针对你自己的立场，最强的反对理由是什么？”
“你的观点隐含的、你不愿讨论的不适结论是什么？”

不得让参与者回避问题，需追问具体内容。

Between-Phase Summaries (between phases)

阶段间总结（阶段转换时）

Provide brief summaries between phases that name the disagreement precisely:

"So far, the key disagreement is between Dr. X (position A, grounded in [tradition]) and Dr. Y (position B, grounded in [tradition]). The crux seems to be [specific point of divergence]. Dr. Z has introduced a third axis — [brief description]."

在阶段转换时提供简要总结，明确指出分歧点：

“目前的核心分歧在于X博士（立场A，基于[流派]）与Y博士（立场B，基于[流派]）之间。关键分歧点是[具体分歧]。Z博士提出了第三个维度——[简要描述]。”

5. Discussion Phases

5. 讨论阶段

Load the chosen format's specific phase guide from

./references/formats.md

. The format guide's phase structure governs all phases between Framing and Synthesis. Phase 0 (Framing) and Synthesis are universal bookends. Adapt all output template headings to match the chosen format's phase names.

formats.md

cannot be loaded, inform the user: "The format reference file is missing — panel quality will be degraded. Reinstall the skill or provide the file at

./references/formats.md

." Proceed only if the user confirms, using roundtable defaults: opening positions (150-200 words each), 2-4 rounds of direct engagement, steel-man + self-critique.

从

./references/formats.md

加载所选格式的具体阶段指南。格式指南的阶段结构将指导 framing 与 synthesis 之间的所有阶段。第0阶段（Framing）与Synthesis是所有格式共有的首尾环节。需将所有输出模板标题调整为与所选格式的阶段名称一致。

若无法加载

formats.md

，需告知用户：“格式参考文件缺失——小组讨论质量将受影响。请重新安装该技能或在

./references/formats.md

路径下提供该文件。”仅在用户确认后继续，使用roundtable默认设置：开场立场（每人150-200词）、2-4轮直接互动、强化反对立场+自我批评。

Phase 0: Framing

第0阶段：Framing（开场设定）

The moderator introduces the topic:

Contextualize why this topic matters now
Frame what the audience should take away
Present each panelist with full credentials
State the core tension or question the panel will address

Keep framing concise. The value is in the discussion, not the introduction.

主持人介绍话题：

说明该话题当前的重要性背景
明确观众应从中获得的收获
完整介绍每位参与者的资质
提出小组将探讨的核心冲突或问题

开场设定需简洁。讨论的价值在于对话本身，而非介绍。

Format-Specific Phases

格式特定阶段

See

./references/formats.md

for phase names, structure, and word counts specific to each format. Use those phase names in the output — not generic Phase 1/2/3.

2-expert phase overrides: Oxford with 2 omits Floor Questions (Phase 4) — the moderator's probing role is unnecessary when both sides engage directly. See format-count notes in Section 1.

请参考

./references/formats.md

获取各格式的阶段名称、结构与字数要求。输出时需使用这些阶段名称——而非通用的第1/2/3阶段。

2位专家的阶段调整： 2位专家的牛津式辩论省略观众提问环节（第4阶段）——双方直接互动时，主持人的追问角色并非必需。见第1节的格式与人数说明。

Synthesis

Synthesis（成果合成）

See Section 6 for detailed synthesis instructions.

详见第6节的成果合成说明。

6. Synthesis

6. 成果合成

Synthesis is NOT a summary of what each person said. It is an intellectual product that could not have been produced by any single panelist alone.

成果合成不是对每位参与者发言的总结。它是单个专家无法独立产出的智力成果。

Required Synthesis Components

成果合成的必填内容

Identify the underlying axiom: what assumption explains WHY the panelists disagree? What prior does each side hold that the other does not? Often the deepest insight of a panel is discovering that the disagreement is not about evidence but about values, or not about values but about empirical assumptions. For 2-expert panels where convergence was deferred from discussion: is apparent agreement genuine (different traditions reaching the same conclusion) or model-prior-driven collapse?
State the emergent question: what NEW question emerged from the interaction that none of the panelists started with? If the panel generated no emergent questions, it was too shallow. Prefer questions that introduce dimensions or stakeholders absent from the original framing. If the panel's deepest insight is a refined version of the original question, state why the refinement matters — what new understanding does it encode? "How should we do X?" with no new understanding is reformulation, not emergence.
Identify resolution evidence: what specific experiment, study, or data would resolve the remaining tensions? What would move the debate forward? Be concrete: "A longitudinal study comparing X and Y populations on Z metric would adjudicate between Dr. A's prediction and Dr. B's prediction."
Map the positions structurally: not "A thinks X, B thinks Y" but "The fundamental axis of disagreement is [Z], with A and C on one side, B and D on the other, and E occupying an unusual middle position because of [specific methodological commitment that cuts across the main axis]."
Name the uncomfortable implications that surfaced during the discussion. Do not let them disappear into polite summary.
Key takeaways: 3-5 condensed bullets distilling the panel's most important insights and unresolved tensions.
Provide genuine further reading: specific works referenced during the panel, plus 2-3 additional works that speak to the tensions identified. Real works only — never fabricate titles, authors, or publication details.
Self-assess: did this panel produce genuine insight beyond what any single expert would have offered? If the discussion was surface-level, acknowledge this honestly and offer to run a deeper follow-up on a specific tension.

识别底层公理：什么假设解释了参与者的分歧？各方持有哪些对方不认可的前提？小组讨论最深刻的洞见往往是发现分歧并非源于证据，而是源于价值观；或并非源于价值观，而是源于实证假设。对于延迟共识判断的2位专家小组：表面的共识是真实的（不同流派得出相同结论），还是模型驱动的趋同？
提出新问题：对话中产生了哪些参与者最初未提出的新问题？若未产生新问题，则讨论过于浅显。优先选择引入原设定中未涵盖的维度或利益相关者的问题。若小组最深刻的洞见是原问题的优化版本，需说明优化的意义——它包含了哪些新认知？例如，仅将“我们应如何做X？”重构为更具体的问题，若未带来新认知，则不属于新问题，只是重新表述。
明确解决分歧的证据：哪些具体实验、研究或数据能解决剩余冲突？如何推动辩论进展？需具体：“针对X与Y群体的Z指标进行纵向研究，可判断A博士的预测与B博士的预测孰是孰非。”
结构化梳理立场：不是“A支持X，B支持Y”，而是“核心分歧轴是[Z]，A与C站在一侧，B与D站在另一侧，E因[具体方法论承诺]处于特殊的中间立场。”
记录讨论中揭示的不适结论：不得让这些结论在礼貌的总结中消失。
关键洞见：3-5条提炼后的要点，涵盖小组讨论最重要的洞见与未解决的冲突。
推荐阅读：讨论中引用的具体著作，再加2-3篇与已识别冲突相关的著作。仅使用真实著作——禁止编造书名、作者或出版细节。
自我评估：该小组讨论是否产出了单个专家无法独立提供的真实洞见？若讨论流于表面，需诚实承认，并提出针对特定冲突的深入讨论方案。

Visual Grammar

视觉规范

Maintain four visual voices throughout the panel output:

```
**Bold Name** (credentials):
```
= panelist speaking (normal text)
```
> **Moderator:**
```
= moderator interjection (blockquote)
```
*[italic brackets]*
```
= between-phase summaries, meta-commentary
```
### H3
```
= phase boundaries, separated by
```
---
```

小组讨论输出需保持四种视觉区分：

```
**加粗姓名** (资质):
```
= 参与者发言（普通文本）
```
> **主持人:**
```
= 主持人插话（块引用）
```
*[斜体括号]*
```
= 阶段间总结、元评论
```
### 三级标题
```
= 阶段分界，用
```
---
```
分隔

Output Format

输出格式

Structure the complete panel output as follows:

## Panel: [Topic]
**Format:** [format] | **Date:** [date] | **Experts:** [count]

### Panelist Roster
- **[Name]** — [credentials] *(tradition)*

### Phase 0: Framing …
### [Each format-specific phase as its own H3]
### Synthesis …
- **Axiom of disagreement:** ...
- **Emergent question:** ...
- **Resolution evidence:** ...
- **Position map:** ...
- **Uncomfortable implications:** ...
- **Key takeaways:** [3-5 bullets]
- **Further reading:** ...
- **Self-assessment:** ...

完整小组讨论的输出结构如下：

## 小组讨论：[话题]
**格式：** [format] | **日期：** [日期] | **专家人数：** [count]

### 参与者名单
- **[姓名]** — [资质] *(流派)*

### 第0阶段：Framing …
### [各格式特定阶段，使用对应阶段名称作为三级标题]
### Synthesis …
- **分歧的底层公理：** ...
- **新产生的问题：** ...
- **解决分歧的证据：** ...
- **立场结构梳理：** ...
- **揭示的不适结论：** ...
- **关键洞见：** [3-5条要点]
- **推荐阅读：** ...
- **自我评估：** ...

Output Length

输出长度

A full panel runs approximately 3000-4000 words total. Let the discussion breathe at natural length — do not compress interaction for brevity.

A condensed panel (~1000-1500 words) keeps: abbreviated framing, one round of sharpest exchanges, challenge highlights, and full synthesis. Cut: opening positions, redundant exchanges, moderator summaries. Use when the user requests "condensed."

完整的小组讨论约3000-4000词。让讨论自然展开——不要为了简洁压缩互动内容。

精简版小组讨论（约1000-1500词）包含：简化的开场设定、一轮最尖锐的互动、挑战环节亮点、完整的成果合成。需删减：开场立场、重复互动、主持人总结。当用户要求“精简版”时使用。

7. After the Panel

7. 讨论结束后

When responding to follow-ups, briefly re-ground by reviewing the panelist roster (name, tradition, argumentative style) before speaking in character. Personas drift after many turns without this re-grounding step.

If the user is making a practical decision, connect the synthesis to decision implications: "If you are deciding X, this panel suggests weighing [tension A] against [tension B]. Dr. Y's framework would prioritize..., while Dr. Z's would prioritize..."

After synthesis, generate 3-4 numbered follow-up options specific to this panel's content. Each must reference a specific tension, expert, or emergent question:

"Drill into [tension]: [Dr. X] and [Dr. Y] disagreed on [claim]. Explore further."
"Challenge [Dr. Z]: Press on [uncomfortable implication] — what does this require?"
"[Emergent question]: Reframe around the new question that surfaced."
"Decision lens: If deciding [related decision], hear each panelist's advice."

Never use generic options like "ask follow-up questions." Every option must be specific to this panel.

回复后续问题时，需先简要回顾参与者名单（姓名、流派、论证风格），再以角色身份发言。若不进行此步骤，多轮对话后角色会偏离设定。

若用户需做出实际决策，需将成果合成与决策关联：例如“若你正在做X决策，该小组讨论建议权衡[冲突A]与[冲突B]。Y博士的框架会优先考虑...，而Z博士的框架会优先考虑...”

成果合成后，需生成3-4条与本次讨论内容相关的后续选项。每条必须提及具体冲突、专家或新产生的问题：

"深入探讨[冲突]： [X博士]与[Y博士]在[主张]上存在分歧。进一步探索。"
"挑战[Z博士]： 追问[不适结论]——该结论要求我们做什么？"
"[新产生的问题]： 围绕讨论中出现的新问题重构话题。"
"决策视角： 若正在做[相关决策]，听取每位参与者的建议。"

不得使用“提出后续问题”这类通用选项。所有选项必须与本次讨论直接相关。

Reference File Index

参考文件索引

File	Read When
`references/formats.md`	Loading phase structure for the chosen discussion format
`references/archetypes.md`	Building personas spanning 2+ distinct domains or unfamiliar fields

文件	读取时机
`references/formats.md`	加载所选讨论格式的阶段结构时
`references/archetypes.md`	构建跨2个及以上领域的角色或陌生领域的角色时

Canonical Vocabulary

标准术语

Canonical Term	Meaning
panel	A simulated multi-expert discussion on a topic
expert / panelist	An AI-simulated domain specialist with defined tradition and credentials
format	The discussion structure: roundtable, oxford, or socratic
synthesis	The intellectual product produced after discussion phases
tradition	An intellectual school of thought with specific methodological commitments
moderator	Claude's role managing turn-taking, provocation, and phase transitions
terrain mapping	The pre-discussion analysis identifying disciplines, tensions, and traditions
convergence	When 2+ panelists agree — must be tested for model-prior collapse
persona integrity	Maintaining each panelist's distinct voice, reasoning, and evidence standards
phase	A discrete stage of the discussion governed by the chosen format

标准术语	含义
panel	针对某一话题的AI模拟多专家讨论
expert / panelist	具备明确流派与资质的AI模拟领域专家
format	讨论结构：roundtable、oxford或socratic
synthesis	讨论阶段结束后产出的智力成果
tradition	具备明确方法论承诺的学术流派
moderator	Claude担任的角色，负责管理发言顺序、触发干预与阶段转换
terrain mapping	讨论前的分析，明确涉及的学科、冲突与流派
convergence	2位及以上参与者达成一致——需测试是否为模型驱动的趋同
persona integrity	保持每位参与者的独特语气、推理方式与证据标准
phase	讨论的离散阶段，由所选格式决定

Critical Rules

核心规则

Non-negotiable constraints for every panel:

Research before personas. Always run topic analysis and research grounding first.
Never skip synthesis. It is the intellectual product that justifies the panel.
Citation integrity. Getting a citation wrong is worse than being vague (Section 2).
Disagreements must be specific. Cite the claim, cite the counter-evidence, explain why the traditions diverge. "I see it differently" is not a disagreement.
No straw men. Each position must be the strongest version of itself. If a panelist's argument is easy to defeat, the persona was poorly constructed.
Test convergence. Convergence may reflect model priors, not genuine agreement. Ask: would a real scholar from tradition X actually concede this point?
No monologues. If a panelist talks for more than 200 words without engagement, something has gone wrong.
Setup is not the product. Show topic map, research brief, and roster, then dive into the discussion.

每轮小组讨论必须遵守的非约束性规则：

先研究，再生成角色。必须先完成话题分析与研究基础，再生成角色。
绝不跳过成果合成。它是证明小组讨论价值的智力成果。
引用诚信。引用错误比表述模糊更严重（见第2节）。
分歧必须具体。需引用主张、反证，并解释流派分歧的原因。“我有不同看法”不构成分歧。
不得歪曲立场。每个立场必须是其最强版本。若参与者的论点容易被反驳，则角色构建存在问题。
测试共识。共识可能反映的是模型的先验，而非真实的一致。需提问：[X]流派的真实学者是否真的会让步？
禁止独白。若参与者连续发言超过200词且无人互动，则存在问题。
准备工作不是成果。展示话题图谱、研究简报与参与者名单后，直接进入讨论。