sdd-full 5.2.2 → 5.2.3

package/.claude/skills/multi-expert-discussion//345/256/214/346/225/264/346/236/266/346/236/204/345/233/276.md

@@ -0,0 +1,615 @@
+# Multi-Expert Discussion 技能 - 完整架构图
+
+本文档包含该技能的9个完整架构图，使用标准Mermaid语法，可直接渲染。
+
+---
+
+## 目录
+
+1. [整体架构图](#1-整体架构图)
+2. [模块拆分图](#2-模块拆分图)
+3. [主业务流程图](#3-主业务流程图)
+4. [异常&降级流程图](#4-异常降级流程图)
+5. [状态机图](#5-状态机图)
+6. [交互时序图](#6-交互时序图)
+7. [数据流图](#7-数据流图)
+8. [核心实体类图](#8-核心实体类图)
+9. [决策规则树图](#9-决策规则树图)
+
+---
+
+## 1. 整体架构图
+
+```mermaid
+graph TB
+    subgraph 接入层
+        A1[用户指令入口<br/>/multi-expert-discussion]
+        A2[静默模式开关]
+        A3[自定义流程解析器]
+    end
+
+    subgraph 调度层
+        B1[问题类型路由器<br/>V5.1五类型映射]
+        B2[YAML工作流引擎]
+        B3[阶段调度器]
+    end
+
+    subgraph 核心逻辑层
+        C1[问题定义与专家招募]
+        C2[圆桌研讨与交叉质询<br/>3A]
+        C3[分支管理Fork/Merge]
+        C4[红队攻击5A]
+        C5[二阶效应推演5B]
+        C6[归因分析中心5C]
+        C7[行动排序5D]
+        C8[熵减行动收束7]
+    end
+
+    subgraph 工具调用层
+        D1[搜索专家<br/>外部信息检索]
+        D2[原子原语库<br/>ask_experts/audit_round等]
+        D3[健康监控<br/>心跳检测/超时/熔断]
+    end
+
+    subgraph 数据持久层
+        E1[阶段备忘录S1]
+        E2[决策链S2]
+        E3[上下文快照S3]
+        E4[共识仓库S7]
+        E5[专家状态S5]
+    end
+
+    subgraph 外部依赖层
+        F1[Web检索API]
+        F2[代码库搜索]
+        F3[Context7文档库]
+    end
+
+    A1 --> B1
+    A2 --> B3
+    A3 --> B2
+    
+    B1 --> C1
+    B2 --> C2
+    B3 --> C3
+    
+    C1 --> D1
+    C2 --> D2
+    C3 --> D3
+    C4 --> D2
+    C5 --> D2
+    C6 --> D2
+    C7 --> D2
+    
+    C1 --> E1
+    C2 --> E2
+    C3 --> E3
+    C4 --> E1
+    C5 --> E1
+    C6 --> E1
+    C7 --> E1
+    C8 --> E2
+    
+    D1 --> F1
+    D1 --> F2
+    D1 --> F3
+```
+
+---
+
+## 2. 模块拆分图
+
+```mermaid
+graph LR
+    Root[Multi-Expert<br/>Discussion 主模块]
+
+    Root --> Router[问题路由模块<br/>problem-routing]
+    Root --> Engine[流程引擎模块<br/>workflow-engine]
+    Root --> Roles[角色管理模块<br/>roles]
+    Root --> Scoring[评分体系模块<br/>scoring]
+    Root --> Mechanisms[机制模块<br/>mechanisms]
+    Root --> RedTeam[红队攻击模块<br/>red-team]
+    Root --> CrossExam[交叉质询模块<br/>cross-examination]
+    Root --> Attribution[归因分析模块<br/>attribution-analysis]
+    Root --> Effects[二阶效应模块<br/>second-order-effects]
+    Root --> Action[行动排序模块<br/>action-prioritization]
+    Root --> Entropy[熵增熵减模块<br/>entropy-framework]
+    Root --> Templates[报告模板模块<br/>templates]
+
+    Roles --> Host[主持人]
+    Roles --> Auditor[审查AI<br/>元审查AI]
+    Roles --> Experts[专家角色<br/>动态生成]
+    Roles --> Devil[魔鬼代言人]
+    Roles --> Red[红队攻击者]
+    Roles --> Inspector[认知监察官]
+    Roles --> Power[权力意志专家]
+    Roles --> Search[搜索专家]
+
+    Mechanisms --> Timeout[超时策略]
+    Mechanisms --> Deadlock[死锁熔断]
+    Mechanisms --> Echo[回音室检测]
+    Mechanisms --> Appeal[上诉裁决]
+    Mechanisms --> Snapshot[快照机制]
+    Mechanisms --> Fork[分叉管理]
+
+    Router --> Convergent[收敛型路由]
+    Router --> Divergent[发散型路由]
+    Router --> Exploration[探索型路由]
+    Router --> V5[V5.1五类型映射]
+
+    Action --> Narrative[叙事先行评估]
+    Action --> Quant[三维量化排序<br/>效果×紧迫度×难度]
+
+    Entropy --> Metrics[元公理层]
+    Entropy --> Commandments[四条戒律]
+    Entropy --> Closure[熵减行动收束]
+```
+
+---
+
+## 3. 主业务流程图
+
+```mermaid
+graph TD
+    Start[用户触发指令<br/>/multi-expert-discussion] --> Parse[参数解析]
+    Parse --> Check{静默模式?}
+    Check -->|是| Silent[静默模式启动<br/>隐藏中间过程]
+    Check -->|否| Normal[正常模式]
+    
+    Normal --> Router[问题类型路由<br/>V5.1五类型判定]
+    Silent --> Router
+    
+    Router --> Load[加载YAML工作流]
+    Load --> Stage0{收敛型?}
+    Stage0 -->|是| Stage0Work[阶段0: 分层前置<br/>判定层→执行层A/B]
+    Stage0 -->|否| Stage1
+    
+    Stage0Work --> Stage1[阶段1: 问题定义<br/>专家招募]
+    Stage1 --> Bias[偏见声明<br/>+利益申报+预设解剖]
+    Bias --> Stage2[阶段2: 初始立场<br/>评分与预标注]
+    Stage2 --> Stage3[阶段3: 圆桌研讨]
+    
+    Stage3 --> Stage3Loop{每轮研讨}
+    Stage3Loop --> ExpertSpeak[专家发言<br/>评分+置信度]
+    ExpertSpeak --> Response[其他专家响应]
+    Response --> Search{需要搜索?}
+    Search -->|是| SearchExec[搜索子阶段<br/>外部验证]
+    Search -->|否| Audit
+    SearchExec --> Audit
+    
+    Audit[审查AI审核<br/>置信度+回音室+上诉] --> Memo[生成阶段备忘录]
+    Memo --> Health[健康检测]
+    
+    Health --> Deadlock{死锁?}
+    Deadlock -->|是| Devil[引入魔鬼代言人]
+    Deadlock -->|否| CheckFork
+    
+    Devil --> CheckFork{分叉条件?}
+    CheckFork -->|是| Stage4[阶段4: Fork<br/>多分支并行探索]
+    CheckFork -->|否| CheckExit{退出条件?}
+    
+    Stage4 --> Stage5[阶段5: Merge<br/>分支对比融合]
+    
+    CheckExit -->|否| Stage3Loop
+    CheckExit -->|是| Stage5
+    
+    Stage5 --> Gray{灰度导航?}
+    Gray -->|是| Stage5A[阶段5A: 红队攻击<br/>3轮+答辩]
+    Gray -->|否| CheckDepth
+    
+    Stage5A --> Stage5B[阶段5B: 二阶效应推演]
+    Stage5B --> CheckDepth
+    
+    CheckDepth{深度分析?} -->|是| Stage5C[阶段5C: 归因分析<br/>五步链路]
+    CheckDepth -->|否| CheckAction
+    
+    Stage5C --> CheckAction
+    CheckAction{决策类?} -->|是| Stage5D[阶段5D: 行动排序<br/>叙事先行+三维量化]
+    CheckAction -->|否| Stage7
+    
+    Stage5D --> Stage7[阶段7: 熵减行动收束]
+    Stage7 --> Report[生成最终报告]
+    Report --> CheckReopen{二次研讨?}
+    CheckReopen -->|是| Stage1
+    CheckReopen -->|否| End[输出回复]
+```
+
+---
+
+## 4. 异常&降级流程图
+
+```mermaid
+graph TD
+    Start[流程执行] --> Check1{参数合法?}
+    Check1 -->|否| Err1[参数非法<br/>返回400+引导澄清]
+    
+    Check1 -->|是| Exec[执行当前阶段]
+    Exec --> Check2{超时?}
+    Check2 -->|是| Err2[超时处理<br/>标记弃权/降级置信度]
+    
+    Exec --> Check3{调用失败?}
+    Check3 -->|是| Err3[重试策略<br/>最多3次+指数退避]
+    Err3 --> CheckRetry{重试成功?}
+    CheckRetry -->|是| Resume[恢复执行]
+    CheckRetry -->|否| Err32[兜底响应<br/>降级输出]
+    
+    Exec --> Check4{权限不足?}
+    Check4 -->|是| Err4[权限拒绝<br/>返回403]
+    
+    Exec --> Check5{无匹配数据?}
+    Check5 -->|是| Err5[无数据提示<br/>引导用户补充]
+    
+    Exec --> Check6{上下文溢出?}
+    Check6 -->|是| Err6[上下文压缩<br/>旧数据摘要化]
+    
+    Exec --> Check7{死锁触发?}
+    Check7 -->|是| Err7[死锁熔断<br/>魔鬼代言人→分歧终止报告]
+    
+    Err2 --> Resume
+    Err6 --> Resume
+    
+    Resume --> Next[进入下一阶段]
+    
+    Err1 --> End
+    Err32 --> End
+    Err4 --> End
+    Err5 --> End
+    Err7 --> End
+```
+
+---
+
+## 5. 状态机图
+
+```mermaid
+stateDiagram-v2
+    [*] --> 初始化: 用户触发指令
+    
+    初始化 --> 问题定义: 加载路由
+    问题定义 --> 专家招募: 边界确认
+    专家招募 --> 初始立场: 偏见声明完成
+    
+    初始立场 --> 圆桌研讨: 准备就绪
+    圆桌研讨 --> 交叉质询: 高置信度互斥
+    圆桌研讨 --> 等待输入: 需要用户确认
+    圆桌研讨 --> 分歧讨论: 出现争议
+    分歧讨论 --> 圆桌研讨: 争议解决
+    等待输入 --> 圆桌研讨: 用户输入完成
+    
+    交叉质询 --> 圆桌研讨: 裁决完成
+    圆桌研讨 --> 分支分叉: 满足分叉条件
+    分支分叉 --> 多分支并行: 分支创建成功
+    
+    多分支并行 --> 分支合并: 所有分支结束
+    分支合并 --> 红队攻击: 灰度导航触发
+    红队攻击 --> 二阶效应推演: 攻击完成
+    二阶效应推演 --> 归因分析: 深度分析触发
+    归因分析 --> 行动排序: 决策类触发
+    行动排序 --> 熵减收束: 排序完成
+    
+    熵减收束 --> 报告生成: 收束完成
+    报告生成 --> [*]: 正常结束
+    
+    圆桌研讨 --> 异常终止: 连续3轮无进展
+    异常终止 --> [*]: 分歧终止报告
+    
+    note right of 圆桌研讨
+        每轮包含:
+        发言→响应→搜索→
+        审核→备忘录→健康检测
+    end note
+```
+
+---
+
+## 6. 交互时序图
+
+```mermaid
+sequenceDiagram
+    participant User as 用户
+    participant Fac as 主持人
+    participant Router as 问题路由器
+    participant Experts as 专家团队
+    participant Search as 搜索专家
+    participant Auditor as 审查AI
+    participant Inspector as 认知监察官
+    participant Red as 红队攻击者
+    participant Fork as 分支管理AI
+    participant Meta as 元审查AI
+    
+    User ->> Fac: /multi-expert-discussion +问题描述
+    Fac ->> Router: 路由请求
+    Router -->> Fac: V5.1类型判定结果
+    Fac ->> Experts: 招募专家团队
+    
+    Experts -->> Fac: 偏见声明+利益申报
+    Fac ->> Experts: 开始初始立场发言
+    Experts -->> Fac: 观点+评分+置信度
+    Fac ->> Auditor: 首轮预标注
+    Auditor -->> Fac: 置信度审核结果
+    
+    loop 每轮研讨
+        Experts ->> Experts: 发言→响应→反驳
+        opt 需要搜索
+            Experts ->> Search: 外部验证请求
+            Search -->> Experts: 检索结果
+        end
+        Experts -->> Fac: 每轮成果
+        Fac ->> Auditor: 本轮审核
+        Auditor -->> Fac: 备忘录
+        Inspector -->> Fac: 监察官审校备注
+    end
+    
+    opt 触发分叉
+        Fac ->> Fork: 创建分支
+        Fork -->> Fac: 分支ID
+        Fork ->> Fork: 分支并行执行
+        Fork -->> Meta: 分支结果汇总
+    end
+    
+    opt 灰度导航
+        Fac ->> Red: 启动红队攻击
+        loop 3轮攻击
+            Red ->> Experts: 攻击维度挑战
+            Experts -->> Red: 答辩
+        end
+        Red -->> Fac: 红队攻击报告
+    end
+    
+    Fac ->> Meta: 最终方案合成
+    Meta -->> Fac: 最终报告
+    Fac ->> Fac: 熵减行动收束
+    Fac -->> User: 完整回复+导航目录
+```
+
+---
+
+## 7. 数据流图
+
+```mermaid
+graph LR
+    subgraph 输入层
+        I1[用户原始输入<br/>问题描述]
+        I2[静默模式标志]
+        I3[自定义流程声明]
+    end
+    
+    subgraph 解析提取层
+        P1[问题特征提取器]
+        P2[路由参数解析器]
+        P3[YAML流程解析器]
+    end
+    
+    subgraph 计算层
+        C1[问题类型判定]
+        C2[专家团队构建]
+        C3[置信度评分计算]
+        C4[死锁指标计算]
+        C5[红队攻击评级]
+        C6[行动排序计算<br/>效果×紧迫度×(6-难度)]
+    end
+    
+    subgraph 存储层
+        M1[阶段备忘录S1]
+        M2[决策链S2]
+        M3[上下文快照S3]
+        M4[共识仓库S7]
+        M5[专家状态S5]
+        M6[偏见声明S6]
+    end
+    
+    subgraph 工具层
+        T1[外部检索<br/>Web/代码库/Context7]
+        T2[原子原语调用<br/>ask_experts等]
+    end
+    
+    subgraph 合成输出层
+        O1[最终报告生成]
+        O2[熵减行动声明]
+        O3[结构化导航目录]
+    end
+    
+    I1 --> P1
+    I2 --> P2
+    I3 --> P3
+    
+    P1 --> C1
+    P2 --> C1
+    P3 --> C2
+    
+    C1 --> M3
+    C2 --> M6
+    C3 --> M1
+    C3 --> M5
+    C4 --> M1
+    C5 --> M1
+    C6 --> M1
+    
+    C1 --> T2
+    C2 --> T2
+    Experts ..>|搜索请求| T1
+    T1 ..>|检索结果| C3
+    
+    M1 --> O1
+    M2 --> O1
+    M4 --> O1
+    M5 --> O1
+    
+    O1 --> O2
+    O2 --> O3
+```
+
+---
+
+## 8. 核心实体类图
+
+```mermaid
+classDiagram
+    class MultiExpertDiscussion {
+        +String problem
+        +String mode(silent/normal)
+        +String type(convergent/divergent/exploration/gray/innovation)
+        +start()
+        +pause()
+        +resume()
+    }
+    
+    class Facilitator {
+        +String name
+        +List~Stage~ stages
+        +List~Expert~ experts
+        +clarifyProblem()
+        +recruitExperts()
+        +scheduleStage()
+        +detectDeadlock()
+    }
+    
+    class Expert {
+        <<Abstract>>
+        +String expertId
+        +String role
+        +Float weight
+        +speak(stage)
+        +respond(argument)
+    }
+    
+    class DynamicExpert {
+        +String domain
+        +BiasDeclaration bias
+        +InterestDeclaration interest
+        +PresuppositionAnatomy anatomy
+    }
+    
+    class RedTeamAttacker {
+        +List~AttackDimension~ dimensions
+        +attack(solution)
+    }
+    
+    class CognitiveInspector {
+        +List~Warning~ warnings
+        +checkBias()
+        +checkEchoChamber()
+        +checkFourCommandments()
+    }
+    
+    class Auditor {
+        +auditOpinion(opinion)
+        +resolveAppeal(appeal)
+        +detectEchoChamber()
+    }
+    
+    class MetaAuditor {
+        +mergeBranches()
+        +compareSolutions()
+    }
+    
+    class StageMemo {
+        +int round
+        +String branchId
+        +List~ExpertOpinion~ opinions
+        +List~ConsensusItem~ consensus
+        +List~DisagreementItem~ disagreements
+    }
+    
+    class ConsensusItem {
+        +String id
+        +String content
+        +Float score
+        +List~String~ agreedBy
+        +boolean immutable
+    }
+    
+    class DecisionChain {
+        +String version
+        +List~StageMemo~ layers
+        +List~ForkPoint~ forkPoints
+    }
+    
+    class Workflow {
+        +String name
+        +String type
+        +List~WorkflowStage~ stages
+        +execute()
+    }
+    
+    MultiExpertDiscussion "1" --> "1" Facilitator
+    MultiExpertDiscussion "1" --> "1" Workflow
+    Facilitator "1" --> "1" Auditor
+    Facilitator "1" --> "1" CognitiveInspector
+    Facilitator "1" --> "*" Expert
+    Expert <|-- DynamicExpert
+    Expert <|-- RedTeamAttacker
+    Auditor <|-- MetaAuditor
+    StageMemo "1" --> "*" ConsensusItem
+    DecisionChain "1" --> "*" StageMemo
+```
+
+---
+
+## 9. 决策规则树图
+
+```mermaid
+graph TD
+    Root[收到用户请求] --> A{是否触发技能?}
+    A -->|否| B[不使用该技能<br/>默认响应]
+    A -->|是| C{问题类型判定<br/>V5.1五类型}
+    
+    C --> D[事实查询]
+    C --> E[步骤流程]
+    C --> F[深度分析]
+    C --> G[灰度导航]
+    C --> H[创新探索]
+    
+    D --> D1{轻量探索?}
+    D1 -->|是| D2[单专家快速响应]
+    D1 -->|否| D3[直接搜索]
+    
+    E --> E1{自信源自检?}
+    E1 -->|是| E2[启动自信源自检]
+    E1 -->|否| E3[直接执行步骤流程]
+    E3 --> E4{决策类?}
+    E4 -->|是| E5[触发行动排序5D]
+    E4 -->|否| E6[正常执行]
+    
+    F --> F1{触发归因分析?}
+    F1 -->|是| F2[执行五步归因链路<br/>第一性→横向→纵向→空间→时间]
+    F1 -->|否| F3[深度优先分析]
+    
+    G --> G1{全层级开启}
+    G1 --> G2[红队攻击5A<br/>3轮+答辩]
+    G2 --> G3[二阶效应推演5B<br/>因果链延长→影响域→时间]
+    G3 --> G4[五维校准<br/>事实/价值/时空/系统/实践]
+    G4 --> G5[行动排序5D]
+    G5 --> G6[熵减行动收束7<br/>强制+日常语言表述]
+    
+    H --> H1{早期过度批判?}
+    H1 -->|是| H2[延迟评判机制启动]
+    H1 -->|否| H3[开放探索]
+    H3 --> H4{可能性探索?}
+    H4 -->|是| H5[触发可能性探索层]
+    
+    B --> End
+    D2 --> End
+    D3 --> End
+    E5 --> End
+    E6 --> End
+    F2 --> End
+    F3 --> End
+    G6 --> End
+    H5 --> End
+```
+
+---
+
+## 使用说明
+
+所有图表均为标准Mermaid语法，你可以：
+
+1. 将图表代码复制到任何支持Mermaid的编辑器中渲染
+2. 使用在线工具如 [Mermaid Live Editor](https://mermaid.live/) 查看
+3. 在支持Mermaid的Markdown查看器中直接渲染
+
+## 文件参考
+
+- [SKILL.md](file:///Users/xutao/Documents/trae_projects/sdd-full-npm/.claude/skills/multi-expert-discussion/SKILL.md) - 主技能文档
+- [万能架构图提示词模板.md](file:///Users/xutao/Documents/trae_projects/sdd-full-npm/.claude/skills/multi-expert-discussion/万能架构图提示词模板.md) - 提示词模板

package/bin.js

@@ -6,7 +6,7 @@ const fs = require('fs');
 const path = require('path');
 
 const SDD = {
-  version: '5.2.2',
+  version: '5.2.3',
   name: 'sdd-full',
   description: '完整的软件设计开发技能包'
 };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sdd-full",
-  "version": "5.2.2",
+  "version": "5.2.3",
   "description": "SDD Full - 完整的软件设计开发技能包",
   "main": "index.js",
   "bin": "./bin.js",