@yeaft/webchat-agent 0.1.248 → 0.1.249

package/conductor/index.js

@@ -1,65 +0,0 @@
-/**
- * Conductor — Module Entry
- *
- * Exports three core subsystems:
- * 1. Personas — 18 persona definitions (4 scenarios)
- * 2. Specialties — specialty behavior definitions
- * 3. CLAUDE.md Generator — dynamic 3-section generation
- * 4. Actor — Claude instance create/manage/release
- * 5. Worktree — per-task worktree management
- */
-
-// ---- Personas ----
-export {
-  getPersonaById,
-  getPersonasByScenario,
-  getPersonaPoolForOrchestrator,
-  getAllPersonas,
-  getScenarios,
-  personasByScenario
-} from './personas/index.js';
-
-// ---- Specialties ----
-export {
-  getSpecialty,
-  getSpecialtiesByScenario,
-  getThinkingMode,
-  isDivergent,
-  getPersonalizedBehavior,
-  getOutputFormat,
-  formatToolRules,
-  personaSupportsSpecialty,
-  allSpecialties,
-  scenarioSpecialties
-} from './specialties/index.js';
-
-// ---- CLAUDE.md Generator ----
-export {
-  generateActorCLAUDEmd,
-  generateActorInstanceId,
-  generateActorDirName
-} from './claudemd-generator.js';
-
-// ---- Actor Management ----
-export {
-  createActor,
-  sendToActor,
-  releaseActor,
-  releaseAllActors,
-  getActor,
-  getActiveActors,
-  getAllActors,
-  getActorPublicState,
-  getActorStates,
-  classifyActorError,
-  actorRegistry
-} from './actor.js';
-
-// ---- Worktree Management ----
-export {
-  createTaskWorktree,
-  cleanupTaskWorktree,
-  cleanupAllConductorWorktrees,
-  getTaskWorktreePath,
-  getActorCwd
-} from './worktree.js';

package/conductor/orchestrator-flows.js

@@ -1,498 +0,0 @@
-/**
- * Conductor Orchestrator — 场景特化流程定义
- *
- * 4 大场景 (Dev / Writing / Trading / Video) 各自定义：
- * - TRIAGE 判定逻辑
- * - 每个 Phase 的 Actor 编排策略
- * - Quality Gate 规则
- * - Acceptance 验收方式
- *
- * 这些流程被 Orchestrator 状态机消费，驱动具体的 Actor 调度。
- */
-
-import { Phase, ThinkingMode } from './task.js';
-
-// =====================================================================
-// Specialty 思维模式映射
-// =====================================================================
-
-const DIVERGENT_SPECIALTIES = new Set([
-  'planning', 'discussion', 'pacing', 'scripting', 'macro-analysis'
-]);
-
-/**
- * 根据 specialty 返回思维模式
- */
-export function getThinkingMode(specialty) {
-  return DIVERGENT_SPECIALTIES.has(specialty)
-    ? ThinkingMode.DIVERGENT
-    : ThinkingMode.CONVERGENT;
-}
-
-// =====================================================================
-// Dev 场景流程
-// =====================================================================
-
-const devFlow = {
-  scenario: 'dev',
-
-  /**
-   * TRIAGE: 判定任务复杂度
-   * 返回 'trivial' | 'small' | 'complex'
-   */
-  triageRules: {
-    trivial: [
-      '任务描述少于 2 句话',
-      '明确指向单个文件/单个函数的修改',
-      '不涉及架构变化，不涉及多文件联动'
-    ],
-    small: [
-      '任务目标清晰，不需要讨论方向',
-      '可能涉及多个文件，但变更模式明确',
-      '不需要先出设计方案'
-    ],
-    complex: [
-      '任务描述模糊或有多种理解方式',
-      '涉及架构决策、技术选型、用户体验取舍',
-      '需要多角色的视角碰撞才能确定方向',
-      '涉及 UI/前端时通常需要先 design'
-    ]
-  },
-
-  /**
-   * DISCUSSION 编排：创建哪些 actor，如何协作
-   */
-  discussion: {
-    defaultActors: [
-      { personaId: 'rams', specialty: 'design', condition: 'ui_involved' },
-      { personaId: 'jobs', specialty: 'discussion' },
-      { personaId: 'torvalds', specialty: 'discussion' }
-    ],
-    outputTarget: 'memory'  // 讨论结果写入 memory.md
-  },
-
-  /**
-   * PLANNING 编排
-   */
-  planning: {
-    actors: [
-      { personaId: 'jobs', specialty: 'planning', role: 'product' },
-      { personaId: 'torvalds', specialty: 'planning', role: 'lead' }
-    ],
-    lead: 'torvalds',       // Lead 负责合并最终 plan
-    minPlanners: 2,
-    outputTarget: 'plan'    // 产出写入 plan.json
-  },
-
-  /**
-   * EXECUTION 编排：按 plan step 创建 coding actor
-   */
-  execution: {
-    primaryCoder: 'torvalds',
-    secondaryCoder: 'beck',
-    supportsParallel: true   // 可并行步骤同时创建多个 actor
-  },
-
-  /**
-   * QUALITY GATE 编排
-   */
-  qualityGate: {
-    reviewer: { personaId: 'martin', specialty: 'review' },
-    tester: { personaId: 'beck', specialty: 'testing' },
-    parallel: true,          // review 和 testing 可并行
-    passThreshold: 9,        // 10分制, ≥9 通过
-    maxRetries: 3,           // 同一步骤 fail ≥3 次 → 升级 discussion（设计文档 6.2.2）
-    codingRequiresReview: true,
-    codingRequiresTesting: true,
-    designRequiresReview: false,
-    docRequiresReview: false
-  },
-
-  /**
-   * ACCEPTANCE 编排
-   */
-  acceptance: {
-    verifier: { personaId: 'jobs', specialty: 'planning' },
-    focusAreas: ['用户体验达标', 'successCriteria 逐条检查']
-  }
-};
-
-// =====================================================================
-// Writing 场景流程
-// =====================================================================
-
-const writingFlow = {
-  scenario: 'writing',
-
-  triageRules: {
-    trivial: [
-      '修改单章内容',
-      '调整已有文本的措辞/表达'
-    ],
-    small: [
-      '续写已有章节（基于已有大纲）',
-      '为已有卷添加章节'
-    ],
-    complex: [
-      '新卷/新书（需要完整的大纲设计）',
-      '涉及人物弧线调整或世界观修改'
-    ]
-  },
-
-  discussion: {
-    defaultActors: [
-      { personaId: 'maoni', specialty: 'discussion' },
-      { personaId: 'zhouzi', specialty: 'discussion' },
-      { personaId: 'jinyong', specialty: 'discussion' }
-    ],
-    focusAreas: {
-      maoni: '故事结构、伏笔布局、人物弧线',
-      zhouzi: '节奏体感、笑点密度、读者情绪过山车',
-      jinyong: '人物厚度、武侠精神内核、大格局叙事'
-    },
-    outputTarget: 'memory'
-  },
-
-  planning: {
-    actors: [
-      { personaId: 'maoni', specialty: 'planning', role: 'lead' },
-      { personaId: 'zhouzi', specialty: 'planning', role: 'execution-review' },
-      { personaId: 'zhouzi', specialty: 'pacing', role: 'pacing' }
-    ],
-    lead: 'maoni',
-    minPlanners: 2,
-    outputTarget: 'plan'
-  },
-
-  execution: {
-    primaryWriter: 'zhouzi',
-    supportsParallel: false   // 章节串行写作
-  },
-
-  qualityGate: {
-    reviewer: { personaId: 'maboyong', specialty: 'editing' },
-    tester: null,             // Writing 无 tester
-    parallel: false,
-    checkAreas: ['设定一致性', '时间线', '人物逻辑', '爽点落地', '文字质量'],
-    maxRetries: 3
-  },
-
-  acceptance: {
-    verifier: { personaId: 'maoni', specialty: 'planning' },
-    focusAreas: ['伏笔按计划埋设/回收', '人物弧线推进', '卷级目标达成']
-  }
-};
-
-// =====================================================================
-// Trading 场景流程
-// =====================================================================
-
-const tradingFlow = {
-  scenario: 'trading',
-
-  triageRules: {
-    trivial: [
-      '紧急交易指令（明确的品种/方向/仓位）',
-      '已有策略的简单执行'
-    ],
-    small: [
-      '已有策略调整（参数微调、止损调整）',
-      '单一品种分析'
-    ],
-    complex: [
-      '新策略制定',
-      '跨资产组合分析',
-      '市场拐点判断'
-    ]
-  },
-
-  /**
-   * Trading 特有：发散分析阶段（替代标准 discussion）
-   * 宏观分析和技术分析并行，然后进入 planning
-   */
-  discussion: {
-    defaultActors: [
-      { personaId: 'dalio', specialty: 'macro-analysis' },
-      { personaId: 'livermore', specialty: 'technical-analysis' }
-    ],
-    parallel: true,
-    outputTarget: 'memory'
-  },
-
-  planning: {
-    actors: [
-      { personaId: 'soros', specialty: 'planning', role: 'lead' }
-    ],
-    lead: 'soros',
-    minPlanners: 1,
-    outputTarget: 'plan'
-  },
-
-  /**
-   * Trading 特有：对抗性讨论（PLANNING 之后自动进入）
-   * 设计文档 6.2.5: PLANNING → DISCUSSION（对抗性审查）→ QG → EXECUTION
-   */
-  postPlanningDiscussion: {
-    actors: [
-      { personaId: 'soros', specialty: 'discussion' },
-      { personaId: 'taleb', specialty: 'discussion' }
-    ],
-    purpose: '策略反脆弱压力测试',
-    challengePoints: [
-      '假设完全相反会怎样？',
-      '信念越强越需要警惕确认偏误',
-      '加仓条件是否是在沉船上往里跳？'
-    ],
-    outputTarget: 'memory'
-  },
-
-  /**
-   * Trading 特有：QG 在 EXECUTION 之前（风控审查通过才能执行）
-   * 设计文档 6.2.5: QG pass → 执行 / fail → 打回修改策略
-   */
-  qualityGateBeforeExecution: true,
-
-  qualityGate: {
-    reviewer: { personaId: 'taleb', specialty: 'risk-review' },
-    tester: null,
-    parallel: false,
-    riskThresholds: {
-      maxPositionSize: 0.02,      // 单仓 ≤ 2%
-      maxTotalExposure: 0.10,     // 总敞口 ≤ 10%
-      minConvexity: 3,            // 凸性 ≥ 3:1
-      requireTailHedge: true
-    },
-    maxRetries: 3
-  },
-
-  execution: {
-    executor: 'jones',
-    executorSpecialty: 'execution',
-    supportsParallel: false
-  },
-
-  acceptance: {
-    verifier: { personaId: 'soros', specialty: 'planning' },
-    focusAreas: ['假设是否仍成立', '信念强度变化', '风险回报比']
-  }
-};
-
-// =====================================================================
-// Video 场景流程
-// =====================================================================
-
-const videoFlow = {
-  scenario: 'video',
-
-  triageRules: {
-    trivial: [
-      '调整单段 prompt 的参数或措辞',
-      '修改单个分镜细节'
-    ],
-    small: [
-      '修改已有脚本/分镜',
-      '局部调整一致性锚点'
-    ],
-    complex: [
-      '新视频项目（需要完整流程）',
-      '涉及风格方向变更'
-    ]
-  },
-
-  /**
-   * Video 使用 planning 阶段确定整体方案，不走标准 discussion
-   */
-  discussion: {
-    defaultActors: [
-      { personaId: 'jia', specialty: 'discussion' },
-      { personaId: 'shi', specialty: 'discussion' }
-    ],
-    outputTarget: 'memory'
-  },
-
-  planning: {
-    actors: [
-      { personaId: 'jia', specialty: 'planning', role: 'lead' }
-    ],
-    lead: 'jia',
-    minPlanners: 1,
-    outputTarget: 'plan'
-  },
-
-  /**
-   * Video EXECUTION: 严格串行的三阶段
-   *
-   * Stage A: scripting (史铁生) → jia review
-   * Stage B: storyboarding (徐克) → jia review
-   * Stage C: prompt-assembly (顾长卫)
-   */
-  execution: {
-    strictSerial: true,
-    stages: [
-      {
-        name: 'scripting',
-        actor: { personaId: 'shi', specialty: 'scripting' },
-        reviewAfter: { personaId: 'jia', specialty: 'review' },
-        description: '基于方案写分段脚本'
-      },
-      {
-        name: 'storyboarding',
-        actor: { personaId: 'tsui', specialty: 'storyboarding' },
-        reviewAfter: { personaId: 'jia', specialty: 'review' },
-        description: '基于脚本设计分镜'
-      },
-      {
-        name: 'prompt-assembly',
-        actor: { personaId: 'gu', specialty: 'prompt-assembly' },
-        reviewAfter: null,
-        description: '组装最终 AI 视频 prompt'
-      }
-    ],
-    supportsParallel: false
-  },
-
-  qualityGate: {
-    reviewer: { personaId: 'gu', specialty: 'continuity-check' },
-    tester: null,
-    parallel: false,
-    checkAreas: ['锚点完整性', '色调统一', '角色描述一致', '时长合理'],
-    maxRetries: 3
-  },
-
-  acceptance: {
-    verifier: { personaId: 'jia', specialty: 'planning' },
-    focusAreas: ['叙事弧线完整', '情绪曲线合理', '视觉一致性', '整体感觉']
-  }
-};
-
-// =====================================================================
-// Flow Registry
-// =====================================================================
-
-const FLOWS = new Map([
-  ['dev', devFlow],
-  ['writing', writingFlow],
-  ['trading', tradingFlow],
-  ['video', videoFlow]
-]);
-
-/**
- * 获取场景流程定义
- * @param {string} scenario — 'dev' | 'writing' | 'trading' | 'video'
- * @returns {ScenarioFlow}
- */
-export function getFlow(scenario) {
-  const flow = FLOWS.get(scenario);
-  if (!flow) {
-    throw new Error(`Unknown scenario: ${scenario}. Available: ${[...FLOWS.keys()].join(', ')}`);
-  }
-  return flow;
-}
-
-/**
- * 获取所有已注册场景
- */
-export function getAvailableScenarios() {
-  return [...FLOWS.keys()];
-}
-
-// =====================================================================
-// Flow Query Helpers
-// =====================================================================
-
-/**
- * 获取 discussion 阶段应创建的 actor 列表
- * @param {ScenarioFlow} flow
- * @param {object} context — { uiInvolved: boolean }
- * @returns {Array<{personaId, specialty}>}
- */
-export function getDiscussionActors(flow, context = {}) {
-  const actors = flow.discussion.defaultActors || [];
-  return actors.filter(a => {
-    if (a.condition === 'ui_involved') return context.uiInvolved;
-    return true;
-  });
-}
-
-/**
- * 获取 planning 阶段应创建的 actor 列表
- */
-export function getPlanningActors(flow) {
-  return flow.planning.actors || [];
-}
-
-/**
- * 获取 planning lead persona ID
- */
-export function getPlanningLead(flow) {
-  return flow.planning.lead;
-}
-
-/**
- * 获取 quality gate 配置
- */
-export function getQualityGateConfig(flow) {
-  return flow.qualityGate;
-}
-
-/**
- * 获取 acceptance verifier
- */
-export function getAcceptanceVerifier(flow) {
-  return flow.acceptance.verifier;
-}
-
-/**
- * 判断某个步骤是否需要经过 quality gate
- *
- * @param {ScenarioFlow} flow
- * @param {string} specialty — 该步骤 actor 的 specialty
- * @returns {boolean}
- */
-export function needsQualityGate(flow, specialty) {
-  const qg = flow.qualityGate;
-  if (specialty === 'coding') {
-    return qg.codingRequiresReview !== false || qg.codingRequiresTesting !== false;
-  }
-  if (specialty === 'writing') {
-    return !!qg.reviewer;
-  }
-  if (specialty === 'design') {
-    return qg.designRequiresReview === true;
-  }
-  if (specialty === 'execution') {
-    // Trading: execution actor 的产出需要风控审查（如果配置了 QG before execution）
-    // 但如果 QG 已在 execution 之前跑过了，这里就不需要重复
-    // 其他场景中 execution specialty 不存在，直接返回 false
-    return flow.qualityGateBeforeExecution !== true
-      && flow.scenario === 'trading'
-      && !!qg.reviewer;
-  }
-  // 其他 specialty（prompt-assembly 等）— 检查是否有 reviewer
-  return false;
-}
-
-/**
- * 获取 post-planning discussion 配置（Trading 对抗性审查）
- * PLANNING 完成后自动进入此 discussion
- */
-export function getPostPlanningDiscussion(flow) {
-  return flow.postPlanningDiscussion || null;
-}
-
-/**
- * 该场景的 QG 是否在 EXECUTION 之前（Trading: 风控通过才能交易）
- */
-export function isQualityGateBeforeExecution(flow) {
-  return flow.qualityGateBeforeExecution === true;
-}
-
-/**
- * Video 特有：获取严格串行 stages
- */
-export function getExecutionStages(flow) {
-  if (flow.execution?.strictSerial) {
-    return flow.execution.stages || [];
-  }
-  return null;
-}