ethan-skill 1.12.0 → 1.14.0

package/dist/agents/orchestrator.js

@@ -0,0 +1,654 @@
+"use strict";
+/**
+ * Multi-Agent 编排器
+ * 将 Pipeline 的步骤分配给不同 Agent，生成协作式超级 Prompt
+ * 执行模型：Prompt 生成（粘贴到 AI 编辑器执行，零 API 成本）
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.buildMultiAgentPrompt = buildMultiAgentPrompt;
+const presets_1 = require("./presets");
+/** 根据 skillId 解析对应的 Agent（兜底 Code Agent） */
+function resolveAgent(skillId, agents, routing) {
+    const agentId = routing[skillId];
+    return agents.find((a) => a.id === agentId) ?? agents.find((a) => a.id === 'coder') ?? agents[0];
+}
+/** 生成 Multi-Agent 编排超级 Prompt */
+function buildMultiAgentPrompt(pipeline, skills, agents, options) {
+    const { context, lang = 'zh', snapshot, mode = 'sequential' } = options;
+    const isEn = lang === 'en';
+    const routing = (0, presets_1.buildSkillRouting)(agents);
+    const participatingAgents = (0, presets_1.getAgentsForPipeline)(pipeline.skillIds, agents, routing);
+    const total = skills.length;
+    if (mode === 'parallel') {
+        return isEn
+            ? buildEnParallelPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total)
+            : buildZhParallelPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total);
+    }
+    if (mode === 'review-loop') {
+        return isEn
+            ? buildEnReviewLoopPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total)
+            : buildZhReviewLoopPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total);
+    }
+    if (mode === 'consensus') {
+        return isEn
+            ? buildEnConsensusPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total)
+            : buildZhConsensusPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total);
+    }
+    // sequential（默认）
+    if (isEn) {
+        return buildEnPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total);
+    }
+    return buildZhPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total);
+}
+// ─── 中文版 ──────────────────────────────────────────────────────────────────
+function buildZhPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total) {
+    const header = `# 🤖 Multi-Agent 编排：${pipeline.name}
+
+**任务背景**：${context}
+**Pipeline**：${pipeline.name}（\`${pipeline.id}\`）
+**执行模式**：Multi-Agent 协作编排
+**参与 Agents**：${participatingAgents.length} 个  |  **总步数**：${total} 步`;
+    // Agent 阵容表
+    const agentStepMap = buildAgentStepMap(skills, agents, routing, false);
+    const rosterRows = participatingAgents.map((a) => {
+        const steps = agentStepMap[a.id] || [];
+        const stepStr = steps.length > 0 ? `步骤 ${steps.join('、')}` : '—';
+        return `| ${a.emoji} **${a.name}** | ${a.role} | ${stepStr} |`;
+    }).join('\n');
+    const roster = `## 🎭 Agent 阵容
+
+| Agent | 职责 | 负责步骤 |
+|-------|------|---------|
+${rosterRows}`;
+    const protocol = `## 📋 协作协议（必须严格遵守）
+
+1. 每个 Agent 严格按分配步骤顺序执行，**不越权执行其他 Agent 的步骤**
+2. 每步完成后输出 **Handoff 摘要**（≤150字），交接给下一个 Agent
+3. 下一个 Agent 必须**先阅读 Handoff 摘要**，再开始执行
+4. 所有步骤完成后，将输出折叠在 \`<details>\` 标签中
+5. 全部步骤完成后，输出一份完整的 Multi-Agent 执行报告
+
+**每步输出格式：**
+
+\`\`\`
+<details>
+<summary>✅ 步骤 N：[Agent名] · [Skill名] — 已完成</summary>
+
+[本步骤完整输出]
+
+</details>
+\`\`\``;
+    // 步骤详情（含 Agent 标签和 Handoff）
+    const stepsSection = skills.map((skill, i) => {
+        const stepNum = i + 1;
+        const agent = resolveAgent(skill.id, agents, routing);
+        const nextAgent = i < skills.length - 1 ? resolveAgent(skills[i + 1].id, agents, routing) : null;
+        const handoffFrom = i > 0 ? resolveAgent(skills[i - 1].id, agents, routing) : null;
+        const handoffNote = handoffFrom && handoffFrom.id !== agent.id
+            ? `> 📨 **Handoff 摘要**（来自 ${handoffFrom.emoji} ${handoffFrom.name}）：\`[上一步核心产出，由上一 Agent 填写]\`\n\n`
+            : i > 0
+                ? `> 📝 **前序输出**（${agent.emoji} ${agent.name} 连续执行）：使用步骤 ${i} 的核心产出作为背景。\n\n`
+                : '';
+        const stepsContent = skill.steps
+            .map((s, si) => {
+            const title = s.title.replace(/^\d+[.、]\s*/, '');
+            const indented = s.content.split('\n').map((l) => `   ${l}`).join('\n');
+            return `   ${si + 1}. **${title}**\n${indented}`;
+        })
+            .join('\n\n');
+        const handoffTo = nextAgent && nextAgent.id !== agent.id
+            ? `\n\n#### ✅ Handoff → ${nextAgent.emoji} ${nextAgent.name}\n\n> 🔖 请在此输出本步核心产出（≤150字），交接给 **${nextAgent.name}**：\n> \`[请填写：完成了什么、关键决策、下一步需要注意什么]\``
+            : '';
+        return `### ${agent.emoji} ${agent.name} — 步骤 ${stepNum}/${total}：${skill.name}
+
+${handoffNote}**目标**：${skill.description}
+
+**执行内容**：
+
+${stepsContent}
+
+**输出格式**：${skill.outputFormat}${handoffTo}`;
+    }).join('\n\n---\n\n');
+    // 最终报告模板
+    const reportTemplate = buildZhReportTemplate(pipeline, skills, agents, participatingAgents, routing, context);
+    const startCmd = `**立即开始执行：${participatingAgents[0].emoji} ${participatingAgents[0].name} 执行步骤 1。**`;
+    const snapshotBlock = snapshot ?? '';
+    return [header, snapshotBlock, '---', roster, '---', protocol, '---', `## 🔄 执行流程`, stepsSection, '---', reportTemplate, startCmd]
+        .filter(Boolean)
+        .join('\n\n');
+}
+function buildZhReportTemplate(pipeline, skills, agents, participatingAgents, routing, context) {
+    // 按 Agent 分组步骤
+    const agentSections = participatingAgents.map((agent) => {
+        const agentSkills = skills.filter((s) => {
+            const a = resolveAgent(s.id, agents, routing);
+            return a.id === agent.id;
+        });
+        const skillList = agentSkills.map((s) => `### ${s.name}\n[完整输出]\n\n---`).join('\n\n');
+        return `## ${agent.emoji} ${agent.name} 产出\n\n${skillList}`;
+    }).join('\n\n');
+    return `## 📄 最终输出要求
+
+全部步骤完成后，输出以下 Multi-Agent 执行报告：
+
+# Multi-Agent 执行报告 — ${context}
+> Pipeline：${pipeline.name} | Agents：${participatingAgents.map((a) => `${a.emoji} ${a.name}`).join(' → ')}
+
+${agentSections}
+
+*由 Ethan Multi-Agent 自动生成*`;
+}
+// ─── 英文版 ──────────────────────────────────────────────────────────────────
+function buildEnPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total) {
+    const header = `# 🤖 Multi-Agent Orchestration: ${pipeline.name}
+
+**Context**: ${context}
+**Pipeline**: ${pipeline.name} (\`${pipeline.id}\`)
+**Mode**: Multi-Agent Collaboration
+**Agents**: ${participatingAgents.length}  |  **Steps**: ${total}`;
+    const agentStepMap = buildAgentStepMap(skills, agents, routing, false);
+    const rosterRows = participatingAgents.map((a) => {
+        const steps = agentStepMap[a.id] || [];
+        const stepStr = steps.length > 0 ? `Steps ${steps.join(', ')}` : '—';
+        return `| ${a.emoji} **${a.nameEn}** | ${a.role} | ${stepStr} |`;
+    }).join('\n');
+    const roster = `## 🎭 Agent Roster
+
+| Agent | Role | Assigned Steps |
+|-------|------|----------------|
+${rosterRows}`;
+    const protocol = `## 📋 Collaboration Protocol (Must Follow)
+
+1. Each Agent executes only its assigned steps — **no overstepping**
+2. After each step, output a **Handoff Summary** (≤150 words) for the next Agent
+3. The next Agent must **read the Handoff Summary** before starting
+4. Wrap each step output in \`<details>\` collapse tags
+5. After all steps complete, output a merged Multi-Agent Execution Report
+
+**Per-step output format:**
+
+\`\`\`
+<details>
+<summary>✅ Step N: [Agent Name] · [Skill Name] — Completed</summary>
+
+[Full step output]
+
+</details>
+\`\`\``;
+    const stepsSection = skills.map((skill, i) => {
+        const stepNum = i + 1;
+        const agent = resolveAgent(skill.id, agents, routing);
+        const nextAgent = i < skills.length - 1 ? resolveAgent(skills[i + 1].id, agents, routing) : null;
+        const handoffFrom = i > 0 ? resolveAgent(skills[i - 1].id, agents, routing) : null;
+        const handoffNote = handoffFrom && handoffFrom.id !== agent.id
+            ? `> 📨 **Handoff Summary** (from ${handoffFrom.emoji} ${handoffFrom.nameEn}): \`[Previous step's key output — filled by prior Agent]\`\n\n`
+            : i > 0
+                ? `> 📝 **Prior Output** (${agent.emoji} ${agent.nameEn} continues): Use Step ${i} core output as background.\n\n`
+                : '';
+        const stepsContent = skill.steps
+            .map((s, si) => {
+            const title = s.title.replace(/^\d+[.、]\s*/, '');
+            const indented = s.content.split('\n').map((l) => `   ${l}`).join('\n');
+            return `   ${si + 1}. **${title}**\n${indented}`;
+        })
+            .join('\n\n');
+        const handoffTo = nextAgent && nextAgent.id !== agent.id
+            ? `\n\n#### ✅ Handoff → ${nextAgent.emoji} ${nextAgent.nameEn}\n\n> 🔖 Output core outcome (≤150 words) for **${nextAgent.nameEn}**:\n> \`[Fill in: what was completed, key decisions, notes for next Agent]\``
+            : '';
+        return `### ${agent.emoji} ${agent.nameEn} — Step ${stepNum}/${total}: ${skill.nameEn.replace(/_/g, ' ')}
+
+${handoffNote}**Goal**: ${skill.descriptionEn ?? skill.description}
+
+**Execution**:
+
+${stepsContent}
+
+**Output Format**: ${skill.outputFormat}${handoffTo}`;
+    }).join('\n\n---\n\n');
+    const reportTemplate = buildEnReportTemplate(pipeline, skills, agents, participatingAgents, routing, context);
+    const startCmd = `**Begin: ${participatingAgents[0].emoji} ${participatingAgents[0].nameEn} executes Step 1 now.**`;
+    const snapshotBlock = snapshot ?? '';
+    return [header, snapshotBlock, '---', roster, '---', protocol, '---', `## 🔄 Execution Flow`, stepsSection, '---', reportTemplate, startCmd]
+        .filter(Boolean)
+        .join('\n\n');
+}
+function buildEnReportTemplate(pipeline, skills, agents, participatingAgents, routing, context) {
+    const agentSections = participatingAgents.map((agent) => {
+        const agentSkills = skills.filter((s) => {
+            const a = resolveAgent(s.id, agents, routing);
+            return a.id === agent.id;
+        });
+        const skillList = agentSkills.map((s) => `### ${s.nameEn.replace(/_/g, ' ')}\n[Full output]\n\n---`).join('\n\n');
+        return `## ${agent.emoji} ${agent.nameEn} Output\n\n${skillList}`;
+    }).join('\n\n');
+    return `## 📄 Final Output Requirements
+
+After all steps complete, output this Multi-Agent Execution Report:
+
+# Multi-Agent Execution Report — ${context}
+> Pipeline: ${pipeline.name} | Agents: ${participatingAgents.map((a) => `${a.emoji} ${a.nameEn}`).join(' → ')}
+
+${agentSections}
+
+*Auto-generated by Ethan Multi-Agent*`;
+}
+/**
+ * 构建 agentId → 步骤序号列表 的映射，用于 Agent 阵容表展示
+ */
+function buildAgentStepMap(skills, agents, routing, _isEn) {
+    const map = {};
+    skills.forEach((skill, i) => {
+        const agent = resolveAgent(skill.id, agents, routing);
+        if (!map[agent.id])
+            map[agent.id] = [];
+        map[agent.id].push(i + 1);
+    });
+    return map;
+}
+// ─── Parallel 模式 ────────────────────────────────────────────────────────────
+function buildZhParallelPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total) {
+    const agentStepMap = buildAgentStepMap(skills, agents, routing, false);
+    const rosterRows = participatingAgents.map((a) => {
+        const steps = agentStepMap[a.id] || [];
+        return `| ${a.emoji} **${a.name}** | ${a.role} | 步骤 ${steps.join('、')} |`;
+    }).join('\n');
+    const header = `# 🤖 Multi-Agent 并行分析：${pipeline.name}
+
+**任务背景**：${context}
+**Pipeline**：${pipeline.name}（\`${pipeline.id}\`）
+**执行模式**：🔀 Parallel — 各 Agent 并行输出，最后汇总整合
+**参与 Agents**：${participatingAgents.length} 个  |  **总步数**：${total} 步
+
+## 🎭 Agent 阵容
+
+| Agent | 职责 | 负责步骤 |
+|-------|------|---------|
+${rosterRows}`;
+    const protocol = `## 📋 并行协议（必须严格遵守）
+
+1. 本次采用 **并行模式**：所有 Agent 基于同一任务背景，**同时**从各自专业视角输出分析
+2. 每个 Agent 输出完成后，进入"🏛️ 汇总整合"阶段
+3. 汇总整合阶段负责合并所有 Agent 产出，提炼出统一结论与行动项
+4. 用 \`<details>\` 折叠各 Agent 的详细输出，最终报告呈现汇总结论
+
+**输出格式：**
+\`\`\`
+<details>
+<summary>✅ [Agent名] 分析完成</summary>
+[Agent 完整输出]
+</details>
+\`\`\``;
+    const taskDesc = `## 📢 任务发布（全体 Agent 同时接收）
+
+> **任务**：${context}
+> **需要从以下维度并行分析并给出产出：**`;
+    const agentBlocks = participatingAgents.map((agent) => {
+        const agentSkills = skills.filter((s) => resolveAgent(s.id, agents, routing).id === agent.id);
+        const skillNames = agentSkills.map((s) => s.name).join('、');
+        const stepsContent = agentSkills.map((skill) => {
+            const stepsList = skill.steps.map((s, si) => {
+                const title = s.title.replace(/^\d+[.、]\s*/, '');
+                const indented = s.content.split('\n').map((l) => `   ${l}`).join('\n');
+                return `   ${si + 1}. **${title}**\n${indented}`;
+            }).join('\n\n');
+            return `#### ${skill.name}\n${stepsList}\n**输出格式**：${skill.outputFormat}`;
+        }).join('\n\n---\n\n');
+        return `### ${agent.emoji} ${agent.name} — 并行分析（${skillNames}）
+
+> 从 **${agent.role}** 视角，对任务进行全面分析：
+
+${stepsContent}`;
+    }).join('\n\n---\n\n');
+    const synthesis = `## 🏛️ 汇总整合阶段
+
+> 所有 Agent 输出完成后，进行以下整合：
+
+1. **共识点**：各 Agent 分析中的一致结论
+2. **分歧与补充**：各 Agent 的独特见解与专项建议
+3. **行动项**（优先级排序）：
+   - 🔴 立即行动
+   - 🟡 近期计划
+   - 🟢 长期优化
+4. **最终建议**：综合各角度的最优方案
+
+---
+
+# 并行分析报告 — ${context}
+> Pipeline：${pipeline.name} | 模式：Parallel | Agents：${participatingAgents.map((a) => `${a.emoji} ${a.name}`).join(' + ')}
+
+## 汇总结论
+[由上述各 Agent 分析综合得出]
+
+*由 Ethan Multi-Agent Parallel 自动生成*`;
+    return [header, snapshot ?? '', '---', protocol, '---', taskDesc, agentBlocks, '---', synthesis]
+        .filter(Boolean)
+        .join('\n\n');
+}
+function buildEnParallelPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total) {
+    const agentStepMap = buildAgentStepMap(skills, agents, routing, true);
+    const rosterRows = participatingAgents.map((a) => {
+        const steps = agentStepMap[a.id] || [];
+        return `| ${a.emoji} **${a.nameEn}** | ${a.role} | Steps ${steps.join(', ')} |`;
+    }).join('\n');
+    const header = `# 🤖 Multi-Agent Parallel Analysis: ${pipeline.name}
+
+**Context**: ${context}
+**Pipeline**: ${pipeline.name} (\`${pipeline.id}\`)
+**Mode**: 🔀 Parallel — All agents analyze simultaneously, then synthesize
+**Agents**: ${participatingAgents.length}  |  **Steps**: ${total}
+
+## 🎭 Agent Roster
+
+| Agent | Role | Assigned Steps |
+|-------|------|----------------|
+${rosterRows}`;
+    const agentBlocks = participatingAgents.map((agent) => {
+        const agentSkills = skills.filter((s) => resolveAgent(s.id, agents, routing).id === agent.id);
+        const skillNames = agentSkills.map((s) => s.nameEn.replace(/_/g, ' ')).join(', ');
+        const stepsContent = agentSkills.map((skill) => {
+            const stepsList = skill.steps.map((s, si) => {
+                const title = s.title.replace(/^\d+[.、]\s*/, '');
+                const indented = s.content.split('\n').map((l) => `   ${l}`).join('\n');
+                return `   ${si + 1}. **${title}**\n${indented}`;
+            }).join('\n\n');
+            return `#### ${skill.nameEn.replace(/_/g, ' ')}\n${stepsList}\n**Output Format**: ${skill.outputFormat}`;
+        }).join('\n\n---\n\n');
+        return `### ${agent.emoji} ${agent.nameEn} — Parallel Analysis (${skillNames})
+
+${stepsContent}`;
+    }).join('\n\n---\n\n');
+    const synthesis = `## 🏛️ Synthesis Phase
+
+1. **Consensus**: Agreed conclusions from all agents
+2. **Divergence & Additions**: Unique insights per agent
+3. **Action Items** (prioritized):
+   - 🔴 Immediate
+   - 🟡 Near-term
+   - 🟢 Long-term
+4. **Final Recommendation**: Optimal approach from all angles
+
+---
+
+# Parallel Analysis Report — ${context}
+> Pipeline: ${pipeline.name} | Mode: Parallel | Agents: ${participatingAgents.map((a) => `${a.emoji} ${a.nameEn}`).join(' + ')}
+
+*Auto-generated by Ethan Multi-Agent Parallel*`;
+    return [header, snapshot ?? '', '---', `## 📢 Task Broadcast\n\n> **Task**: ${context}\n> All agents analyze in parallel from their domain.`, agentBlocks, '---', synthesis]
+        .filter(Boolean)
+        .join('\n\n');
+}
+// ─── Review-Loop 模式 ─────────────────────────────────────────────────────────
+function buildZhReviewLoopPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total) {
+    const coderAgent = agents.find((a) => a.id === 'coder') ?? participatingAgents[0];
+    const reviewerAgent = agents.find((a) => a.id === 'reviewer') ?? agents.find((a) => a.id === 'qa') ?? participatingAgents[participatingAgents.length - 1];
+    const header = `# 🤖 Multi-Agent Review-Loop：${pipeline.name}
+
+**任务背景**：${context}
+**Pipeline**：${pipeline.name}（\`${pipeline.id}\`）
+**执行模式**：🔄 Review-Loop — 实现 → 审查 → 修改（最多 2 轮迭代）
+**核心 Agents**：${coderAgent.emoji} ${coderAgent.name}（执行）+ ${reviewerAgent.emoji} ${reviewerAgent.name}（审查）
+**总步数**：${total} 步`;
+    const protocol = `## 📋 Review-Loop 协议（必须严格遵守）
+
+1. **Loop 1 — 实现阶段**：${coderAgent.emoji} ${coderAgent.name} 按步骤执行全部任务
+2. **Loop 1 — 审查阶段**：${reviewerAgent.emoji} ${reviewerAgent.name} 对实现结果进行全面审查
+   - 若审查通过（无 Blocker）→ 直接输出最终报告
+   - 若有 Blocker/Major 问题 → 进入 Loop 2
+3. **Loop 2 — 修改阶段**：${coderAgent.emoji} ${coderAgent.name} 仅修复审查指出的问题
+4. **Loop 2 — 确认阶段**：${reviewerAgent.emoji} ${reviewerAgent.name} 确认修复，输出最终报告
+5. **最多 2 轮迭代**，第 2 轮结束后无论如何输出最终报告`;
+    const stepsContent = skills.map((skill, i) => {
+        const stepNum = i + 1;
+        const stepsList = skill.steps.map((s, si) => {
+            const title = s.title.replace(/^\d+[.、]\s*/, '');
+            const indented = s.content.split('\n').map((l) => `   ${l}`).join('\n');
+            return `   ${si + 1}. **${title}**\n${indented}`;
+        }).join('\n\n');
+        return `#### 步骤 ${stepNum}：${skill.name}\n${stepsList}\n**输出格式**：${skill.outputFormat}`;
+    }).join('\n\n---\n\n');
+    const loop1 = `## 🔄 Loop 1 — 实现阶段
+
+${coderAgent.emoji} **${coderAgent.name}** 按顺序执行以下所有步骤：
+
+${stepsContent}
+
+---
+
+## 🔄 Loop 1 — 审查阶段
+
+${reviewerAgent.emoji} **${reviewerAgent.name}** 对上述全部实现进行审查：
+
+1. 检查每个步骤的输出是否满足任务目标
+2. 识别 🚫 Blocker / ⚠️ Major / 💡 Minor 问题
+3. 评估整体质量与完整性
+
+**审查结论：**
+- [ ] ✅ 通过（无 Blocker/Major）→ 跳过 Loop 2，直接输出最终报告
+- [ ] ⚠️ 需要修改（有 Blocker/Major）→ 进入 Loop 2
+
+**问题清单（如有）：**
+| 问题级别 | 步骤 | 问题描述 | 修改建议 |
+|---------|------|---------|---------|
+| [级别] | [步骤N] | [描述] | [建议] |`;
+    const loop2 = `## 🔄 Loop 2 — 修改阶段（仅在 Loop 1 审查不通过时执行）
+
+${coderAgent.emoji} **${coderAgent.name}** 根据审查意见，**仅修复 Blocker 和 Major 问题**：
+
+> 📨 **审查意见摘要**（来自 ${reviewerAgent.emoji} ${reviewerAgent.name}）：\`[Loop 1 审查问题列表]\`
+
+针对每个 Blocker/Major 问题输出修改内容。
+
+---
+
+## 🔄 Loop 2 — 确认阶段
+
+${reviewerAgent.emoji} **${reviewerAgent.name}** 确认修复是否满足要求：
+
+- [ ] ✅ 修复通过 → 输出最终报告
+- [ ] ❌ 仍有问题 → 记录遗留问题，强制输出最终报告（已达迭代上限）`;
+    const finalReport = `## 📄 最终输出
+
+# Review-Loop 执行报告 — ${context}
+> Pipeline：${pipeline.name} | 模式：Review-Loop | Agents：${coderAgent.emoji} ${coderAgent.name} ⇄ ${reviewerAgent.emoji} ${reviewerAgent.name}
+
+## 最终实现产出
+[Loop 1 或 Loop 2 修改后的完整输出]
+
+## 审查摘要
+- 发现问题：[Blocker X 个 | Major Y 个 | Minor Z 个]
+- 已修复：[描述]
+- 遗留（如有）：[描述]
+
+*由 Ethan Multi-Agent Review-Loop 自动生成*`;
+    return [header, snapshot ?? '', '---', protocol, '---', loop1, '---', loop2, '---', finalReport]
+        .filter(Boolean)
+        .join('\n\n');
+}
+function buildEnReviewLoopPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total) {
+    const coderAgent = agents.find((a) => a.id === 'coder') ?? participatingAgents[0];
+    const reviewerAgent = agents.find((a) => a.id === 'reviewer') ?? agents.find((a) => a.id === 'qa') ?? participatingAgents[participatingAgents.length - 1];
+    const header = `# 🤖 Multi-Agent Review-Loop: ${pipeline.name}
+
+**Context**: ${context}
+**Pipeline**: ${pipeline.name} (\`${pipeline.id}\`)
+**Mode**: 🔄 Review-Loop — Implement → Review → Fix (max 2 iterations)
+**Core Agents**: ${coderAgent.emoji} ${coderAgent.nameEn} (execute) + ${reviewerAgent.emoji} ${reviewerAgent.nameEn} (review)
+**Steps**: ${total}`;
+    const stepsContent = skills.map((skill, i) => {
+        const stepNum = i + 1;
+        const stepsList = skill.steps.map((s, si) => {
+            const title = s.title.replace(/^\d+[.、]\s*/, '');
+            const indented = s.content.split('\n').map((l) => `   ${l}`).join('\n');
+            return `   ${si + 1}. **${title}**\n${indented}`;
+        }).join('\n\n');
+        return `#### Step ${stepNum}: ${skill.nameEn.replace(/_/g, ' ')}\n${stepsList}\n**Output Format**: ${skill.outputFormat}`;
+    }).join('\n\n---\n\n');
+    const loop1 = `## 🔄 Loop 1 — Implementation
+
+${coderAgent.emoji} **${coderAgent.nameEn}** executes all steps in order:
+
+${stepsContent}
+
+---
+
+## 🔄 Loop 1 — Review
+
+${reviewerAgent.emoji} **${reviewerAgent.nameEn}** reviews all outputs:
+
+- [ ] ✅ Pass (no Blockers/Major) → Skip Loop 2, output final report
+- [ ] ⚠️ Issues found → Enter Loop 2`;
+    const loop2 = `## 🔄 Loop 2 — Fix (only if Loop 1 review failed)
+
+${coderAgent.emoji} **${coderAgent.nameEn}** fixes Blocker and Major issues only.
+
+---
+
+## 🔄 Loop 2 — Confirm
+
+${reviewerAgent.emoji} **${reviewerAgent.nameEn}** confirms fixes, then outputs final report.`;
+    const finalReport = `## 📄 Final Output
+
+# Review-Loop Report — ${context}
+> Pipeline: ${pipeline.name} | Mode: Review-Loop | Agents: ${coderAgent.emoji} ${coderAgent.nameEn} ⇄ ${reviewerAgent.emoji} ${reviewerAgent.nameEn}
+
+*Auto-generated by Ethan Multi-Agent Review-Loop*`;
+    return [header, snapshot ?? '', '---', loop1, '---', loop2, '---', finalReport]
+        .filter(Boolean)
+        .join('\n\n');
+}
+// ─── Consensus 模式 ───────────────────────────────────────────────────────────
+function buildZhConsensusPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total) {
+    const agentStepMap = buildAgentStepMap(skills, agents, routing, false);
+    const rosterRows = participatingAgents.map((a) => {
+        const steps = agentStepMap[a.id] || [];
+        return `| ${a.emoji} **${a.name}** | ${a.role} | 步骤 ${steps.join('、')} |`;
+    }).join('\n');
+    const header = `# 🤖 Multi-Agent 共识决策：${pipeline.name}
+
+**任务背景**：${context}
+**Pipeline**：${pipeline.name}（\`${pipeline.id}\`）
+**执行模式**：🏛️ Consensus — 各 Agent 独立提案，共识整合最优解
+**参与 Agents**：${participatingAgents.length} 个  |  **总步数**：${total} 步
+
+## 🎭 Agent 阵容
+
+| Agent | 职责 | 负责步骤 |
+|-------|------|---------|
+${rosterRows}`;
+    const protocol = `## 📋 共识协议（必须严格遵守）
+
+1. **提案阶段**：每个 Agent 从自己的视角，**独立**地为任务给出完整方案
+2. Agent 之间**互不干扰**，不参考其他 Agent 的方案
+3. **共识阶段**：比较所有提案，提炼共识、识别分歧
+4. **最终推荐**：基于共识，输出最优综合方案与明确行动计划`;
+    const proposalBlocks = participatingAgents.map((agent, idx) => {
+        const agentSkills = skills.filter((s) => resolveAgent(s.id, agents, routing).id === agent.id);
+        const stepsContent = agentSkills.map((skill) => {
+            const stepsList = skill.steps.map((s, si) => {
+                const title = s.title.replace(/^\d+[.、]\s*/, '');
+                const indented = s.content.split('\n').map((l) => `   ${l}`).join('\n');
+                return `   ${si + 1}. **${title}**\n${indented}`;
+            }).join('\n\n');
+            return `#### ${skill.name}\n${stepsList}\n**输出格式**：${skill.outputFormat}`;
+        }).join('\n\n---\n\n');
+        return `### 提案 ${idx + 1}：${agent.emoji} ${agent.name}
+
+> 以下是 **${agent.name}** 基于 **${agent.role}** 视角的独立方案：
+
+${stepsContent}`;
+    }).join('\n\n---\n\n');
+    const consensus = `## 🏛️ 共识整合阶段
+
+> 所有提案提交完毕，现在进行共识分析：
+
+### 1. 共识点（各方认同的结论）
+[列出各 Agent 方案中一致的内容]
+
+### 2. 分歧点（不同视角的差异）
+| 维度 | ${participatingAgents.map((a) => a.name).join(' | ')} | 分析 |
+|------|${participatingAgents.map(() => '---').join('|')}|------|
+| [方面] | [各 Agent 观点] | [如何取舍] |
+
+### 3. 最终推荐方案
+基于共识整合，推荐以下最优方案：
+
+[综合各 Agent 优势的最终方案]
+
+### 4. 行动计划
+| 优先级 | 行动项 | 负责方 |
+|--------|--------|--------|
+| 🔴 P0 | [立即执行] | [Agent] |
+| 🟡 P1 | [近期计划] | [Agent] |
+
+---
+
+# 共识决策报告 — ${context}
+> Pipeline：${pipeline.name} | 模式：Consensus | Agents：${participatingAgents.map((a) => `${a.emoji} ${a.name}`).join(' + ')}
+
+*由 Ethan Multi-Agent Consensus 自动生成*`;
+    return [header, snapshot ?? '', '---', protocol, '---', `## 📝 提案阶段（各 Agent 独立输出）`, proposalBlocks, '---', consensus]
+        .filter(Boolean)
+        .join('\n\n');
+}
+function buildEnConsensusPrompt(pipeline, skills, agents, participatingAgents, routing, context, snapshot, total) {
+    const agentStepMap = buildAgentStepMap(skills, agents, routing, true);
+    const rosterRows = participatingAgents.map((a) => {
+        const steps = agentStepMap[a.id] || [];
+        return `| ${a.emoji} **${a.nameEn}** | ${a.role} | Steps ${steps.join(', ')} |`;
+    }).join('\n');
+    const header = `# 🤖 Multi-Agent Consensus: ${pipeline.name}
+
+**Context**: ${context}
+**Pipeline**: ${pipeline.name} (\`${pipeline.id}\`)
+**Mode**: 🏛️ Consensus — Each agent proposes independently, then synthesize
+**Agents**: ${participatingAgents.length}  |  **Steps**: ${total}
+
+## 🎭 Agent Roster
+
+| Agent | Role | Assigned Steps |
+|-------|------|----------------|
+${rosterRows}`;
+    const proposalBlocks = participatingAgents.map((agent, idx) => {
+        const agentSkills = skills.filter((s) => resolveAgent(s.id, agents, routing).id === agent.id);
+        const stepsContent = agentSkills.map((skill) => {
+            const stepsList = skill.steps.map((s, si) => {
+                const title = s.title.replace(/^\d+[.、]\s*/, '');
+                const indented = s.content.split('\n').map((l) => `   ${l}`).join('\n');
+                return `   ${si + 1}. **${title}**\n${indented}`;
+            }).join('\n\n');
+            return `#### ${skill.nameEn.replace(/_/g, ' ')}\n${stepsList}\n**Output Format**: ${skill.outputFormat}`;
+        }).join('\n\n---\n\n');
+        return `### Proposal ${idx + 1}: ${agent.emoji} ${agent.nameEn}
+
+> Independent proposal from **${agent.nameEn}** (${agent.role}):
+
+${stepsContent}`;
+    }).join('\n\n---\n\n');
+    const consensus = `## 🏛️ Consensus Phase
+
+### 1. Agreed Points
+[Common conclusions across all proposals]
+
+### 2. Divergences
+| Dimension | ${participatingAgents.map((a) => a.nameEn).join(' | ')} | Resolution |
+|-----------|${participatingAgents.map(() => '---').join('|')}|------------|
+
+### 3. Final Recommendation
+[Synthesized optimal approach]
+
+### 4. Action Plan
+| Priority | Action | Owner |
+|----------|--------|-------|
+| 🔴 P0 | [Immediate] | [Agent] |
+
+---
+
+# Consensus Report — ${context}
+> Pipeline: ${pipeline.name} | Mode: Consensus | Agents: ${participatingAgents.map((a) => `${a.emoji} ${a.nameEn}`).join(' + ')}
+
+*Auto-generated by Ethan Multi-Agent Consensus*`;
+    return [header, snapshot ?? '', '---', `## 📝 Proposal Phase (Each Agent Independent)`, proposalBlocks, '---', consensus]
+        .filter(Boolean)
+        .join('\n\n');
+}
+//# sourceMappingURL=orchestrator.js.map