explain_plan

Instructions

Implementation Reference

• src/server.ts:289-302 (registration)

  Tool registration for 'explain_plan' in the ListToolsRequestSchema handler, defining its name, description, and inputSchema (requires 'task' string).

  {
    name: 'explain_plan',
    description: '解释 Tech Lead 会如何分配任务（不实际执行）',
    inputSchema: {
      type: 'object',
      properties: {
        task: {
          type: 'string',
          description: '要分析的任务描述',
        },
      },
      required: ['task'],
    },
  },

• src/server.ts:683-718 (handler)

  Handler for 'explain_plan' tool. Calls TechLead.analyze(task) to get a task analysis without executing it, then formats the response as a markdown plan showing workflow type, expert assignments, and execution order.

          case 'explain_plan': {
            const { task } = args as { task: string };
            
            // 让 Tech Lead 分析任务但不执行
            const analysis = await techLead.analyze(task);
            
            const expertPlan = analysis.experts
              .map((e: { id: string; name: string; tier: string; role: string }, i: number) => {
                const subtask = analysis.subtasks.find(t => t.expertId === e.id);
                return `${i + 1}. **${e.name}** (${e.tier})\n   - 角色: ${e.role.slice(0, 100)}...\n   - 任务: ${subtask?.description || '待分配'}`;
              })
              .join('\n\n');
            
            const plan = `# 🧠 任务执行计划
  
  ## 任务分析
  **原始任务**: ${task}
  
  ## Tech Lead 分析结果
  
  ### 工作流类型
  \`${analysis.workflow}\` ${analysis.workflow === 'sequential' ? '(顺序执行)' : analysis.workflow === 'parallel' ? '(并行执行)' : '(并行+审查)'}
  
  ### 专家分配 (${analysis.experts.length} 个)
  ${expertPlan}
  
  ### 执行顺序
  ${analysis.workflow === 'parallel' ? '所有专家将并行执行任务' : analysis.experts.map((e: { name: string }, i: number) => `${i + 1}. ${e.name}`).join(' → ')}
  
  ---
  > 💡 这只是计划预览，使用 \`team_work\` 工具实际执行任务`;
  
            return {
              content: [{ type: 'text', text: plan }],
            };
          }

• src/agents/tech-lead.ts:159-173 (helper)

  TechLead.analyze() - the core AI-driven analysis method called by the explain_plan handler. Sends the task to an LLM with a system prompt asking it to dynamically create experts, decompose tasks, and choose a workflow.

  async analyze(task: string, context?: string): Promise<TaskAnalysis> {
    // 构建消息
    const userContent = context
      ? `用户需求：${task}\n\n额外上下文：${context}`
      : `用户需求：${task}`;
  
    const messages: ChatMessage[] = [
      { role: 'system', content: SYSTEM_PROMPT },
      { role: 'user', content: userContent },
    ];
  
    // 获取分析结果
    const response = await this.adapter.chat(messages);
    return this.parseAnalysis(response);
  }

• src/agents/tech-lead.ts:50-63 (schema)

  TaskAnalysis interface - the type definition for the analysis result returned by techLead.analyze(), including summary, experts, subtasks, workflow type, and needsReview flag.

  export interface TaskAnalysis {
    /** 需求摘要 */
    readonly summary: string;
    /** 动态生成的专家列表 */
    readonly experts: readonly DynamicExpert[];
    /** 子任务列表 */
    readonly subtasks: readonly SubTask[];
    /** 工作流类型 */
    readonly workflow: WorkflowType;
    /** 是否需要最终审查 */
    readonly needsReview: boolean;
    /** 额外说明 */
    readonly notes?: string;
  }