diff --git a/agents/zhuyuan-dev-agent/modules/plan-expander.js b/agents/zhuyuan-dev-agent/modules/plan-expander.js
index c76d83b..c2ada37 100644
--- a/agents/zhuyuan-dev-agent/modules/plan-expander.js
+++ b/agents/zhuyuan-dev-agent/modules/plan-expander.js
@@ -1,77 +1,99 @@
-// ═══════════════════════════════════════
-// 铸渊开发Agent · 任务补全引擎
+// HLDP-ZY://agents/zhuyuan-dev-agent/modules/plan-expander.js
+// @guardian: ICE-GL-ZY001 · 铸渊
+// @sovereign: TCS-0002∞ · 冰朔
+// @copyright: 国作登字-2026-A-00037559
+// 铸渊开发Agent · 任务补全引擎 v2.0
 // 高层面规划 → LLM补全 → 详细可执行步骤
-// ═══════════════════════════════════════
+// v2.0: 修复LLM提示词缺少target/content → 增加光湖术语上下文 → 退化继承机制
 
 const fs = require("fs");
 const path = require("path");
 
-/**
- * 补全单条高级任务为详细步骤
- * @param {object} task - { id, action, description, context }
- * @param {function} modelCaller - LLM 调用函数 (prompt) => string
- * @returns {object} 补全后的任务
- */
 async function expandTask(task, modelCaller) {
-  const prompt = [
-    "你是一个开发Agent的任务补全引擎。将高级描述补全为可执行的详细步骤。",
-    "",
-    "原任务:",
-    "  ID: " + task.id,
-    "  动作: " + task.action,
+  const taskInfo = [
+    "【原任务】",
+    "  任务ID: " + task.id,
+    "  动作: " + (task.action || "未知"),
+    "  目标路径/命令: " + (task.target || "未指定"),
+    "  写入内容: " + (task.content || "未指定"),
     "  描述: " + (task.description || "无"),
     "  上下文: " + (task.context || "无"),
     "",
-    "请严格基于以上原任务拆解。不要改变原任务的action/target/content。只补充detail、depends_on等元信息。如果原任务已经足够具体，直接返回：",
-    "[{\"step\":1,\"action\":\"write_file|run_cmd|call_api|check\",\"target\":\"文件路径或命令\",\"detail\":\"具体做什么\",\"depends_on\":null}]"
+    "【光湖术语解释】",
+    "  铸渊 = AI开发Agent，运行在Linux服务器上，可执行Shell命令和写文件",
+    "  ice-core = 冰核域，冰朔主权的核心域",
+    "  BS-SG-002 = 新加坡服务器，代码face节点",
+    "  HLDP = 光湖开发协议，所有模块代码必须包含@guardian @sovereign标签",
+    "  光湖语言世界 = AI原生操作系统，铸渊是其开发Agent",
+    "",
+    "【规则 · 必须遵守】",
+    "1. substep的target字段必须等于原任务的target字段（完整路径/完整命令），不得修改、不得用占位符替代",
+    "2. substep的action字段必须等于原任务的action字段",
+    "3. 如果原任务为 write_file 且提供了content，substep只需用相同的content",
+    "4. 如果原任务为 run_cmd，substep的target必须是具体的Shell命令（如 'ls -la /opt'），不能是占位符如'待指定'或'无'",
+    "5. 如果原任务信息已经足够具体，只返回一个substep，内容与原任务完全一致",
+    "6. 返回纯JSON数组，不要任何解释文字",
+    "",
+    '格式: [{"step":1,"action":"write_file|run_cmd|call_api|check","target":"文件路径或命令","detail":"具体做什么","content":"写入内容","depends_on":null}]',
+    "",
+    "现在请补全以上原任务的详细步骤："
   ].join("\n");
 
   try {
-    const raw = await modelCaller(prompt);
+    const raw = await modelCaller(taskInfo);
     const jsonMatch = raw.match(/\[[\s\S]*\]/);
     if (jsonMatch) {
-      const substeps = JSON.parse(jsonMatch[0]);
+      let substeps = JSON.parse(jsonMatch[0]);
+      substeps = substeps.map(ss => {
+        if (!ss.target || ss.target === "待指定文件路径" || ss.target === "待指定" || ss.target === "无" || ss.target === "未指定") {
+          console.warn("[Expander] LLM返回模糊target='" + ss.target + "', 回退到原任务target='" + task.target + "'");
+          ss.target = task.target || "";
+        }
+        if ((!ss.content || ss.content === "未指定") && task.content) {
+          ss.content = task.content;
+        }
+        if (!ss.detail || ss.detail.includes("需要补充") || ss.detail.includes("待指定")) {
+          ss.detail = task.description || ("执行: " + ss.action + " -> " + (ss.target || "?"));
+        }
+        return ss;
+      });
       return { ...task, substeps, expanded: true, expanded_at: new Date().toISOString() };
     }
   } catch (e) {
     console.warn("[Expander] LLM补全失败，退化为单步模式: " + e.message);
   }
 
-  // 退化：LLM不可用时，原任务本身就是一个步骤
   return {
     ...task,
-    substeps: [{ step: 1, action: task.action, target: task.target || "", detail: task.description || "" }],
+    substeps: [{
+      step: 1,
+      action: task.action || "unknown",
+      target: task.target || "",
+      content: task.content || "",
+      detail: task.description || ("执行: " + (task.action || "?") + " -> " + (task.target || "?")),
+      depends_on: null
+    }],
     expanded: false,
     expanded_at: new Date().toISOString()
   };
 }
 
-/**
- * 完整规划补全
- * @param {object} plan - 完整规划 { plan_id, tasks[], ... }
- * @param {function} modelCaller - LLM 调用函数
- * @param {string} plansDir - 规划文件存储目录
- */
 async function expandPlan(plan, modelCaller, plansDir) {
-  console.log("[Expander] 开始补全规划: " + plan.plan_id + " · " + plan.tasks.length + "个任务");
-
+  console.log("[Expander v2] 开始补全规划: " + plan.plan_id + " · " + (plan.tasks?.length || 0) + "个任务");
   const expandedTasks = [];
-  for (let i = 0; i < plan.tasks.length; i++) {
+  for (let i = 0; i < (plan.tasks || []).length; i++) {
     const task = plan.tasks[i];
-    console.log("[Expander] 补全 " + (i + 1) + "/" + plan.tasks.length + ": " + task.id);
+    console.log("[Expander v2] 补全 " + (i + 1) + "/" + plan.tasks.length + ": " + task.id);
     const expanded = await expandTask(task, modelCaller);
     expandedTasks.push(expanded);
   }
-
   plan.tasks = expandedTasks;
   plan._status = "expanded";
   plan._expanded_at = new Date().toISOString();
-
-  // 写回规划文件
   const planFile = path.join(plansDir, plan.plan_id + ".json");
   fs.writeFileSync(planFile, JSON.stringify(plan, null, 2), "utf-8");
-
-  console.log("[Expander] 补全完成 · 共 " + expandedTasks.reduce((s, t) => s + (t.substeps?.length || 0), 0) + " 个子步骤");
+  const totalSubsteps = expandedTasks.reduce((s, t) => s + (t.substeps?.length || 0), 0);
+  console.log("[Expander v2] 补全完成 · 共 " + totalSubsteps + " 个子步骤");
   return plan;
 }