diff --git a/pipeline_service/sdlc_ability.py b/pipeline_service/sdlc_ability.py
index 9fbe110..6fd5793 100644
--- a/pipeline_service/sdlc_ability.py
+++ b/pipeline_service/sdlc_ability.py
@@ -407,6 +407,22 @@ async def _h_create_task(sor, p, ctx):
     if iteration_name:
         params["iteration_id"] = iteration_name
 
+    # 迭代级查重：requirement 是迭代的初始任务，同一迭代内不应重复创建。
+    # 判断粒度是「迭代」而非「项目」——每个迭代有自己独立的初始任务与任务链，
+    # 新迭代（start_next_iteration）才需要新的需求分析任务。
+    if role == 'agent.requirement' and iteration_name:
+        dup_recs = await sor.sqlExe(
+            "SELECT id, title, state FROM pipeline_tasks WHERE tenant_id=${pid}$ "
+            "AND role='agent.requirement' AND state != 'cancelled' "
+            "AND JSON_UNQUOTE(JSON_EXTRACT(params,'$.iteration_id'))=${nm}$ LIMIT 1",
+            {"pid": pid, "nm": iteration_name})
+        if dup_recs:
+            d = dup_recs[0]
+            return (f"⚠️ 拒绝创建：当前迭代「{iteration_name}」已有需求分析任务「{getattr(d, 'title', '')}」"
+                    f"（状态 {getattr(d, 'state', '')}）。需求分析是迭代的初始任务，同一迭代不应重复创建；"
+                    f"请 list_tasks 查看现状、更新现有需求或推进下一阶段；"
+                    f"若要新起一轮需求分析，请用 start_next_iteration 开启新迭代。")
+
     await sor.C("pipeline_tasks", {
         "id": tid, "tenant_id": pid, "pipeline_id": "role_task",
         "owner_id": ctx.get("user_id") or "user", "title": title,