diff --git a/pipeline_service/sdlc_ability.py b/pipeline_service/sdlc_ability.py
index 6fd5793..15870ba 100644
--- a/pipeline_service/sdlc_ability.py
+++ b/pipeline_service/sdlc_ability.py
@@ -217,6 +217,11 @@ SDL_PROMPT = """你是「开发产线」的驾驶舱 agent，负责软件项目
 - 除功能需求外，需求阶段还要明确部署环境需求：测试环境 + 生产环境的资源规格、依赖服务、端口、环境变量、高可用/备份/域名证书/监控告警等。
 - 用户未提供时，在会话中逐步追问明确（用 ask_user / list_questions），不要跳过；明确后再创建 requirement 任务。
 
+## 迭代粒度（重要）
+- 每个迭代有自己独立的初始任务（requirement 需求分析）和完整任务链（requirement → design → develop → deploy_test → test → deploy_prod）。
+- 创建 requirement 任务前，必须先 list_tasks 查「当前迭代」是否已有需求分析任务；同一迭代已有（含 approved/completed）→ 更新现有需求或推进下一阶段，禁止重复创建；只有新迭代（start_next_iteration）才需要新的初始任务。
+- 诊断卡点、派发任务、推进项目都以「当前迭代」为单位（diagnose_project 已按当前迭代统计），不要混着整个项目看。
+
 ## 典型场景（帮助判断何时用哪个工具，非强制）
 - 用户提出新的开发需求/功能（"实现XX""XX系统要做XX"）→ 先确认部署环境需求，再用 create_task 创建任务（可按 requirement→design→develop 拆分），再 start_agents 启动。**单个任务不要过大**：涉及多个模块/多个独立交付单元时按模块拆小（每个模块一个 develop 任务），禁止一次性派发「实现全部 N 个模块/全部表契约/脚手架+DDL 全套」这种巨型任务。**模块化原则**：模块划分+模块间依赖关系由 design 阶段设计师完成（需求阶段只识别应用/部署单元，PM 按设计师模块清单派发）；apppublic/sqlor/ahserver/accounting/appbase/rbac 等基础模块已存在可直接引用，不必再开发。
 - 用户问进展/状态 → list_tasks / diagnose_project / check_progress。
@@ -554,10 +559,18 @@ async def _h_diagnose_project(sor, p, ctx):
     if not pid:
         return "请先切换到项目"
 
+    # 迭代级诊断：诊断单位是「当前迭代」(in_progress)，而非整个项目。
+    # 项目只是迭代的容器；PM 诊断卡点、派发任务、推进都以当前迭代为单位。
+    from .iteration_capability import get_current_iteration
+    cur_iter = await get_current_iteration(sor, pid)
+    iter_name = cur_iter.get('iteration_name', '') if cur_iter else ''
+    iter_cond = " AND JSON_UNQUOTE(JSON_EXTRACT(params,'$.iteration_id'))=${nm}$ " if iter_name else ""
+    iter_bind = {"nm": iter_name} if iter_name else {}
+
     async def _cnt(state):
         recs = await sor.sqlExe(
-            "SELECT COUNT(*) as c FROM pipeline_tasks WHERE tenant_id=${pid}$ AND state=${st}$",
-            {"pid": pid, "st": state})
+            "SELECT COUNT(*) as c FROM pipeline_tasks WHERE tenant_id=${pid}$ AND state=${st}$" + iter_cond,
+            {"pid": pid, "st": state, **iter_bind})
         return getattr(recs[0], "c", 0) if recs else 0
 
     sc = await _cnt("submitted")
@@ -575,7 +588,7 @@ async def _h_diagnose_project(sor, p, ctx):
         pass
 
     lines = [
-        "项目诊断:",
+        f"项目诊断（当前迭代：{iter_name or '无 in_progress 迭代(回退项目级)'}）:",
         f"- 待执行: {sc}",
         f"- 运行中: {rc}",
         f"- 待审核: {vc}",
@@ -585,15 +598,15 @@ async def _h_diagnose_project(sor, p, ctx):
 
     if fc:
         failed_rows = await sor.sqlExe(
-            "SELECT id, title, role FROM pipeline_tasks WHERE tenant_id=${pid}$ AND state='failed' ORDER BY created_at DESC LIMIT 10",
-            {"pid": pid})
+            "SELECT id, title, role FROM pipeline_tasks WHERE tenant_id=${pid}$ AND state='failed'" + iter_cond + "ORDER BY created_at DESC LIMIT 10",
+            {"pid": pid, **iter_bind})
         lines.append("失败任务:")
         for fr in (failed_rows or []):
             lines.append(f"  - [{getattr(fr, 'role', '?')}] {getattr(fr, 'title', '')} (id={getattr(fr, 'id', '')})")
     if sc:
         submitted_rows = await sor.sqlExe(
-            "SELECT id, title, role FROM pipeline_tasks WHERE tenant_id=${pid}$ AND state='submitted' ORDER BY created_at ASC LIMIT 10",
-            {"pid": pid})
+            "SELECT id, title, role FROM pipeline_tasks WHERE tenant_id=${pid}$ AND state='submitted'" + iter_cond + "ORDER BY created_at ASC LIMIT 10",
+            {"pid": pid, **iter_bind})
         lines.append("待执行任务:")
         for sr in (submitted_rows or []):
             lines.append(f"  - [{getattr(sr, 'role', '?')}] {getattr(sr, 'title', '')} (id={getattr(sr, 'id', '')})")
@@ -601,9 +614,8 @@ async def _h_diagnose_project(sor, p, ctx):
         running_rows = await sor.sqlExe(
             "SELECT id, title, role, claimed_by, "
             "TIMESTAMPDIFF(MINUTE, updated_at, NOW()) AS mins "
-            "FROM pipeline_tasks WHERE tenant_id=${pid}$ AND state='running' "
-            "ORDER BY updated_at ASC LIMIT 10",
-            {"pid": pid})
+            "FROM pipeline_tasks WHERE tenant_id=${pid}$ AND state='running'" + iter_cond + "ORDER BY updated_at ASC LIMIT 10",
+            {"pid": pid, **iter_bind})
         lines.append("运行中任务:")
         for rr in (running_rows or []):
             rrole = getattr(rr, 'role', '?')