diff --git a/docs/design/12-pipeline-design.md b/docs/design/12-pipeline-design.md
index 68be849..035db20 100644
--- a/docs/design/12-pipeline-design.md
+++ b/docs/design/12-pipeline-design.md
@@ -55,8 +55,16 @@ class PipelineStage:
     on_success: str                  # 成功后转到哪个 stage（或 "done"/"failed"）
     on_failure: str                  # 失败后转到哪个 stage（可指向前序 stage = 退回）
     max_retries: int = 0             # 最大退回重试次数（0=不限，超限转 failed）
-    retry_count: int = 0             # 当前已重试次数（运行时状态）
     is_terminal: bool = False        # 是否终态
+    # 注意：retry_count 不在此处——PipelineStage 是注册表模板，所有同类型任务共享
+
+@dataclass
+class TaskPipelineState:
+    """每个任务的 Pipeline 运行时状态（存在 DB 或内存）"""
+    task_id: str
+    pipeline_type: str
+    current_stage: str
+    stage_retry_counts: Dict[str, int] = field(default_factory=dict)  # stage_id → 已重试次数
 
 @dataclass
 class Pipeline:
@@ -73,7 +81,7 @@ class Pipeline:
 | **顺序** | `on_success` 指向下一个 stage | working → review → done |
 | **条件分支** | `on_success` / `on_failure` 指向不同 stage | review 通过→done，review 拒绝→working |
 | **退回** | `on_failure` 指向前序 stage | review 失败退回 working |
-| **循环/重试** | `max_retries` + `retry_count` 控制 | review 最多退回 3 次，超过转 failed |
+| **循环/重试** | `max_retries` + `TaskPipelineState.stage_retry_counts` 控制 | review 最多退回 3 次，超过转 failed |
 
 异常状态（failed/paused/escalated/blocked/cancelled）不受 Pipeline 约束，任何 stage 都可以转入异常状态。
 
@@ -313,26 +321,50 @@ class PipelineEngine:
         self.registry = registry
         self.bb = bb
     
-    def get_next_stage(self, task: Task, outcome: str) -> Optional[str]:
-        """根据当前 stage 和执行结果，计算下一个 stage"""
-        pipeline = self.registry.get(task.task_type)
+    def get_or_create_state(self, task: Task) -> Optional[TaskPipelineState]:
+        """获取或创建任务的 Pipeline 运行时状态"""
+        pipeline_type = task.task_type
+        if not pipeline_type:
+            return None  # 自由模式
+        pipeline = self.registry.get(pipeline_type)
+        if not pipeline:
+            return None
+        # 从 DB 或内存加载；首次则初始化
+        state = self.bb.get_pipeline_state(task.id)
+        if not state:
+            state = TaskPipelineState(
+                task_id=task.id,
+                pipeline_type=pipeline_type,
+                current_stage=pipeline.entry,
+            )
+            self.bb.save_pipeline_state(state)
+        return state
+    
+    def get_next_stage(self, task_state: TaskPipelineState, outcome: str) -> Optional[str]:
+        """根据当前 stage 和执行结果，计算下一个 stage（纯函数，不修改 PipelineStage 模板）"""
+        pipeline = self.registry.get(task_state.pipeline_type)
         if not pipeline:
             return None  # 自由模式，不干预
         
-        current_stage = pipeline.stages.get(task.status)
+        current_stage = pipeline.stages.get(task_state.current_stage)
         if not current_stage:
             return None
         
         if outcome == "success":
+            task_state.current_stage = current_stage.on_success
             return current_stage.on_success
         elif outcome == "failure":
             # 检查 max_retries
             if current_stage.max_retries > 0:
-                current_stage.retry_count += 1
-                if current_stage.retry_count > current_stage.max_retries:
+                retries = task_state.stage_retry_counts.get(current_stage.id, 0)
+                retries += 1
+                task_state.stage_retry_counts[current_stage.id] = retries
+                if retries > current_stage.max_retries:
+                    task_state.current_stage = "failed"
                     return "failed"
             return current_stage.on_failure
         else:
+            task_state.current_stage = "failed"
             return "failed"
     
     def get_agent_for_stage(self, task: Task) -> Optional[str]:
@@ -369,11 +401,15 @@ class PipelineEngine:
 
 ### §8.3 Ticker 集成
 
+**TaskPipelineState 存储方案**：建议在 tasks 表新增 `pipeline_state_json` 列（TEXT），序列化存储 `TaskPipelineState`。Blackboard 的 `get_pipeline_state()` / `save_pipeline_state()` 负责序列化/反序列化。独立表方案亦可，但鉴于 state 结构简单且与任务 1:1，放 tasks 列更方便。
+
 Ticker 在任务状态流转时调用 PipelineEngine：
-1. 任务完成（spawn 返回）→ `classify_outcome()` 判定 success/failure
-2. `PipelineEngine.get_next_stage()` 计算下一个 stage
-3. 如果下一个 stage 不是终态 → 更新任务状态 + `PipelineEngine.get_agent_for_stage()` 获取下一个 Agent → spawn
-4. 如果是终态 → 标记 done/failed
+1. `PipelineEngine.get_or_create_state(task)` 获取运行时状态
+2. 任务完成（spawn 返回）→ `classify_outcome()` 判定 success/failure
+3. `PipelineEngine.get_next_stage(task_state, outcome)` 计算下一个 stage
+4. 持久化更新 `TaskPipelineState`（`bb.save_pipeline_state(task_state)`）
+5. 如果下一个 stage 不是终态 → 更新任务状态 + `PipelineEngine.get_agent_for_stage()` 获取下一个 Agent → spawn
+6. 如果是终态 → 标记 done/failed
 
 ## §9 实施路线