From 6a649aba07647b97e4c46049bf1d079327bbb0d7 Mon Sep 17 00:00:00 2001
From: cfdaily <cfdaily@gitee.com>
Date: Fri, 5 Jun 2026 11:03:30 +0800
Subject: [PATCH] auto-sync: 2026-06-05 11:03:30

---
 tests/conftest.py                             |  51 ++
 tests/e2e/__init__.py                         |   0
 tests/{ => e2e}/test_e2e_v27.py               |   4 +
 tests/{ => e2e}/test_e2e_v31.py               |   4 +
 .../test_four_phase.py}                       |   4 +
 tests/integration/__init__.py                 |   0
 tests/{ => integration}/test_api.py           |   2 +
 .../test_dispatcher_integration.py            | 128 ++++
 tests/{ => integration}/test_main.py          |   2 +
 .../test_review_integration.py}               |  69 +-
 tests/integration/test_spawner_integration.py |  73 ++
 tests/{ => integration}/test_sse.py           |   2 +
 .../test_ticker_integration.py}               |  39 +-
 tests/{ => integration}/test_v27_subtasks.py  |   2 +
 tests/test_bootstrap.py                       | 212 ------
 tests/test_e2e_four_phase.py                  | 629 ------------------
 tests/unit/__init__.py                        |   0
 tests/{ => unit}/test_blackboard.py           |   2 +
 tests/unit/test_bootstrap.py                  | 191 ++++++
 tests/{ => unit}/test_cli.py                  |   2 +
 tests/{ => unit}/test_counter.py              |   2 +
 tests/{ => unit}/test_dispatcher.py           | 173 +----
 tests/{ => unit}/test_experience.py           |   2 +
 tests/{ => unit}/test_health.py               |   2 +
 tests/{ => unit}/test_inbox.py                |   2 +
 tests/{ => unit}/test_registry.py             |   2 +
 tests/unit/test_review.py                     |  87 +++
 tests/{ => unit}/test_router.py               |   2 +
 tests/{ => unit}/test_skill_system.py         |   2 +
 tests/{ => unit}/test_spawner.py              | 188 +-----
 30 files changed, 602 insertions(+), 1276 deletions(-)
 create mode 100644 tests/conftest.py
 create mode 100644 tests/e2e/__init__.py
 rename tests/{ => e2e}/test_e2e_v27.py (99%)
 rename tests/{ => e2e}/test_e2e_v31.py (99%)
 rename tests/{test_four_phase_unit.py => e2e/test_four_phase.py} (99%)
 create mode 100644 tests/integration/__init__.py
 rename tests/{ => integration}/test_api.py (99%)
 create mode 100644 tests/integration/test_dispatcher_integration.py
 rename tests/{ => integration}/test_main.py (98%)
 rename tests/{test_review.py => integration/test_review_integration.py} (71%)
 create mode 100644 tests/integration/test_spawner_integration.py
 rename tests/{ => integration}/test_sse.py (99%)
 rename tests/{test_ticker.py => integration/test_ticker_integration.py} (92%)
 rename tests/{ => integration}/test_v27_subtasks.py (99%)
 delete mode 100644 tests/test_bootstrap.py
 delete mode 100644 tests/test_e2e_four_phase.py
 create mode 100644 tests/unit/__init__.py
 rename tests/{ => unit}/test_blackboard.py (99%)
 create mode 100644 tests/unit/test_bootstrap.py
 rename tests/{ => unit}/test_cli.py (99%)
 rename tests/{ => unit}/test_counter.py (99%)
 rename tests/{ => unit}/test_dispatcher.py (62%)
 rename tests/{ => unit}/test_experience.py (99%)
 rename tests/{ => unit}/test_health.py (99%)
 rename tests/{ => unit}/test_inbox.py (99%)
 rename tests/{ => unit}/test_registry.py (99%)
 create mode 100644 tests/unit/test_review.py
 rename tests/{ => unit}/test_router.py (99%)
 rename tests/{ => unit}/test_skill_system.py (99%)
 rename tests/{ => unit}/test_spawner.py (68%)

diff --git a/tests/conftest.py b/tests/conftest.py
new file mode 100644
index 0000000..e7d2f76
--- /dev/null
+++ b/tests/conftest.py
@@ -0,0 +1,51 @@
+import uuid
+import pytest
+from pathlib import Path
+from fastapi.testclient import TestClient
+
+
+def pytest_configure(config):
+    markers = {
+        "unit": "单元测试：纯逻辑，mock 外部依赖",
+        "integration": "集成测试：API 端点 + 真实/临时 DB",
+        "e2e": "端到端测试：真实 daemon + Agent（手动触发）",
+        "slow": "慢测试（>5s）",
+        "broadcast": "广播认领相关",
+        "mail": "邮件系统相关",
+        "state_machine": "状态机转换",
+        "classify": "Classify Outcome 相关",
+        "review": "审查/Rebuttal 相关",
+    }
+    for name, desc in markers.items():
+        config.addinivalue_line("markers", f"{name}: {desc}")
+
+
+@pytest.fixture
+def isolated_data_root(tmp_path):
+    """隔离的 data_root，测试结束自动清理"""
+    data_root = tmp_path / "test_data"
+    data_root.mkdir()
+    return data_root
+
+
+@pytest.fixture
+def isolated_registry(isolated_data_root):
+    """隔离的 registry.db"""
+    from src.blackboard.registry import ProjectRegistry
+    registry = ProjectRegistry(isolated_data_root)
+    return registry
+
+
+@pytest.fixture
+def client_with_isolation(isolated_data_root):
+    """带数据隔离的 TestClient"""
+    import src.utils as utils
+    original = utils.get_data_root
+    utils.get_data_root = lambda: isolated_data_root
+
+    from src.main import app
+    client = TestClient(app)
+
+    yield client
+
+    utils.get_data_root = original
diff --git a/tests/e2e/__init__.py b/tests/e2e/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/test_e2e_v27.py b/tests/e2e/test_e2e_v27.py
similarity index 99%
rename from tests/test_e2e_v27.py
rename to tests/e2e/test_e2e_v27.py
index c686206..c9f3128 100644
--- a/tests/test_e2e_v27.py
+++ b/tests/e2e/test_e2e_v27.py
@@ -1,3 +1,7 @@
+import pytest
+
+pytestmark = pytest.mark.e2e
+
 """v2.7 端到端测试 — 全链路真实环境
 
 覆盖：项目管理 → Task CRUD → SubTask → Stage进度 → 状态聚合 → 依赖链 → 超时 → Mail → 真实Agent调度
diff --git a/tests/test_e2e_v31.py b/tests/e2e/test_e2e_v31.py
similarity index 99%
rename from tests/test_e2e_v31.py
rename to tests/e2e/test_e2e_v31.py
index f05ec11..47efaba 100644
--- a/tests/test_e2e_v31.py
+++ b/tests/e2e/test_e2e_v31.py
@@ -1,3 +1,7 @@
+import pytest
+
+pytestmark = pytest.mark.e2e
+
 """v3.1 端到端测试 — 新增场景覆盖
 
 覆盖 v3.1 新增功能：
diff --git a/tests/test_four_phase_unit.py b/tests/e2e/test_four_phase.py
similarity index 99%
rename from tests/test_four_phase_unit.py
rename to tests/e2e/test_four_phase.py
index eddbb54..f2173a2 100644
--- a/tests/test_four_phase_unit.py
+++ b/tests/e2e/test_four_phase.py
@@ -1,3 +1,7 @@
+import pytest
+
+pytestmark = pytest.mark.e2e
+
 """#01 四相循环 单元测试
 
 不依赖 daemon / Agent，纯逻辑验证。覆盖：
diff --git a/tests/integration/__init__.py b/tests/integration/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/test_api.py b/tests/integration/test_api.py
similarity index 99%
rename from tests/test_api.py
rename to tests/integration/test_api.py
index e7f0af7..65045c0 100644
--- a/tests/test_api.py
+++ b/tests/integration/test_api.py
@@ -12,6 +12,8 @@ from src.blackboard.models import Task
 from src.blackboard.registry import ProjectRegistry
 from src.main import app
 
+pytestmark = pytest.mark.integration
+
 
 @pytest.fixture
 def project_env(tmp_path):
diff --git a/tests/integration/test_dispatcher_integration.py b/tests/integration/test_dispatcher_integration.py
new file mode 100644
index 0000000..ee17fe3
--- /dev/null
+++ b/tests/integration/test_dispatcher_integration.py
@@ -0,0 +1,128 @@
+"""F9 Agent 调度器集成测试 — rollback/on_complete DB 交互"""
+
+import asyncio
+import json
+import pytest
+from pathlib import Path
+
+from src.blackboard.models import Task
+from src.blackboard.operations import Blackboard
+from src.daemon.dispatcher import Dispatcher
+
+pytestmark = pytest.mark.integration
+
+
+# ---------------------------------------------------------------------------
+# 司马懿评审补充：_rollback_current_agent + on_complete 统一（v2.8 #07.2）
+# ---------------------------------------------------------------------------
+
+class TestRollbackAndOnComplete:
+    """司马懿评审遗漏 #3 + #4: crash 后 current_agent 回退 + on_complete 统一路径"""
+
+    def test_rollback_current_agent_on_crash(self, tmp_path):
+        """executor crash → _rollback_current_agent 回退 current_agent → assignee"""
+        db_path = tmp_path / "blackboard.db"
+        bb = Blackboard(db_path)
+        bb.create_task(Task(
+            id="t1", title="T", status="working",
+            assigned_by="daemon", assignee="zhangfei-dev",
+        ))
+
+        conn = bb._conn()
+        try:
+            conn.execute("UPDATE tasks SET current_agent=? WHERE id=?", ("zhangfei-dev", "t1"))
+            conn.commit()
+        finally:
+            conn.close()
+
+        dispatcher = Dispatcher(registered_agents=["zhangfei-dev"])
+        dispatcher._rollback_current_agent(db_path, "t1", "zhangfei-dev")
+
+        import sqlite3
+        conn2 = sqlite3.connect(str(db_path))
+        conn2.row_factory = sqlite3.Row
+        row = conn2.execute("SELECT current_agent, assignee FROM tasks WHERE id=?", ("t1",)).fetchone()
+        conn2.close()
+
+        assert row["current_agent"] == row["assignee"] == "zhangfei-dev"
+
+    def test_rollback_different_agent(self, tmp_path):
+        """current_agent ≠ agent_id → 不回退（安全检查）"""
+        db_path = tmp_path / "blackboard.db"
+        bb = Blackboard(db_path)
+        bb.create_task(Task(
+            id="t1", title="T", status="working",
+            assigned_by="daemon", assignee="zhangfei-dev",
+        ))
+
+        conn = bb._conn()
+        try:
+            conn.execute("UPDATE tasks SET current_agent=? WHERE id=?", ("simayi-challenger", "t1"))
+            conn.commit()
+        finally:
+            conn.close()
+
+        dispatcher = Dispatcher(registered_agents=["zhangfei-dev"])
+        dispatcher._rollback_current_agent(db_path, "t1", "wrong-agent")
+
+        import sqlite3
+        conn2 = sqlite3.connect(str(db_path))
+        conn2.row_factory = sqlite3.Row
+        row = conn2.execute("SELECT current_agent FROM tasks WHERE id=?", ("t1",)).fetchone()
+        conn2.close()
+
+        assert row["current_agent"] == "simayi-challenger"
+
+    def test_on_complete_crash_rollback_executor(self, tmp_path):
+        """executor crash → rollback current_agent + _task_auto_complete（标 review）"""
+        db_path = tmp_path / "blackboard.db"
+        bb = Blackboard(db_path)
+        bb.create_task(Task(
+            id="t1", title="T", status="working",
+            assigned_by="daemon", assignee="zhangfei-dev",
+        ))
+
+        conn = bb._conn()
+        try:
+            conn.execute("UPDATE tasks SET current_agent=? WHERE id=?", ("zhangfei-dev", "t1"))
+            conn.commit()
+        finally:
+            conn.close()
+
+        dispatcher = Dispatcher(registered_agents=["zhangfei-dev"])
+        dispatcher._rollback_current_agent(db_path, "t1", "zhangfei-dev")
+
+        import sqlite3
+        conn2 = sqlite3.connect(str(db_path))
+        conn2.row_factory = sqlite3.Row
+        row = conn2.execute("SELECT current_agent FROM tasks WHERE id=?", ("t1",)).fetchone()
+        conn2.close()
+        assert row["current_agent"] == "zhangfei-dev"
+
+    def test_on_complete_crash_rollback_review(self, tmp_path):
+        """review crash → rollback current_agent + 保持 review 状态"""
+        db_path = tmp_path / "blackboard.db"
+        bb = Blackboard(db_path)
+        bb.create_task(Task(
+            id="t1", title="T", status="review",
+            assigned_by="daemon", assignee="zhangfei-dev",
+        ))
+
+        conn = bb._conn()
+        try:
+            conn.execute("UPDATE tasks SET current_agent=? WHERE id=?", ("simayi-challenger", "t1"))
+            conn.commit()
+        finally:
+            conn.close()
+
+        dispatcher = Dispatcher(registered_agents=["simayi-challenger"])
+        dispatcher._rollback_current_agent(db_path, "t1", "simayi-challenger")
+
+        import sqlite3
+        conn2 = sqlite3.connect(str(db_path))
+        conn2.row_factory = sqlite3.Row
+        row = conn2.execute("SELECT current_agent, status FROM tasks WHERE id=?", ("t1",)).fetchone()
+        conn2.close()
+
+        assert row["current_agent"] == "zhangfei-dev"
+        assert row["status"] == "review"
diff --git a/tests/test_main.py b/tests/integration/test_main.py
similarity index 98%
rename from tests/test_main.py
rename to tests/integration/test_main.py
index a1ebdfb..0b49061 100644
--- a/tests/test_main.py
+++ b/tests/integration/test_main.py
@@ -5,6 +5,8 @@ from fastapi.testclient import TestClient
 
 from src.main import app, config, load_config
 
+pytestmark = pytest.mark.integration
+
 
 @pytest.fixture
 def client():
diff --git a/tests/test_review.py b/tests/integration/test_review_integration.py
similarity index 71%
rename from tests/test_review.py
rename to tests/integration/test_review_integration.py
index 48b9ad2..5b3db9e 100644
--- a/tests/test_review.py
+++ b/tests/integration/test_review_integration.py
@@ -1,11 +1,9 @@
-"""F12 Review Pipeline + F13 Guardrail + F14 Rebuttal 单元测试
+"""F12 Review Pipeline + F13 Guardrail 集成测试
 
-按 test-plan-v2.6.md §F12-F14：
+按 test-plan-v2.6.md §F12-F13：
 - F12 T1: 验证流水线四步（P0）
 - F12 T2: 评分计算（P0）
 - F13 T1: Guardrail 门控（P0）
-- F14 T1: 反驳权流程（P0）
-- F14 T2: 最大轮次限制（P0）
 """
 
 import json
@@ -16,12 +14,13 @@ from unittest.mock import MagicMock
 from src.blackboard.models import Task
 from src.blackboard.operations import Blackboard
 from src.daemon.review import (
-    RebuttalManager,
     ReviewPipeline,
     ReviewResult,
     ReviewVerdict,
 )
 
+pytestmark = pytest.mark.integration
+
 
 @pytest.fixture
 def db_path(tmp_path):
@@ -177,63 +176,3 @@ class TestGuardrail:
         outputs = [{"content": "valid output here", "type": "text"}]
         result = p.run_review(task, outputs=outputs)
         assert result["gate"] == "optional"
-
-
-# ---------------------------------------------------------------------------
-# F14 T1: 反驳权流程
-# ---------------------------------------------------------------------------
-
-class TestRebuttal:
-    def test_submit_rebuttal_accepted(self, bb):
-        task = Task(id="t1", title="T", status="pending", assigned_by="d")
-        bb.create_task(task)
-
-        rm = RebuttalManager(bb=bb)
-        result = rm.submit_rebuttal("t1", "agent-1", "I disagree with the review")
-        assert result["status"] == "accepted"
-        assert result["round"] == 1
-        assert result["escalation_target"] == "simayi-challenger"
-
-    def test_second_round_escalates_to_pangtong(self, bb):
-        task = Task(id="t1", title="T", status="pending", assigned_by="d")
-        bb.create_task(task)
-
-        rm = RebuttalManager(bb=bb)
-        rm.submit_rebuttal("t1", "agent-1", "Round 1")
-        result = rm.submit_rebuttal("t1", "agent-1", "Round 2")
-        assert result["status"] == "accepted"
-        assert result["round"] == 2
-        assert result["escalation_target"] == "pangtong-fujunshi"
-
-
-# ---------------------------------------------------------------------------
-# F14 T2: 最大轮次限制
-# ---------------------------------------------------------------------------
-
-class TestRebuttalLimits:
-    def test_max_rounds_rejected(self, bb):
-        task = Task(id="t1", title="T", status="pending", assigned_by="d")
-        bb.create_task(task)
-
-        rm = RebuttalManager(bb=bb)
-        rm.submit_rebuttal("t1", "a", "R1")
-        rm.submit_rebuttal("t1", "a", "R2")
-        result = rm.submit_rebuttal("t1", "a", "R3")
-        assert result["status"] == "rejected"
-        assert "Max" in result["reason"]
-
-    def test_rebuttal_without_bb(self):
-        rm = RebuttalManager(bb=None)
-        result = rm.submit_rebuttal("t1", "a", "reason")
-        assert result["status"] == "accepted"
-        assert result["round"] == 1
-
-    def test_rebuttal_observation_recorded(self, bb):
-        task = Task(id="t1", title="T", status="pending", assigned_by="d")
-        bb.create_task(task)
-
-        rm = RebuttalManager(bb=bb)
-        rm.submit_rebuttal("t1", "agent-1", "test reason", evidence="file.txt")
-        obs = bb.get_observations(task_id="t1")
-        rebuttals = [o for o in obs if "Rebuttal round" in (o.body or "")]
-        assert len(rebuttals) == 1
diff --git a/tests/integration/test_spawner_integration.py b/tests/integration/test_spawner_integration.py
new file mode 100644
index 0000000..2d6bae1
--- /dev/null
+++ b/tests/integration/test_spawner_integration.py
@@ -0,0 +1,73 @@
+"""F9 Agent Spawner 集成测试 — 超时/失败/spawn 真实流程"""
+
+import asyncio
+import pytest
+from pathlib import Path
+
+from src.blackboard.operations import Blackboard
+from src.daemon.spawner import AgentSpawner
+
+pytestmark = pytest.mark.integration
+
+
+@pytest.fixture
+def db_path(tmp_path):
+    return tmp_path / "blackboard.db"
+
+
+@pytest.fixture
+def bb(db_path):
+    return Blackboard(db_path)
+
+
+@pytest.fixture
+def real_spawner(db_path):
+    return AgentSpawner(db_path=db_path, dry_run=False, agent_timeout=2.0)
+
+
+# ---------------------------------------------------------------------------
+# T2: 超时处理
+# ---------------------------------------------------------------------------
+
+class TestTimeout:
+    def test_timeout_kills_process(self, tmp_path):
+        """超时后 kill 进程"""
+        db_path = tmp_path / "blackboard.db"
+        Blackboard(db_path)  # init
+        spawner = AgentSpawner(db_path=db_path, dry_run=False, agent_timeout=0.5)
+
+        # Spawn a long-running process (sleep 10)
+        session_id = asyncio.run(
+            spawner.spawn_full_agent(
+                "test-agent",
+                "sleep 10",
+                task_id=None,
+            )
+        )
+        # Wait for timeout
+        asyncio.run(asyncio.sleep(1.0))
+
+        session = spawner.get_session(session_id)
+        if session:
+            assert session["status"] in ("timed_out", "running", "completed")
+
+
+# ---------------------------------------------------------------------------
+# T3: spawn 失败
+# ---------------------------------------------------------------------------
+
+class TestSpawnFailure:
+    def test_nonexistent_command(self, real_spawner, db_path, bb):
+        """命令不存在 → spawn_failed"""
+        bb.create_task(
+            __import__("src.blackboard.models", fromlist=["Task"]).Task(
+                id="t1", title="T", status="pending", assigned_by="d"
+            )
+        )
+
+        try:
+            asyncio.run(
+                real_spawner.spawn_full_agent("test", "msg", task_id="t1")
+            )
+        except Exception:
+            pass  # Expected - command may fail
diff --git a/tests/test_sse.py b/tests/integration/test_sse.py
similarity index 99%
rename from tests/test_sse.py
rename to tests/integration/test_sse.py
index 34413b1..7403fef 100644
--- a/tests/test_sse.py
+++ b/tests/integration/test_sse.py
@@ -12,6 +12,8 @@ import json
 import pytest
 from unittest.mock import AsyncMock, MagicMock
 
+pytestmark = pytest.mark.integration
+
 from src.daemon.sse import (
     Hook,
     HookManager,
diff --git a/tests/test_ticker.py b/tests/integration/test_ticker_integration.py
similarity index 92%
rename from tests/test_ticker.py
rename to tests/integration/test_ticker_integration.py
index db2f7bb..d4822a9 100644
--- a/tests/test_ticker.py
+++ b/tests/integration/test_ticker_integration.py
@@ -24,6 +24,8 @@ from src.blackboard.registry import ProjectRegistry
 from src.blackboard.queries import Queries
 from src.daemon.ticker import Ticker
 
+pytestmark = pytest.mark.integration
+
 
 # ---------------------------------------------------------------------------
 # Fixtures
@@ -446,20 +448,15 @@ class TestCheckTimeoutsUnified:
         return registry, db_path, bb
 
     def test_crash_limit_working(self, timeout_project):
-        """E12.1: executor crash 3 次/30min → _check_timeouts 标 failed
+        """E12.1: executor crash 3 次/30min → _check_timeouts 标 failed"""
 
-        #07.2 将 crash_limit 从 _dispatch_reviews 移到 _check_timeouts，
-        覆盖 working 和 review 状态。
-        """
         registry, db_path, bb = timeout_project
 
-        # 创建 working 任务
         bb.create_task(Task(
             id="t-crash", title="Crash Task", status="working",
             assigned_by="daemon", current_agent="agent-a",
         ))
 
-        # 模拟 3 次 crash 的 task_attempts
         from datetime import datetime, timedelta
         conn = bb._conn()
         try:
@@ -476,20 +473,13 @@ class TestCheckTimeoutsUnified:
             conn.close()
 
         ticker = Ticker(registry, tick_interval=30)
-        # 如果有 dispatcher + _check_crash_limit，它会在 _check_timeouts 中触发
-        # 测试基本结构：_check_timeouts 应该能处理 working 状态任务
         result = ticker._check_timeouts(db_path)
-        # 即使没有 dispatcher（_check_crash_limit 需要），超时检查本身不应崩溃
         assert isinstance(result, list)
 
     def test_crash_limit_review(self, timeout_project):
-        """E12.2: reviewer crash 3 次/30min → _check_timeouts 标 failed
-
-        #07.2 统一后，review 状态的 crash_limit 也走 _check_timeouts。
-        """
+        """E12.2: reviewer crash 3 次/30min → _check_timeouts 标 failed"""
         registry, db_path, bb = timeout_project
 
-        # 创建 review 状态任务
         bb.create_task(Task(
             id="t-review-crash", title="Review Crash Task", status="review",
             assigned_by="daemon", current_agent="simayi-challenger",
@@ -498,32 +488,24 @@ class TestCheckTimeoutsUnified:
         ticker = Ticker(registry, tick_interval=30)
         result = ticker._check_timeouts(db_path)
         assert isinstance(result, list)
-        # _check_timeouts 不应崩溃，review 状态在统一逻辑中被正确处理
 
     def test_updated_at_fallback(self, timeout_project):
-        """E12.3: mail auto-working 无 started_at/claimed_at → updated_at fallback
-
-        #07.3 ACT-1: _check_timeouts 使用 updated_at 作为最后 fallback，
-        确保 PM2 重启后 mail 孤儿任务也能被回收。
-        """
+        """E12.3: mail auto-working 无 started_at/claimed_at → updated_at fallback"""
         registry, db_path, bb = timeout_project
 
         from datetime import datetime, timedelta
 
-        # 创建 working 任务，只有 updated_at（模拟 mail auto-working）
         old_time = (datetime.utcnow() - timedelta(minutes=60)).isoformat()
         bb.create_task(Task(
             id="t-mail-orphan", title="Mail Orphan", status="working",
             assigned_by="daemon", current_agent="pangtong-fujunshi",
         ))
-        # 手动设置 updated_at（模拟 PM2 重启前的时间戳）
         conn = bb._conn()
         try:
             conn.execute(
                 "UPDATE tasks SET updated_at = ? WHERE id = ?",
                 (old_time, "t-mail-orphan"),
             )
-            # 确保 started_at 和 claimed_at 为 NULL
             conn.execute(
                 "UPDATE tasks SET started_at = NULL, claimed_at = NULL WHERE id = ?",
                 ("t-mail-orphan",),
@@ -534,25 +516,18 @@ class TestCheckTimeoutsUnified:
 
         ticker = Ticker(registry, tick_interval=30, default_task_timeout_minutes=30)
         reclaimed = ticker._check_timeouts(db_path)
-        # updated_at fallback 应让这个任务被回收
         assert "t-mail-orphan" in reclaimed, \
             "Mail orphan with only updated_at should be reclaimed via fallback"
 
     def test_process_dead_keeps_review_status(self, timeout_project):
-        """E12.4: review agent 进程死 → 保持 review 状态（不推 pending）
-
-        #07.2: process_dead 对 review 状态的处理——保持 review，
-        等 _dispatch_reviews 下个 tick 自然 dispatch。
-        """
+        """E12.4: review agent 进程死 → 保持 review 状态（不推 pending）"""
         registry, db_path, bb = timeout_project
 
-        # 创建 review 状态任务
         bb.create_task(Task(
             id="t-review-dead", title="Review Dead Process", status="review",
             assigned_by="daemon", current_agent="simayi-challenger",
         ))
 
-        # 设置较新的时间戳（不应因超时被回收）
         from datetime import datetime
         conn = bb._conn()
         try:
@@ -567,6 +542,4 @@ class TestCheckTimeoutsUnified:
         ticker = Ticker(registry, tick_interval=30, default_task_timeout_minutes=30)
         reclaimed = ticker._check_timeouts(db_path)
 
-        # 没有 process_dead 的模拟（无 counter/spawner），纯超时路径
-        # review 任务时间戳较新 → 不应被超时回收
         assert "t-review-dead" not in reclaimed
diff --git a/tests/test_v27_subtasks.py b/tests/integration/test_v27_subtasks.py
similarity index 99%
rename from tests/test_v27_subtasks.py
rename to tests/integration/test_v27_subtasks.py
index e3e3e6e..24fd88f 100644
--- a/tests/test_v27_subtasks.py
+++ b/tests/integration/test_v27_subtasks.py
@@ -6,6 +6,8 @@ from pathlib import Path
 
 import pytest
 
+pytestmark = pytest.mark.integration
+
 from src.blackboard.db import init_db, get_connection
 from src.blackboard.models import Task
 from src.blackboard.operations import Blackboard
diff --git a/tests/test_bootstrap.py b/tests/test_bootstrap.py
deleted file mode 100644
index c12809f..0000000
--- a/tests/test_bootstrap.py
+++ /dev/null
@@ -1,212 +0,0 @@
-"""F11 Bootstrap 拼装单元测试
-
-按 test-plan-v2.6.md §F11：
-- T1: 各 role 拼装（P0）
-- T2: token 估算（P0）
-- T3: 缺失组件降级（P1）
-- T4: 模板变量替换（P1）
-"""
-
-import pytest
-from pathlib import Path
-
-from src.blackboard.models import Task
-from src.daemon.bootstrap import BootstrapBuilder, estimate_tokens
-
-
-@pytest.fixture
-def builder():
-    return BootstrapBuilder(max_tokens=4096)
-
-
-@pytest.fixture
-def builder_with_templates(tmp_path):
-    template_dir = tmp_path / "templates"
-    template_dir.mkdir()
-    (template_dir / "executor.md").write_text("# Executor Role\nYou execute tasks.")
-    (template_dir / "reviewer.md").write_text("# Reviewer Role\nYou review code.")
-    (template_dir / "planner.md").write_text("# Planner Role\nYou plan tasks.")
-    return BootstrapBuilder(template_dir=template_dir, max_tokens=4096)
-
-
-# ---------------------------------------------------------------------------
-# T1: 各 role 拼装
-# ---------------------------------------------------------------------------
-
-class TestRoleBootstrap:
-    def test_executor_bootstrap(self, builder):
-        b = builder.build(
-            role="executor",
-            task_context={"task_id": "t1", "title": "Write tests"},
-        )
-        assert "Write tests" in b
-        assert "t1" in b
-
-    def test_reviewer_bootstrap(self, builder):
-        b = builder.build(
-            role="reviewer",
-            task_context={"task_id": "t2", "title": "Review PR"},
-        )
-        assert "Review PR" in b
-
-    def test_planner_bootstrap(self, builder):
-        b = builder.build(
-            role="planner",
-            task_context={"task_id": "t3", "title": "Plan sprint"},
-        )
-        assert "Plan sprint" in b
-
-    def test_executor_with_guardrail(self, builder):
-        b = builder.build(
-            role="executor",
-            guardrail_rules="## Guardrail\nNo dangerous ops",
-        )
-        assert "Guardrail" in b
-
-    def test_reviewer_no_guardrail(self, builder):
-        b = builder.build(
-            role="reviewer",
-            guardrail_rules="## Guardrail\nNo dangerous ops",
-        )
-        assert "Guardrail" not in b  # reviewer 不注入 guardrail
-
-    def test_executor_with_review_protocol(self, builder):
-        b = builder.build(
-            role="executor",
-            review_protocols="## Review Protocol\nCheck tests",
-        )
-        assert "Review Protocol" in b
-
-    def test_reviewer_with_review_protocol(self, builder):
-        b = builder.build(
-            role="reviewer",
-            review_protocols="## Review Protocol\nCheck quality",
-        )
-        assert "Review Protocol" in b
-
-    def test_with_template(self, builder_with_templates):
-        b = builder_with_templates.build(role="executor")
-        assert "Executor Role" in b
-        assert "You execute tasks" in b
-
-
-# ---------------------------------------------------------------------------
-# T2: token 估算
-# ---------------------------------------------------------------------------
-
-class TestTokenEstimation:
-    def test_estimate_tokens_basic(self):
-        assert estimate_tokens("hello") > 0
-
-    def test_estimate_tokens_long_text(self):
-        text = "a" * 1000
-        tokens = estimate_tokens(text)
-        assert 200 < tokens < 400  # ~333
-
-    def test_bootstrap_under_limit(self, builder):
-        b = builder.build(
-            role="executor",
-            task_context={"title": "Short task"},
-        )
-        assert estimate_tokens(b) <= 4096
-
-    def test_bootstrap_over_limit_truncates(self):
-        builder = BootstrapBuilder(max_tokens=10)
-        b = builder.build(
-            role="executor",
-            task_context={"title": "A" * 10000},
-        )
-        assert estimate_tokens(b) <= 20  # 接近限制
-        assert "truncated" in b
-
-
-# ---------------------------------------------------------------------------
-# T3: 缺失组件降级
-# ---------------------------------------------------------------------------
-
-class TestGracefulDegradation:
-    def test_no_task_context(self, builder):
-        b = builder.build(role="executor")
-        assert b  # 不为空
-
-    def test_no_project_context(self, builder):
-        b = builder.build(
-            role="executor",
-            task_context={"title": "Task"},
-        )
-        assert "Task" in b
-
-    def test_no_template(self, builder):
-        b = builder.build(role="executor")
-        assert b  # 没有 template 也不崩溃
-
-    def test_empty_experiences(self, builder):
-        b = builder.build(role="executor", experiences=[])
-        assert b
-
-    def test_template_dir_not_exists(self, tmp_path):
-        builder = BootstrapBuilder(template_dir=tmp_path / "nonexistent")
-        b = builder.build(role="executor")
-        assert b  # 不崩溃
-
-
-# ---------------------------------------------------------------------------
-# T4: 便捷方法 + 项目上下文
-# ---------------------------------------------------------------------------
-
-class TestBuildForTask:
-    def test_build_for_task_object(self, builder):
-        task = Task(
-            id="t1", title="Build Feature", status="pending",
-            assigned_by="daemon", task_type="coding",
-            description="Implement X with tests",
-            must_haves="Unit tests, Documentation",
-            risk_level="high",
-        )
-        b = builder.build_for_task(task, role="executor")
-        assert "Build Feature" in b
-        assert "Implement X" in b
-        assert "high" in b
-
-    def test_build_for_task_with_project(self, builder):
-        task = Task(id="t1", title="T", status="pending", assigned_by="d")
-        b = builder.build_for_task(
-            task, role="executor",
-            project_config={"name": "My Project", "agents": ["a1", "a2"]},
-        )
-        assert "My Project" in b
-        assert "a1" in b
-
-    def test_with_experiences(self, builder):
-        task = Task(id="t1", title="T", status="pending", assigned_by="d")
-        experiences = [
-            {"category": "pitfall", "summary": "Always test edge cases"},
-            {"category": "best_practice", "summary": "Use type hints"},
-        ]
-        b = builder.build_for_task(task, role="executor", experiences=experiences)
-        assert "pitfall" in b
-        assert "Always test edge cases" in b
-
-    def test_with_skills(self, builder):
-        skills = [
-            {"name": "code-review", "description": "Review code quality"},
-        ]
-        b = builder.build(
-            role="executor",
-            skill_descriptions=skills,
-        )
-        assert "code-review" in b
-        assert "Review code quality" in b
-
-    def test_with_depends_on_outputs(self, builder):
-        b = builder.build(
-            role="executor",
-            task_context={
-                "title": "T",
-                "depends_on_outputs": [
-                    {"task_id": "t0", "summary": "Data downloaded"},
-                ],
-            },
-        )
-        assert "前序产出" in b
-        assert "Data downloaded" in b
diff --git a/tests/test_e2e_four_phase.py b/tests/test_e2e_four_phase.py
deleted file mode 100644
index e3c9824..0000000
--- a/tests/test_e2e_four_phase.py
+++ /dev/null
@@ -1,629 +0,0 @@
-"""#01 四相循环 E2E 集成测试
-
-需要 daemon 运行 + RUN_INTEGRATION=1。覆盖：
-  E1  comment + @mention 端到端（真实 Agent spawn）
-  E2  一轮结束 → 庞统 review spawn（真实庞统 spawn）
-  E3  多轮迭代（庞统真实 spawn + 真实创建 sub task）
-  E4  round 上限强制停止
-  E5  mention 重试（可靠制造 Agent busy）
-  E6  failed sub 触发 review（BUG-2 验证）
-  B1-B6 边界测试
-
-全部真实环境执行，不 mock Agent 行为。
-"""
-
-import json
-import os
-import sqlite3
-import sys
-import time
-import uuid
-from pathlib import Path
-from typing import Any, Dict, List
-
-import pytest
-import requests as http_requests
-
-# ── 路径设置 ──
-DEPLOY_DIR = Path.home() / ".sanguo_projects" / "sanguo_moziplus_v2"
-sys.path.insert(0, str(DEPLOY_DIR))
-
-from src.utils import get_data_root
-
-# ── 常量 ──
-API_BASE = "http://localhost:8083"
-POLL_INTERVAL = 5
-MAX_WAIT_DISPATCH = 120
-MAX_WAIT_PANGTONG = 900
-E2E_PREFIX = "e2e-01-"
-DATA_ROOT = get_data_root()
-
-
-# ── 工具函数 ──
-
-def _check_environment():
-    try:
-        resp = http_requests.get(f"{API_BASE}/api/daemon/status", timeout=5)
-        data = resp.json()
-        if data.get("status") != "running" or not data.get("ticker_running"):
-            pytest.skip(f"Daemon not ready: {data}")
-    except Exception as e:
-        pytest.skip(f"Production API not available: {e}")
-
-
-def _cleanup_project(pid: str):
-    try:
-        http_requests.post(f"{API_BASE}/api/projects/{pid}/archive", timeout=5)
-    except Exception:
-        pass
-
-
-def _create_project(plist: list, prefix: str = "E01",
-                    agents: list = None) -> str:
-    pid = f"{E2E_PREFIX}{uuid.uuid4().hex[:6]}"
-    config = {"agents": agents or ["zhangfei-dev", "simayi-challenger", "zhaoyun-data"]}
-    resp = http_requests.post(f"{API_BASE}/api/projects", json={
-        "id": pid, "name": f"{prefix}-{pid}", "config": config,
-    }, timeout=10)
-    assert resp.status_code == 200, f"Create project failed: {resp.text}"
-    plist.append(pid)
-    return pid
-
-
-def _create_task(pid: str, **kwargs) -> str:
-    tid = kwargs.pop("id", None) or f"e2e-task-{uuid.uuid4().hex[:8]}"
-    body = {"id": tid, "status": "pending", "priority": 5, **kwargs}
-    resp = http_requests.post(
-        f"{API_BASE}/api/projects/{pid}/tasks", json=body, timeout=10)
-    assert resp.status_code == 200, f"Create task failed: {resp.text}"
-    return tid
-
-
-def _get_task(pid: str, tid: str) -> Dict:
-    resp = http_requests.get(
-        f"{API_BASE}/api/projects/{pid}/tasks/{tid}?expand=all", timeout=10)
-    assert resp.status_code == 200
-    return resp.json()
-
-
-def _list_tasks(pid: str, **params) -> List[Dict]:
-    resp = http_requests.get(
-        f"{API_BASE}/api/projects/{pid}/tasks", params=params, timeout=10)
-    assert resp.status_code == 200
-    data = resp.json()
-    # API 返回 {"tasks": [...]} 或直接 [...]
-    if isinstance(data, dict) and "tasks" in data:
-        return data["tasks"]
-    return data
-
-
-def _update_status(pid: str, tid: str, status: str,
-                   agent: str = "test", detail: str = "") -> Dict:
-    body = {"status": status, "agent": agent}
-    if detail:
-        body["detail"] = detail
-    resp = http_requests.post(
-        f"{API_BASE}/api/projects/{pid}/tasks/{tid}/status",
-        json=body, timeout=10)
-    return resp.json()
-
-
-def _add_comment(pid: str, tid: str, author: str, body: str,
-                 mentions: list = None) -> int:
-    resp = http_requests.post(
-        f"{API_BASE}/api/projects/{pid}/tasks/{tid}/comments",
-        json={"author": author, "body": body, "comment_type": "general",
-              "mentions": mentions or []},
-        timeout=10)
-    assert resp.status_code == 200
-    data = resp.json()
-    return data.get("comment_id") or data.get("id")
-
-
-def _get_db_path(pid: str) -> Path:
-    return DATA_ROOT / pid / "blackboard.db"
-
-
-def _query_mentions(db_path: Path, status: str = None) -> list:
-    conn = sqlite3.connect(str(db_path))
-    conn.row_factory = sqlite3.Row
-    try:
-        if status:
-            rows = conn.execute(
-                "SELECT * FROM mention_queue WHERE status=?", (status,)
-            ).fetchall()
-        else:
-            rows = conn.execute("SELECT * FROM mention_queue").fetchall()
-        return [dict(r) for r in rows]
-    finally:
-        conn.close()
-
-
-def _set_round_count(db_path: Path, tid: str, count: int):
-    conn = sqlite3.connect(str(db_path))
-    try:
-        conn.execute("UPDATE tasks SET round_count=? WHERE id=?", (count, tid))
-        conn.commit()
-    finally:
-        conn.close()
-
-
-def _get_round_count(db_path: Path, tid: str) -> int:
-    conn = sqlite3.connect(str(db_path))
-    conn.row_factory = sqlite3.Row
-    try:
-        row = conn.execute(
-            "SELECT round_count FROM tasks WHERE id=?", (tid,)).fetchone()
-        return row["round_count"] if row else 0
-    finally:
-        conn.close()
-
-
-def _count_subtasks(db_path: Path, parent_tid: str) -> int:
-    conn = sqlite3.connect(str(db_path))
-    try:
-        row = conn.execute(
-            "SELECT COUNT(*) FROM tasks WHERE parent_task=?", (parent_tid,)
-        ).fetchone()
-        return row[0]
-    finally:
-        conn.close()
-
-
-def _wait_round(db_path: Path, tid: str, min_count: int,
-                timeout: int = MAX_WAIT_DISPATCH) -> int:
-    deadline = time.time() + timeout
-    while time.time() < deadline:
-        time.sleep(POLL_INTERVAL)
-        rc = _get_round_count(db_path, tid)
-        if rc >= min_count:
-            return rc
-    rc = _get_round_count(db_path, tid)
-    pytest.fail(f"round_count < {min_count} after {timeout}s (now={rc})")
-
-
-def _wait_subtasks(db_path: Path, parent_tid: str, min_count: int,
-                   timeout: int = MAX_WAIT_PANGTONG) -> int:
-    deadline = time.time() + timeout
-    while time.time() < deadline:
-        time.sleep(POLL_INTERVAL)
-        cnt = _count_subtasks(db_path, parent_tid)
-        if cnt >= min_count:
-            return cnt
-    cnt = _count_subtasks(db_path, parent_tid)
-    pytest.fail(f"subtasks < {min_count} after {timeout}s (now={cnt})")
-
-
-def _push_done(pid: str, tid: str, agent: str = "test"):
-    for s in ("claimed", "working", "review", "done"):
-        _update_status(pid, tid, s, agent=agent)
-
-
-def _push_failed(pid: str, tid: str, agent: str = "test"):
-    _update_status(pid, tid, "claimed", agent=agent)
-    _update_status(pid, tid, "working", agent=agent)
-    _update_status(pid, tid, "failed", agent=agent, detail="E2E forced failure")
-
-
-# ===================================================================
-# E1: comment + @mention 端到端
-# ===================================================================
-
-@pytest.mark.integration
-@pytest.mark.skipif(not os.environ.get("RUN_INTEGRATION"),
-                    reason="RUN_INTEGRATION=1 required")
-class TestE01MentionE2E:
-    @pytest.fixture(autouse=True)
-    def setup(self):
-        _check_environment()
-        self._p = []
-        yield
-        for p in self._p:
-            _cleanup_project(p)
-
-    def test_mention_spawn_e2e(self):
-        pid = _create_project(self._p, "E01")
-        tid = _create_task(pid, title="E2E mention", description="mention test",
-                           assignee="simayi-challenger")
-        _add_comment(pid, tid, "simayi-challenger",
-                     "@zhaoyun-data 请查看", mentions=["zhaoyun-data"])
-        print(f"\n🚀 E1: comment written")
-
-        db = _get_db_path(pid)
-        deadline = time.time() + MAX_WAIT_DISPATCH
-        while time.time() < deadline:
-            time.sleep(POLL_INTERVAL)
-            ms = _query_mentions(db)
-            if ms and ms[0]["status"] in ("notified", "failed"):
-                break
-        else:
-            pytest.fail(f"mention not processed after {MAX_WAIT_DISPATCH}s")
-
-        m = _query_mentions(db)[0]
-        assert m["mentioned_agent"] == "zhaoyun-data"
-        assert m["status"] == "notified", f"spawn failed: {m['status']}"
-        print(f"   ✅ mention e2e OK (spawned)")
-
-
-# ===================================================================
-# E2: 一轮结束 → 庞统 review spawn
-# ===================================================================
-
-@pytest.mark.integration
-@pytest.mark.skipif(not os.environ.get("RUN_INTEGRATION"),
-                    reason="RUN_INTEGRATION=1 required")
-class TestE02RoundComplete:
-    @pytest.fixture(autouse=True)
-    def setup(self):
-        _check_environment()
-        self._p = []
-        yield
-        for p in self._p:
-            _cleanup_project(p)
-
-    def test_round_complete_triggers_review(self):
-        pid = _create_project(self._p, "E02",
-                              agents=["pangtong-fujunshi", "zhangfei-dev", "simayi-challenger"])
-        parent = _create_task(pid, id=f"parent-{uuid.uuid4().hex[:6]}",
-                              title="E2E Parent", description="round test")
-        s1 = _create_task(pid, title="Sub1", description="s1",
-                          assignee="zhangfei-dev", parent_task=parent)
-        s2 = _create_task(pid, title="Sub2", description="s2",
-                          assignee="simayi-challenger", parent_task=parent)
-        _push_done(pid, s1, "zhangfei-dev")
-        _push_done(pid, s2, "simayi-challenger")
-        print(f"\n🚀 E2: subs done, waiting for review")
-
-        db = _get_db_path(pid)
-        rc = _wait_round(db, parent, 1)
-        print(f"   round_count={rc}")
-        print(f"   ✅ E2 round complete OK")
-
-
-# ===================================================================
-# E3: 多轮迭代（庞统真实 spawn + 真实创建 sub task）
-# ===================================================================
-
-@pytest.mark.integration
-@pytest.mark.slow
-@pytest.mark.skipif(not os.environ.get("RUN_INTEGRATION"),
-                    reason="RUN_INTEGRATION=1 required")
-class TestE03MultiRound:
-    """Round 1 → 庞统真实 review → 创建新 sub → Round 2
-
-    庞统完整链路：spawn → 读黑板 → 创建 sub task。
-    耗时 10-20 分钟。
-    """
-
-    @pytest.fixture(autouse=True)
-    def setup(self):
-        _check_environment()
-        self._p = []
-        yield
-        for p in self._p:
-            _cleanup_project(p)
-
-    def test_multi_round_full_chain(self):
-        pid = _create_project(self._p, "E03",
-                              agents=["pangtong-fujunshi", "zhangfei-dev", "simayi-challenger"])
-        parent = _create_task(
-            pid, id=f"parent-{uuid.uuid4().hex[:6]}",
-            title="E2E Multi-Round: Hello World",
-            description="创建 hello.py 输出 Hello World。第一轮创建，第二轮验证。",
-            task_type="coding",
-            must_haves=json.dumps({"capability": "python"}))
-        s1 = _create_task(pid, title="Round1: 创建 hello.py",
-                          description="创建 hello.py: print('Hello World')",
-                          assignee="zhangfei-dev", parent_task=parent)
-        _push_done(pid, s1, "zhangfei-dev")
-        print(f"\n🚀 E3 R1: sub done, waiting pangtong review")
-
-        db = _get_db_path(pid)
-        rc1 = _wait_round(db, parent, 1, timeout=MAX_WAIT_PANGTONG)
-        print(f"   R1: round_count={rc1}, waiting for new sub tasks...")
-
-        # 等庞统创建新 sub task
-        cnt = _wait_subtasks(db, parent, 2, timeout=MAX_WAIT_PANGTONG)
-        print(f"   pangtong created new subs (total={cnt})")
-
-        # 推新 sub 到 done
-        tasks = _list_tasks(pid, parent_task=parent)
-        new = [t for t in tasks if t["id"] != s1
-               and t["status"] not in ("done", "failed")]
-        assert len(new) >= 1, f"no new subs: {[t['id']+'='+t['status'] for t in tasks]}"
-        for t in new:
-            _push_done(pid, t["id"], "zhangfei-dev")
-            print(f"   pushed '{t['title']}' → done")
-
-        rc2 = _wait_round(db, parent, 2, timeout=MAX_WAIT_PANGTONG)
-        print(f"   R2: round_count={rc2}")
-        print(f"   ✅ E3 multi-round OK (pangtong real spawn + new sub)")
-
-
-# ===================================================================
-# E4: round 上限强制停止
-# ===================================================================
-
-@pytest.mark.integration
-@pytest.mark.skipif(not os.environ.get("RUN_INTEGRATION"),
-                    reason="RUN_INTEGRATION=1 required")
-class TestE04RoundLimit:
-    @pytest.fixture(autouse=True)
-    def setup(self):
-        _check_environment()
-        self._p = []
-        yield
-        for p in self._p:
-            _cleanup_project(p)
-
-    def test_round_limit(self):
-        pid = _create_project(self._p, "E04",
-                              agents=["pangtong-fujunshi", "zhangfei-dev"])
-        parent = _create_task(pid, id=f"parent-{uuid.uuid4().hex[:6]}",
-                              title="E2E Limit", description="limit test")
-        s1 = _create_task(pid, title="Sub", description="limit sub",
-                          assignee="zhangfei-dev", parent_task=parent)
-        _push_done(pid, s1, "zhangfei-dev")
-
-        db = _get_db_path(pid)
-        _set_round_count(db, parent, 5)
-        print(f"\n🚀 E4: round_count=5, waiting 2 ticks")
-
-        time.sleep(60)
-        rc = _get_round_count(db, parent)
-        assert rc == 5, f"round_count changed: {rc}"
-        print(f"   ✅ E4 round limit OK (rc=5 unchanged)")
-
-
-# ===================================================================
-# E5: mention 重试（可靠 Agent busy）
-# ===================================================================
-
-@pytest.mark.integration
-@pytest.mark.skipif(not os.environ.get("RUN_INTEGRATION"),
-                    reason="RUN_INTEGRATION=1 required")
-class TestE05MentionRetry:
-    """可靠制造 busy：先让 zhaoyun-data 有个 working task，再 @ 它。
-
-    1. 创建 blocker task → ticker dispatch → zhaoyun-data working
-    2. 创建 mention task → @zhaoyun-data → busy → retry
-    3. 推 blocker done → zhaoyun-data 空闲 → mention spawn 成功
-    """
-
-    @pytest.fixture(autouse=True)
-    def setup(self):
-        _check_environment()
-        self._p = []
-        yield
-        for p in self._p:
-            _cleanup_project(p)
-
-    def test_mention_retry_on_busy(self):
-        pid = _create_project(self._p, "E05",
-                              agents=["zhaoyun-data", "simayi-challenger"])
-
-        # 1. blocker task
-        blocker = _create_task(pid, title="E2E Blocker",
-                               description="占用 zhaoyun-data",
-                               assignee="zhaoyun-data")
-        print(f"\n🚀 E5: blocker {blocker}, waiting for zhaoyun-data spawn")
-
-        deadline = time.time() + MAX_WAIT_DISPATCH
-        busy = False
-        while time.time() < deadline:
-            time.sleep(POLL_INTERVAL)
-            t = _get_task(pid, blocker)
-            if t["status"] in ("claimed", "working"):
-                busy = True
-                print(f"   zhaoyun-data busy: status={t['status']}")
-                break
-
-        if not busy:
-            _update_status(pid, blocker, "claimed", agent="zhaoyun-data")
-            _update_status(pid, blocker, "working", agent="zhaoyun-data")
-            print(f"   forced blocker → working")
-
-        # 2. @zhaoyun-data on another task
-        mtid = _create_task(pid, title="E2E Mention Task",
-                            description="mention retry test",
-                            assignee="simayi-challenger")
-        _add_comment(pid, mtid, "simayi-challenger",
-                     "@zhaoyun-data 请查看", mentions=["zhaoyun-data"])
-        print(f"   mention created while zhaoyun-data busy")
-
-        db = _get_db_path(pid)
-        time.sleep(35)  # 1 tick
-
-        ms = _query_mentions(db)
-        assert len(ms) >= 1, "no mention written"
-        print(f"   mention: status={ms[0]['status']} retry={ms[0]['retry_count']}")
-
-        # 3. release zhaoyun-data
-        _update_status(pid, blocker, "review", agent="zhaoyun-data")
-        _update_status(pid, blocker, "done", agent="zhaoyun-data")
-        print(f"   blocker → done, zhaoyun-data free")
-
-        # 4. wait for mention resolution
-        deadline2 = time.time() + MAX_WAIT_DISPATCH
-        while time.time() < deadline2:
-            time.sleep(POLL_INTERVAL)
-            ms2 = _query_mentions(db)
-            if ms2 and ms2[0]["status"] in ("notified", "failed"):
-                break
-
-        mf = _query_mentions(db)[0]
-        print(f"   final: status={mf['status']} retry={mf['retry_count']}")
-
-        assert mf["status"] in ("notified", "failed"), f"unresolved: {mf}"
-        if mf["retry_count"] > 0:
-            print(f"   ✅ E5 retry verified (retry_count={mf['retry_count']})")
-        else:
-            print(f"   ✅ E5 resolved directly (timing OK)")
-        print(f"   ✅ E5 mention retry OK")
-
-
-# ===================================================================
-# E6: failed sub 触发 review
-# ===================================================================
-
-@pytest.mark.integration
-@pytest.mark.skipif(not os.environ.get("RUN_INTEGRATION"),
-                    reason="RUN_INTEGRATION=1 required")
-class TestE06FailedSubReview:
-    @pytest.fixture(autouse=True)
-    def setup(self):
-        _check_environment()
-        self._p = []
-        yield
-        for p in self._p:
-            _cleanup_project(p)
-
-    def test_failed_sub_review(self):
-        pid = _create_project(self._p, "E06",
-                              agents=["pangtong-fujunshi", "zhangfei-dev", "simayi-challenger"])
-        parent = _create_task(pid, id=f"parent-{uuid.uuid4().hex[:6]}",
-                              title="E2E Failed Sub", description="BUG-2 test")
-        s1 = _create_task(pid, title="Done", description="s1",
-                          assignee="zhangfei-dev", parent_task=parent)
-        s2 = _create_task(pid, title="Failed", description="s2",
-                          assignee="simayi-challenger", parent_task=parent)
-        _push_done(pid, s1, "zhangfei-dev")
-        _push_failed(pid, s2, "simayi-challenger")
-        print(f"\n🚀 E6: done+failed, waiting for review")
-
-        db = _get_db_path(pid)
-        rc = _wait_round(db, parent, 1)
-        print(f"   round_count={rc}")
-        print(f"   ✅ E6 failed-sub review OK (BUG-2 verified)")
-
-
-# ===================================================================
-# B1-B6: 边界测试
-# ===================================================================
-
-@pytest.mark.integration
-@pytest.mark.skipif(not os.environ.get("RUN_INTEGRATION"),
-                    reason="RUN_INTEGRATION=1 required")
-class TestBoundary:
-    @pytest.fixture(autouse=True)
-    def setup(self):
-        _check_environment()
-        self._p = []
-        yield
-        for p in self._p:
-            _cleanup_project(p)
-
-    def test_B1_empty_mentions(self):
-        """空 mentions 列表 → 不写入 mention_queue"""
-        pid = _create_project(self._p, "B1")
-        tid = _create_task(pid, title="B1")
-        _add_comment(pid, tid, "test", "no mentions", mentions=[])
-        db = _get_db_path(pid)
-        assert len(_query_mentions(db)) == 0
-        print(f"   ✅ B1: empty mentions → no queue entry")
-
-    def test_B2_nonexistent_agent(self):
-        """不存在的 agent_id → mention 写入成功，spawn 失败后 retry 递增"""
-        pid = _create_project(self._p, "B2", agents=["simayi-challenger"])
-        tid = _create_task(pid, title="B2")
-        _add_comment(pid, tid, "test", "@ghost-agent 你在哪",
-                     mentions=["ghost-agent"])
-
-        db = _get_db_path(pid)
-        # 等写入
-        time.sleep(2)
-        ms = _query_mentions(db)
-        assert len(ms) >= 1
-        assert ms[0]["mentioned_agent"] == "ghost-agent"
-        assert ms[0]["status"] == "pending"
-
-        # 等 ticker 尝试 spawn → 失败 → retry_count 递增
-        deadline = time.time() + MAX_WAIT_DISPATCH
-        retried = False
-        while time.time() < deadline:
-            time.sleep(POLL_INTERVAL)
-            ms2 = _query_mentions(db)
-            if ms2 and ms2[0]["retry_count"] > 0:
-                retried = True
-                print(f"   ghost-agent retry_count={ms2[0]['retry_count']}")
-                break
-            if ms2 and ms2[0]["status"] == "failed":
-                retried = True
-                print(f"   ghost-agent failed (retries exhausted)")
-                break
-
-        assert retried, "ghost-agent mention never retried"
-        print(f"   ✅ B2: nonexistent agent → retry OK")
-
-    def test_B3_duplicate_mentions(self):
-        """同一 comment @ 同一 agent 多次 → 去重只写 1 条"""
-        pid = _create_project(self._p, "B3")
-        tid = _create_task(pid, title="B3")
-        cid = _add_comment(pid, tid, "test", "@agent-a @agent-a",
-                           mentions=["agent-a", "agent-a"])
-
-        db = _get_db_path(pid)
-        time.sleep(2)
-        ms = _query_mentions(db)
-        # mentions 列表有两个 agent-a，但 record_mentions 会去重
-        agent_a_count = sum(1 for m in ms if m["mentioned_agent"] == "agent-a")
-        assert agent_a_count <= 2  # 允许 1 或 2（取决于去重在哪一层）
-        print(f"   ✅ B3: duplicate mention count={agent_a_count}")
-
-    def test_B4_parent_no_subs(self):
-        """parent 无 sub task → _check_round_complete 不触发"""
-        pid = _create_project(self._p, "B4",
-                              agents=["pangtong-fujunshi"])
-        parent = _create_task(pid, id=f"parent-{uuid.uuid4().hex[:6]}",
-                              title="B4 Parent", description="no subs")
-        db = _get_db_path(pid)
-
-        # 等 2 tick
-        time.sleep(60)
-        rc = _get_round_count(db, parent)
-        assert rc == 0, f"round_count should stay 0, got {rc}"
-        print(f"   ✅ B4: parent with no subs → no review")
-
-    def test_B5_parent_done_no_subs(self):
-        """parent done 且无 sub → 不触发"""
-        pid = _create_project(self._p, "B5",
-                              agents=["pangtong-fujunshi"])
-        parent = _create_task(pid, id=f"parent-{uuid.uuid4().hex[:6]}",
-                              title="B5 Parent", description="done no subs")
-        _update_status(pid, parent, "claimed", agent="test")
-        _update_status(pid, parent, "working", agent="test")
-        _update_status(pid, parent, "review", agent="test")
-        _update_status(pid, parent, "done", agent="test")
-
-        db = _get_db_path(pid)
-        time.sleep(60)
-        rc = _get_round_count(db, parent)
-        assert rc == 0
-        print(f"   ✅ B5: parent done no subs → no review")
-
-    def test_B6_comment_deleted_mention_handling(self):
-        """comment 被删除后 mention 仍存在于 queue"""
-        pid = _create_project(self._p, "B6")
-        tid = _create_task(pid, title="B6")
-        _add_comment(pid, tid, "test", "@simayi-challenger check this",
-                     mentions=["simayi-challenger"])
-
-        db = _get_db_path(pid)
-        time.sleep(2)
-        ms = _query_mentions(db)
-        assert len(ms) >= 1
-
-        # 删除 comment（通过 DB 直接操作模拟）
-        conn = sqlite3.connect(str(db))
-        try:
-            conn.execute("DELETE FROM comments WHERE task_id=?", (tid,))
-            conn.commit()
-        finally:
-            conn.close()
-
-        # mention 仍在 queue（没有 FK cascade 或 JOIN 过滤时）
-        ms2 = _query_mentions(db)
-        # 如果 get_pending_mentions JOIN comments，可能过滤掉
-        # 这里验证 mention 行本身仍存在
-        assert len(ms2) >= 1, "mention row should persist even after comment deleted"
-        print(f"   ✅ B6: mention persists after comment deleted")
diff --git a/tests/unit/__init__.py b/tests/unit/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/tests/test_blackboard.py b/tests/unit/test_blackboard.py
similarity index 99%
rename from tests/test_blackboard.py
rename to tests/unit/test_blackboard.py
index 5cee9e0..e17d1e0 100644
--- a/tests/test_blackboard.py
+++ b/tests/unit/test_blackboard.py
@@ -24,6 +24,8 @@ from src.blackboard.models import (
 from src.blackboard.operations import Blackboard
 from src.blackboard.queries import Queries
 
+pytestmark = pytest.mark.unit
+
 
 @pytest.fixture
 def tmp_db(tmp_path):
diff --git a/tests/unit/test_bootstrap.py b/tests/unit/test_bootstrap.py
new file mode 100644
index 0000000..cd26cf3
--- /dev/null
+++ b/tests/unit/test_bootstrap.py
@@ -0,0 +1,191 @@
+"""#11 Bootstrap 四段式拼装单元测试（v2.1）
+
+覆盖：
+- T1: build(task, role) 4 段结构
+- T2: token 估算 + 预算告警
+- T3: 缺失组件降级
+- T4: build_for_task 便捷方法
+- T5: _read_skill fallback
+- T6: ROLE_SKILL_MAP 覆盖
+"""
+
+import pytest
+
+pytestmark = pytest.mark.unit
+from unittest.mock import patch
+from pathlib import Path
+
+from src.daemon.bootstrap import BootstrapBuilder, estimate_tokens
+
+
+@pytest.fixture
+def builder():
+    return BootstrapBuilder(max_tokens=4096)
+
+
+# ---------------------------------------------------------------------------
+# T1: build(task, role) 4 段结构
+# ---------------------------------------------------------------------------
+
+class TestFourSectionBuild:
+    def test_basic_executor_build(self, builder):
+        b = builder.build(
+            task={"task_id": "t1", "title": "Write tests", "description": "Write unit tests",
+                  "must_haves": "100% coverage", "status": "claimed"},
+            role="executor",
+        )
+        # 段 1: 任务上下文
+        assert "Write tests" in b
+        assert "t1" in b
+        assert "100% coverage" in b
+        # 段 4: 硬约束
+        assert "review" in b
+        assert "handoff" in b
+
+    def test_basic_reviewer_build(self, builder):
+        b = builder.build(
+            task={"task_id": "t2", "title": "Review PR"},
+            role="reviewer",
+        )
+        assert "Review PR" in b
+        # 段 4: reviewer 硬约束
+        assert "pass/fail" in b or "pass" in b
+
+    def test_planner_build(self, builder):
+        b = builder.build(
+            task={"task_id": "t3", "title": "Plan sprint"},
+            role="planner",
+        )
+        assert "Plan sprint" in b
+
+    def test_depends_on_outputs_injected(self, builder):
+        b = builder.build(
+            task={
+                "title": "T",
+                "depends_on_outputs": [
+                    {"task_id": "t0", "summary": "Data downloaded"},
+                ],
+            },
+            role="executor",
+        )
+        assert "前序产出" in b
+        assert "Data downloaded" in b
+
+    def test_no_depends_on_omitted(self, builder):
+        b = builder.build(
+            task={"title": "T"},
+            role="executor",
+        )
+        assert "前序产出" not in b
+
+
+# ---------------------------------------------------------------------------
+# T2: token 估算 + 预算告警
+# ---------------------------------------------------------------------------
+
+class TestTokenEstimation:
+    def test_estimate_tokens_basic(self):
+        assert estimate_tokens("hello") > 0
+
+    def test_estimate_tokens_long_text(self):
+        text = "a" * 1000
+        tokens = estimate_tokens(text)
+        assert 200 < tokens < 400
+
+
+# ---------------------------------------------------------------------------
+# T3: 缺失组件降级
+# ---------------------------------------------------------------------------
+
+class TestGracefulDegradation:
+    def test_empty_task(self, builder):
+        b = builder.build(task={}, role="executor")
+        assert b  # 不为空
+        assert "硬约束" in b
+
+    def test_partial_task(self, builder):
+        b = builder.build(task={"title": "Only title"}, role="executor")
+        assert "Only title" in b
+
+
+# ---------------------------------------------------------------------------
+# T4: build_for_task 便捷方法
+# ---------------------------------------------------------------------------
+
+class TestBuildForTask:
+    def test_build_for_task_object(self, builder):
+        """用 mock task 对象测试 build_for_task"""
+        class MockTask:
+            id = "t1"
+            title = "Build Feature"
+            description = "Implement X with tests"
+            must_haves = "Unit tests, Documentation"
+            status = "claimed"
+        task = MockTask()
+        b = builder.build_for_task(task, role="executor")
+        assert "Build Feature" in b
+        assert "Implement X" in b
+
+    def test_build_for_task_ignores_kwargs(self, builder):
+        """build_for_task 忽略旧参数"""
+        class MockTask:
+            id = "t1"
+            title = "T"
+            description = ""
+            must_haves = ""
+            status = ""
+        task = MockTask()
+        # 旧参数 project_config/experiences 不应报错
+        b = builder.build_for_task(
+            task, role="executor",
+            project_config={"name": "Old"},
+            experiences=[{"x": 1}],
+        )
+        assert "T" in b
+        # 不应出现旧参数内容
+        assert "Old" not in b
+
+
+# ---------------------------------------------------------------------------
+# T5: _read_skill fallback
+# ---------------------------------------------------------------------------
+
+class TestReadSkillFallback:
+    def test_missing_skill_file_returns_empty(self, builder):
+        """Skill 文件不存在时返回空字符串，不抛异常"""
+        result = builder._read_skill("nonexistent-skill-xyz")
+        assert result == ""
+
+    def test_existing_skill_file_read(self, builder):
+        """能读取实际存在的 Skill 文件"""
+        # blackboard-executor 应该存在（P1 已创建）
+        result = builder._read_skill("blackboard-executor")
+        assert "执行" in result or "executor" in result.lower()
+
+
+# ---------------------------------------------------------------------------
+# T6: ROLE_SKILL_MAP 覆盖
+# ---------------------------------------------------------------------------
+
+class TestRoleSkillMap:
+    def test_all_roles_mapped(self):
+        assert set(BootstrapBuilder.ROLE_SKILL_MAP.keys()) == {
+            "executor", "reviewer", "reviewer-simayi",
+            "reviewer-pangtong", "planner", "claim",
+        }
+
+    def test_unknown_role_warns(self, builder):
+        """未映射的 role 输出 warning"""
+        import logging
+        with patch("src.daemon.bootstrap.logger") as mock_logger:
+            builder.build(task={"title": "T"}, role="unknown_role")
+            mock_logger.warning.assert_called_with(
+                "No skill mapping for role: %s", "unknown_role"
+            )
+
+    def test_discussion_role_no_warning(self, builder, caplog):
+        """discussion 角色不应触发 warning"""
+        import logging
+        with caplog.at_level(logging.WARNING, logger="moziplus-v2.bootstrap"):
+            builder.build(task={"title": "T"}, role="discussion")
+        assert "No skill mapping" not in caplog.text
diff --git a/tests/test_cli.py b/tests/unit/test_cli.py
similarity index 99%
rename from tests/test_cli.py
rename to tests/unit/test_cli.py
index 25d00f1..a47b311 100644
--- a/tests/test_cli.py
+++ b/tests/unit/test_cli.py
@@ -6,6 +6,8 @@ from pathlib import Path
 
 import pytest
 
+pytestmark = pytest.mark.unit
+
 from src.cli.blackboard import run_blackboard_cli, run_admin_cli
 from src.blackboard.operations import Blackboard
 from src.blackboard.models import Task
diff --git a/tests/test_counter.py b/tests/unit/test_counter.py
similarity index 99%
rename from tests/test_counter.py
rename to tests/unit/test_counter.py
index bb3cc18..93c17a9 100644
--- a/tests/test_counter.py
+++ b/tests/unit/test_counter.py
@@ -10,6 +10,8 @@
 import asyncio
 import pytest
 
+pytestmark = pytest.mark.unit
+
 from src.daemon.counter import ActiveAgentCounter
 
 
diff --git a/tests/test_dispatcher.py b/tests/unit/test_dispatcher.py
similarity index 62%
rename from tests/test_dispatcher.py
rename to tests/unit/test_dispatcher.py
index eb46bd7..0e3df18 100644
--- a/tests/test_dispatcher.py
+++ b/tests/unit/test_dispatcher.py
@@ -1,14 +1,4 @@
-"""F9 Agent 调度器单元测试
-
-按 test-plan-v2.6.md §F9：
-- T1: 三级决策树（P0）
-- T2: 调度不阻塞（P0）
-- T3: 队列满拒绝（P0）
-- T4: 任务优先级排序（P1）
-
-v2.8 新增（#07 AgentBusyError 分类）：
-- E14.3: Dispatcher 错误区分（1 个测试）
-"""
+"""F9 Agent 调度器单元测试 — 三级决策树 + 批量决策"""
 
 import asyncio
 import json
@@ -21,6 +11,8 @@ from src.blackboard.models import Task
 from src.daemon.dispatcher import Dispatcher, DispatchLevel
 from src.daemon.spawner import AgentBusyError
 
+pytestmark = pytest.mark.unit
+
 
 # ---------------------------------------------------------------------------
 # Fixtures
@@ -184,7 +176,7 @@ class TestDispatch:
 
 class TestConcurrencyControl:
     def test_counter_busy_skips(self, dispatcher, task_pending):
-        """Agent 忙 → skip（v2.8 #07: AgentBusyError 从 spawn_full_agent 抛出）"""
+        """Agent 忙 → skip"""
         mock_spawner = MagicMock()
         mock_spawner.spawn_full_agent = AsyncMock(
             side_effect=AgentBusyError("zhangfei-dev", reason="counter_blocked")
@@ -196,7 +188,7 @@ class TestConcurrencyControl:
         assert "busy" in result["reason"].lower()
 
     def test_counter_releases_on_error(self, dispatcher, task_pending):
-        """spawn 失败后释放 counter（v2.8 #07: counter.release 由 spawn_full_agent 内部保证）"""
+        """spawn 失败后释放 counter"""
         mock_spawner = MagicMock()
         mock_spawner.spawn_full_agent = AsyncMock(side_effect=RuntimeError("fail"))
         dispatcher.spawner = mock_spawner
@@ -277,159 +269,4 @@ class TestDispatcherErrorClassification:
         dispatcher.spawner = mock_spawner
 
         result = asyncio.run(dispatcher.dispatch(task_pending))
-        # counter_blocked 通常在 can_acquire 阶段被拦，结果为 skipped
-        # 如果穿透到 spawn_full_agent，则为 error
         assert result["status"] in ("skipped", "error")
-
-
-# ---------------------------------------------------------------------------
-# 司马懿评审补充：_rollback_current_agent + on_complete 统一（v2.8 #07.2）
-# ---------------------------------------------------------------------------
-
-class TestRollbackAndOnComplete:
-    """司马懿评审遗漏 #3 + #4: crash 后 current_agent 回退 + on_complete 统一路径"""
-
-    def test_rollback_current_agent_on_crash(self, tmp_path):
-        """executor crash → _rollback_current_agent 回退 current_agent → assignee
-
-        #07.2 核心改动：executor crash 后也回退 current_agent，
-        避免 _dispatch_reviews 的 exclude_current 卡死。
-        """
-        from src.blackboard.operations import Blackboard
-        from src.blackboard.models import Task
-
-        db_path = tmp_path / "blackboard.db"
-        bb = Blackboard(db_path)
-        bb.create_task(Task(
-            id="t1", title="T", status="working",
-            assigned_by="daemon", assignee="zhangfei-dev",
-        ))
-
-        # 通过状态变更设置 current_agent（create_task 不持久化 current_agent）
-        conn = bb._conn()
-        try:
-            conn.execute("UPDATE tasks SET current_agent=? WHERE id=?", ("zhangfei-dev", "t1"))
-            conn.commit()
-        finally:
-            conn.close()
-
-        dispatcher = Dispatcher(registered_agents=["zhangfei-dev"])
-        dispatcher._rollback_current_agent(db_path, "t1", "zhangfei-dev")
-
-        import sqlite3
-        conn2 = sqlite3.connect(str(db_path))
-        conn2.row_factory = sqlite3.Row
-        row = conn2.execute("SELECT current_agent, assignee FROM tasks WHERE id=?", ("t1",)).fetchone()
-        conn2.close()
-
-        assert row["current_agent"] == row["assignee"] == "zhangfei-dev"
-
-    def test_rollback_different_agent(self, tmp_path):
-        """current_agent ≠ agent_id → 不回退（安全检查）
-
-        _rollback_current_agent 的 WHERE 条件包含 current_agent=?，
-        如果 agent_id 不匹配 current_agent 则不执行更新。
-        """
-        from src.blackboard.operations import Blackboard
-        from src.blackboard.models import Task
-
-        db_path = tmp_path / "blackboard.db"
-        bb = Blackboard(db_path)
-        bb.create_task(Task(
-            id="t1", title="T", status="working",
-            assigned_by="daemon", assignee="zhangfei-dev",
-        ))
-
-        # 设置 current_agent 为 simayi-challenger
-        conn = bb._conn()
-        try:
-            conn.execute("UPDATE tasks SET current_agent=? WHERE id=?", ("simayi-challenger", "t1"))
-            conn.commit()
-        finally:
-            conn.close()
-
-        dispatcher = Dispatcher(registered_agents=["zhangfei-dev"])
-        # 用错误的 agent_id 回退
-        dispatcher._rollback_current_agent(db_path, "t1", "wrong-agent")
-
-        import sqlite3
-        conn2 = sqlite3.connect(str(db_path))
-        conn2.row_factory = sqlite3.Row
-        row = conn2.execute("SELECT current_agent FROM tasks WHERE id=?", ("t1",)).fetchone()
-        conn2.close()
-
-        # current_agent 不变
-        assert row["current_agent"] == "simayi-challenger"
-
-    def test_on_complete_crash_rollback_executor(self, tmp_path):
-        """executor crash → rollback current_agent + _task_auto_complete（标 review）
-
-        #07.2: crash 回退在 if _is_review 之前执行。
-        """
-        from src.blackboard.operations import Blackboard
-        from src.blackboard.models import Task
-
-        db_path = tmp_path / "blackboard.db"
-        bb = Blackboard(db_path)
-        bb.create_task(Task(
-            id="t1", title="T", status="working",
-            assigned_by="daemon", assignee="zhangfei-dev",
-        ))
-
-        # 设置 current_agent
-        conn = bb._conn()
-        try:
-            conn.execute("UPDATE tasks SET current_agent=? WHERE id=?", ("zhangfei-dev", "t1"))
-            conn.commit()
-        finally:
-            conn.close()
-
-        dispatcher = Dispatcher(registered_agents=["zhangfei-dev"])
-
-        # executor crash: rollback current_agent
-        dispatcher._rollback_current_agent(db_path, "t1", "zhangfei-dev")
-
-        import sqlite3
-        conn2 = sqlite3.connect(str(db_path))
-        conn2.row_factory = sqlite3.Row
-        row = conn2.execute("SELECT current_agent FROM tasks WHERE id=?", ("t1",)).fetchone()
-        conn2.close()
-        assert row["current_agent"] == "zhangfei-dev"  # rollback 到 assignee
-
-    def test_on_complete_crash_rollback_review(self, tmp_path):
-        """review crash → rollback current_agent + 保持 review 状态
-
-        #07.2: crash 回退在 if _is_review 之前执行。
-        review crash 后 current_agent 回退，但任务保持 review 状态。
-        """
-        from src.blackboard.operations import Blackboard
-        from src.blackboard.models import Task
-
-        db_path = tmp_path / "blackboard.db"
-        bb = Blackboard(db_path)
-        bb.create_task(Task(
-            id="t1", title="T", status="review",
-            assigned_by="daemon", assignee="zhangfei-dev",
-        ))
-
-        # 设置 current_agent 为 reviewer
-        conn = bb._conn()
-        try:
-            conn.execute("UPDATE tasks SET current_agent=? WHERE id=?", ("simayi-challenger", "t1"))
-            conn.commit()
-        finally:
-            conn.close()
-
-        dispatcher = Dispatcher(registered_agents=["simayi-challenger"])
-        dispatcher._rollback_current_agent(db_path, "t1", "simayi-challenger")
-
-        import sqlite3
-        conn2 = sqlite3.connect(str(db_path))
-        conn2.row_factory = sqlite3.Row
-        row = conn2.execute("SELECT current_agent, status FROM tasks WHERE id=?", ("t1",)).fetchone()
-        conn2.close()
-
-        # current_agent 回退到 assignee
-        assert row["current_agent"] == "zhangfei-dev"
-        # review 状态不变
-        assert row["status"] == "review"
diff --git a/tests/test_experience.py b/tests/unit/test_experience.py
similarity index 99%
rename from tests/test_experience.py
rename to tests/unit/test_experience.py
index af23c08..27c37cb 100644
--- a/tests/test_experience.py
+++ b/tests/unit/test_experience.py
@@ -9,6 +9,8 @@
 
 import json
 import pytest
+
+pytestmark = pytest.mark.unit
 from pathlib import Path
 
 from src.daemon.experience import (
diff --git a/tests/test_health.py b/tests/unit/test_health.py
similarity index 99%
rename from tests/test_health.py
rename to tests/unit/test_health.py
index 9e28ad0..c77d1f8 100644
--- a/tests/test_health.py
+++ b/tests/unit/test_health.py
@@ -10,6 +10,8 @@
 import asyncio
 import json
 import pytest
+
+pytestmark = pytest.mark.unit
 from pathlib import Path
 
 from src.blackboard.operations import Blackboard
diff --git a/tests/test_inbox.py b/tests/unit/test_inbox.py
similarity index 99%
rename from tests/test_inbox.py
rename to tests/unit/test_inbox.py
index c46bce7..57c73c5 100644
--- a/tests/test_inbox.py
+++ b/tests/unit/test_inbox.py
@@ -12,6 +12,8 @@ import asyncio
 import json
 import threading
 import pytest
+
+pytestmark = pytest.mark.unit
 from pathlib import Path
 
 from src.daemon.inbox import InboxWatcher
diff --git a/tests/test_registry.py b/tests/unit/test_registry.py
similarity index 99%
rename from tests/test_registry.py
rename to tests/unit/test_registry.py
index bf13400..8e21f03 100644
--- a/tests/test_registry.py
+++ b/tests/unit/test_registry.py
@@ -7,6 +7,8 @@ from pathlib import Path
 
 import pytest
 
+pytestmark = pytest.mark.unit
+
 from src.blackboard.registry import ProjectRegistry
 from src.blackboard.operations import Blackboard
 from src.blackboard.models import Task
diff --git a/tests/unit/test_review.py b/tests/unit/test_review.py
new file mode 100644
index 0000000..edc7626
--- /dev/null
+++ b/tests/unit/test_review.py
@@ -0,0 +1,87 @@
+"""F14 Rebuttal 单元测试
+
+按 test-plan-v2.6.md §F14：
+- F14 T1: 反驳权流程（P0）
+- F14 T2: 最大轮次限制（P0）
+"""
+
+import json
+import pytest
+from pathlib import Path
+from unittest.mock import MagicMock
+
+from src.blackboard.models import Task
+from src.blackboard.operations import Blackboard
+from src.daemon.review import RebuttalManager
+
+pytestmark = pytest.mark.unit
+
+
+@pytest.fixture
+def db_path(tmp_path):
+    return tmp_path / "blackboard.db"
+
+
+@pytest.fixture
+def bb(db_path):
+    return Blackboard(db_path)
+
+
+# ---------------------------------------------------------------------------
+# F14 T1: 反驳权流程
+# ---------------------------------------------------------------------------
+
+class TestRebuttal:
+    def test_submit_rebuttal_accepted(self, bb):
+        task = Task(id="t1", title="T", status="pending", assigned_by="d")
+        bb.create_task(task)
+
+        rm = RebuttalManager(bb=bb)
+        result = rm.submit_rebuttal("t1", "agent-1", "I disagree with the review")
+        assert result["status"] == "accepted"
+        assert result["round"] == 1
+        assert result["escalation_target"] == "simayi-challenger"
+
+    def test_second_round_escalates_to_pangtong(self, bb):
+        task = Task(id="t1", title="T", status="pending", assigned_by="d")
+        bb.create_task(task)
+
+        rm = RebuttalManager(bb=bb)
+        rm.submit_rebuttal("t1", "agent-1", "Round 1")
+        result = rm.submit_rebuttal("t1", "agent-1", "Round 2")
+        assert result["status"] == "accepted"
+        assert result["round"] == 2
+        assert result["escalation_target"] == "pangtong-fujunshi"
+
+
+# ---------------------------------------------------------------------------
+# F14 T2: 最大轮次限制
+# ---------------------------------------------------------------------------
+
+class TestRebuttalLimits:
+    def test_max_rounds_rejected(self, bb):
+        task = Task(id="t1", title="T", status="pending", assigned_by="d")
+        bb.create_task(task)
+
+        rm = RebuttalManager(bb=bb)
+        rm.submit_rebuttal("t1", "a", "R1")
+        rm.submit_rebuttal("t1", "a", "R2")
+        result = rm.submit_rebuttal("t1", "a", "R3")
+        assert result["status"] == "rejected"
+        assert "Max" in result["reason"]
+
+    def test_rebuttal_without_bb(self):
+        rm = RebuttalManager(bb=None)
+        result = rm.submit_rebuttal("t1", "a", "reason")
+        assert result["status"] == "accepted"
+        assert result["round"] == 1
+
+    def test_rebuttal_observation_recorded(self, bb):
+        task = Task(id="t1", title="T", status="pending", assigned_by="d")
+        bb.create_task(task)
+
+        rm = RebuttalManager(bb=bb)
+        rm.submit_rebuttal("t1", "agent-1", "test reason", evidence="file.txt")
+        obs = bb.get_observations(task_id="t1")
+        rebuttals = [o for o in obs if "Rebuttal round" in (o.body or "")]
+        assert len(rebuttals) == 1
diff --git a/tests/test_router.py b/tests/unit/test_router.py
similarity index 99%
rename from tests/test_router.py
rename to tests/unit/test_router.py
index 90fb838..b114b32 100644
--- a/tests/test_router.py
+++ b/tests/unit/test_router.py
@@ -1,6 +1,8 @@
 """Router 单元测试 — 三种路由模式 + 校验 + fallback"""
 
 import pytest
+
+pytestmark = pytest.mark.unit
 from unittest.mock import MagicMock, patch
 
 from src.daemon.router import (
diff --git a/tests/test_skill_system.py b/tests/unit/test_skill_system.py
similarity index 99%
rename from tests/test_skill_system.py
rename to tests/unit/test_skill_system.py
index e262025..25b82fb 100644
--- a/tests/test_skill_system.py
+++ b/tests/unit/test_skill_system.py
@@ -9,6 +9,8 @@
 
 import json
 import pytest
+
+pytestmark = pytest.mark.unit
 from pathlib import Path
 
 from src.daemon.skill_system import (
diff --git a/tests/test_spawner.py b/tests/unit/test_spawner.py
similarity index 68%
rename from tests/test_spawner.py
rename to tests/unit/test_spawner.py
index f4f774e..d6708cd 100644
--- a/tests/test_spawner.py
+++ b/tests/unit/test_spawner.py
@@ -1,16 +1,4 @@
-"""F9 Agent Spawner 单元测试
-
-按 test-plan-v2.6.md §F9 Spawner：
-- T1: spawn 成功（P0）
-- T2: 超时处理（P0）
-- T3: spawn 失败（P0）
-- T4: session 清理（P1）
-
-v2.8 新增（#07 Acquire-First + Compact Hanging + AgentBusyError）：
-- E11: Spawner Acquire-First Phase 0-4（6 个测试）
-- E13: Compact Hanging 不标 failed（3 个测试）
-- E14: AgentBusyError 分类（3 个测试）
-"""
+"""F9 Agent Spawner 单元测试 — classify/session 管理"""
 
 import asyncio
 import pytest
@@ -19,6 +7,8 @@ from pathlib import Path
 from src.blackboard.operations import Blackboard
 from src.daemon.spawner import AgentSpawner, AgentBusyError
 
+pytestmark = pytest.mark.unit
+
 
 @pytest.fixture
 def db_path(tmp_path):
@@ -35,13 +25,8 @@ def spawner(db_path):
     return AgentSpawner(db_path=db_path, dry_run=True)
 
 
-@pytest.fixture
-def real_spawner(db_path):
-    return AgentSpawner(db_path=db_path, dry_run=False, agent_timeout=2.0)
-
-
 # ---------------------------------------------------------------------------
-# T1: spawn 成功
+# T1: spawn 成功（dry_run）
 # ---------------------------------------------------------------------------
 
 class TestSpawnSuccess:
@@ -78,57 +63,6 @@ class TestSpawnSuccess:
         assert len(set(ids)) == 3  # 每个 session_id 唯一
 
 
-# ---------------------------------------------------------------------------
-# T2: 超时处理
-# ---------------------------------------------------------------------------
-
-class TestTimeout:
-    def test_timeout_kills_process(self, tmp_path):
-        """超时后 kill 进程"""
-        db_path = tmp_path / "blackboard.db"
-        Blackboard(db_path)  # init
-        spawner = AgentSpawner(db_path=db_path, dry_run=False, agent_timeout=0.5)
-
-        # Spawn a long-running process (sleep 10)
-        session_id = asyncio.run(
-            spawner.spawn_full_agent(
-                "test-agent",
-                "sleep 10",  # will be passed as --message, actual agent may ignore
-                task_id=None,  # no task to avoid DB writes for non-existent task
-            )
-        )
-        # Wait for timeout
-        asyncio.run(asyncio.sleep(1.0))
-
-        session = spawner.get_session(session_id)
-        if session:
-            # Process should have been killed
-            assert session["status"] in ("timed_out", "running", "completed")
-
-
-# ---------------------------------------------------------------------------
-# T3: spawn 失败
-# ---------------------------------------------------------------------------
-
-class TestSpawnFailure:
-    def test_nonexistent_command(self, real_spawner, db_path, bb):
-        """命令不存在 → spawn_failed"""
-        bb.create_task(
-            __import__("src.blackboard.models", fromlist=["Task"]).Task(
-                id="t1", title="T", status="pending", assigned_by="d"
-            )
-        )
-
-        # Spawner will try to run "openclaw" which may not exist in test env
-        # This test is about error handling, not the actual command
-        try:
-            asyncio.run(
-                real_spawner.spawn_full_agent("test", "msg", task_id="t1")
-            )
-        except Exception:
-            pass  # Expected - command may fail
-
-
 # ---------------------------------------------------------------------------
 # T4: session 清理
 # ---------------------------------------------------------------------------
@@ -164,44 +98,29 @@ class TestAcquireFirst:
     """E11: #07.1 Acquire-First 重构后的 Phase 0-4 测试"""
 
     def test_phase0_revive_before_acquire(self, spawner):
-        """E11.1 Phase 0: timeout/failed 状态 → revive → acquire 成功
-
-        Phase 0 在 counter acquire 之前执行，修复 timeout/failed 的 session。
-        这里用 dry_run spawner 测试 _revive_session 的调用路径。
-        """
-        # 在 dry_run 模式下，Phase 0 检查 session state
-        # 如果 session 不存在（正常情况），status=None → 不触发 revive
-        # 测试正常路径：不超时/失败 → 直接 acquire + spawn
+        """E11.1 Phase 0: timeout/failed 状态 → revive → acquire 成功"""
         session_id = asyncio.run(
             spawner.spawn_full_agent(
                 "test-agent", "do something",
                 task_id="t1", use_main_session=True,
             )
         )
-        assert session_id  # dry_run 返回 "main" 或 session_id
+        assert session_id
 
     def test_phase0_stuck_detection(self, spawner):
-        """E11.2 Phase 0: status=running + lock PID 死 → revive
-
-        当 session state 显示 running 但 lock PID 已死，Phase 0 应自动 revive。
-        """
-        # 模拟：通过 mock _check_session_state 返回假死状态
+        """E11.2 Phase 0: status=running + lock PID 死 → revive"""
         original_check = spawner._check_session_state
         call_count = [0]
 
         def mock_check(agent_id):
             call_count[0] += 1
             if call_count[0] == 1:
-                # Phase 0: 假死状态
                 return {"status": "running", "lock_pid_alive": False}
-            # Phase 2: revive 后正常
             return {"status": "idle", "lock_pid_alive": False}
 
         spawner._check_session_state = mock_check
 
         revive_called = [False]
-        original_revive = spawner._revive_session
-
         def mock_revive(agent_id):
             revive_called[0] = True
             return True
@@ -217,22 +136,16 @@ class TestAcquireFirst:
         assert revive_called[0], "Phase 0 should have called _revive_session for stuck session"
 
     def test_phase1_counter_acquire_exclusive(self, spawner):
-        """E11.3 Phase 1: counter acquire 互斥
-
-        同一 agent 并发 spawn → 第二个 AgentBusyError(reason 含 counter)。
-        使用 max_concurrent_sessions=1 确保同 agent 第二次 acquire 失败。
-        """
+        """E11.3 Phase 1: counter acquire 互斥"""
         from src.daemon.counter import ActiveAgentCounter
         counter = ActiveAgentCounter(max_global=5, max_concurrent_sessions=1)
         spawner.counter = counter
 
-        # 第一次 acquire 成功
         session_id = asyncio.run(
             spawner.spawn_full_agent("test-agent", "task", task_id="t1")
         )
         assert session_id
 
-        # 第二次 acquire → counter blocked（同 agent 已有活跃 session）
         with pytest.raises(AgentBusyError) as exc_info:
             asyncio.run(
                 spawner.spawn_full_agent("test-agent", "task2", task_id="t2")
@@ -240,15 +153,11 @@ class TestAcquireFirst:
         assert "counter" in exc_info.value.reason or "blocked" in exc_info.value.reason
 
     def test_phase2_session_check_under_lock(self, spawner):
-        """E11.4 Phase 2: session check 在锁保护下执行
-
-        counter acquire 后 → session locked → release counter → AgentBusyError。
-        """
+        """E11.4 Phase 2: session check 在锁保护下执行"""
         from src.daemon.counter import ActiveAgentCounter
         counter = ActiveAgentCounter(max_global=5, max_per_agent=1)
         spawner.counter = counter
 
-        # 模拟 session locked
         spawner._check_session_state = lambda agent_id: {
             "status": "idle",
             "lock_pid_alive": True,
@@ -263,14 +172,10 @@ class TestAcquireFirst:
                 )
             )
         assert "session_locked" in exc_info.value.reason
-        # counter 应被 release（不会泄漏）
         assert counter.global_active == 0
 
     def test_phase2_multiple_blockers(self, spawner):
-        """E11.5 Phase 2: 多 blocker 并列收集
-
-        session locked + compact 同时存在 → detail.blockers 包含两者。
-        """
+        """E11.5 Phase 2: 多 blocker 并列收集"""
         spawner._check_session_state = lambda agent_id: {
             "status": "idle",
             "lock_pid_alive": True,
@@ -292,10 +197,7 @@ class TestAcquireFirst:
         assert "session_compacting" in blocker_reasons
 
     def test_phase3_on_checks_passed_exception_rollback(self, spawner):
-        """E11.6 Phase 3: on_checks_passed 抛异常 → counter 自动 release
-
-        回调异常不应导致 counter 泄漏。
-        """
+        """E11.6 Phase 3: on_checks_passed 抛异常 → counter 自动 release"""
         from src.daemon.counter import ActiveAgentCounter
         counter = ActiveAgentCounter(max_global=5, max_per_agent=1)
         spawner.counter = counter
@@ -311,7 +213,6 @@ class TestAcquireFirst:
                     on_checks_passed=bad_callback,
                 )
             )
-        # counter 应被 release
         assert counter.global_active == 0
 
 
@@ -323,10 +224,7 @@ class TestCompactHanging:
     """E13: compact_hanging 后不标 failed，只 release counter → 任务保持 working"""
 
     def test_compact_hanging_releases_counter(self):
-        """E13.1: compact 超限 → compact_hanging → release counter → 任务保持 working
-
-        compact_hanging outcome 时 counter 应被 release，任务不应被标 failed。
-        """
+        """E13.1: compact 超限 → compact_hanging → release counter → 任务保持 working"""
         from src.daemon.counter import ActiveAgentCounter
         from src.blackboard.models import Task as TaskModel
         counter = ActiveAgentCounter(max_global=5, max_per_agent=1)
@@ -340,7 +238,6 @@ class TestCompactHanging:
             spawner = AgentSpawner(db_path=db_path, dry_run=True)
             spawner.counter = counter
 
-            # 模拟 compact_hanging outcome 的 on_complete
             outcomes = []
             async def mock_on_complete(aid, outcome):
                 outcomes.append((aid, outcome))
@@ -350,7 +247,6 @@ class TestCompactHanging:
                 on_complete=mock_on_complete,
             ))
 
-            # 手动模拟 counter release（compact_hanging 路径）
             counter.release("test-agent", sid)
             assert counter.global_active == 0
         finally:
@@ -358,15 +254,11 @@ class TestCompactHanging:
                 db_path.unlink()
 
     def test_retry_agent_busy_releases_counter(self, spawner):
-        """E13.3: _do_retry 遇 AgentBusyError → release counter → 任务保持 working
-
-        retry 遇 session busy 时应 release counter，不持有。
-        """
+        """E13.3: _do_retry 遇 AgentBusyError → release counter → 任务保持 working"""
         from src.daemon.counter import ActiveAgentCounter
         counter = ActiveAgentCounter(max_global=5, max_per_agent=1)
         spawner.counter = counter
 
-        # 模拟 retry 场景：spawn 遇 AgentBusyError
         spawner._check_session_state = lambda agent_id: {
             "status": "running",
             "lock_pid_alive": True,
@@ -380,7 +272,6 @@ class TestCompactHanging:
                     task_id="t1", use_main_session=True,
                 )
             )
-        # counter 不应泄漏
         assert counter.global_active == 0
 
 
@@ -397,7 +288,6 @@ class TestAgentBusyErrorClassification:
         counter = ActiveAgentCounter(max_global=1, max_per_agent=1)
         spawner.counter = counter
 
-        # Fill counter
         asyncio.run(counter.acquire("test-agent"))
 
         with pytest.raises(AgentBusyError) as exc_info:
@@ -411,19 +301,14 @@ class TestAgentBusyErrorClassification:
         """E14.2: session locked/running/compacting → 具体 reason + detail.blockers"""
         test_cases = [
             {
-                # session locked: lock PID alive + not expired
                 "state": {"status": "idle", "lock_pid_alive": True, "lock_expired": False},
                 "expected": "session_locked",
             },
             {
-                # session running: status=running + lock alive
-                # 注意：running + lock alive 同时触发 session_locked 和 session_running，
-                # primary_reason 取第一个 blocker（session_locked）
                 "state": {"status": "running", "lock_pid_alive": True, "lock_expired": False},
-                "expected": "session_locked",  # session_locked 排在 session_running 之前
+                "expected": "session_locked",
             },
             {
-                # session compacting: recent compact
                 "state": {"status": "idle", "lock_pid_alive": False, "recent_compact": True},
                 "expected": "session_compacting",
             },
@@ -448,11 +333,7 @@ class TestAgentBusyErrorClassification:
                 f"Case {i}: expected detail to be set"
 
     def test_session_running_in_blockers(self, spawner):
-        """session_running 出现在 blockers 列表中（session_locked 优先）
-
-        status=running + lock alive → session_locked 和 session_running 同时存在，
-        session_locked 排第一（primary_reason），但 blockers 列表包含 session_running。
-        """
+        """session_running 出现在 blockers 列表中（session_locked 优先）"""
         spawner._check_session_state = lambda aid: {
             "status": "running", "lock_pid_alive": True, "lock_expired": False,
         }
@@ -464,9 +345,7 @@ class TestAgentBusyErrorClassification:
                     task_id="t1", use_main_session=True,
                 )
             )
-        # primary_reason 是 session_locked
         assert "session_locked" in exc_info.value.reason
-        # blockers 列表包含 session_running
         blockers = exc_info.value.detail.get("blockers", [])
         blocker_reasons = [b[0] for b in blockers]
         assert "session_running" in blocker_reasons
@@ -494,22 +373,15 @@ class TestPhase25AndStuck:
     """司马懿评审遗漏 #1 + session_stuck 遗漏补充"""
 
     def test_phase25_stuck_fallback(self, spawner):
-        """Phase 0 不触发（status 非 running），Phase 2 检测到假死 → revive → 成功 spawn
-
-        Phase 2.5 是 #07 v1.1 加的兜底：Phase 0 时 session 正常（idle），
-        但 Phase 2 检查时变为 running + lock PID 死。Phase 2.5 应 revive → 重检 → idle → spawn 成功。
-        """
+        """Phase 2.5: Phase 2 检测到假死 → revive → 成功 spawn"""
         call_count = [0]
 
         def mock_check(agent_id):
             call_count[0] += 1
             if call_count[0] <= 1:
-                # Phase 0: 正常 idle，不触发 revive
                 return {"status": "idle", "lock_pid_alive": False}
             if call_count[0] == 2:
-                # Phase 2: 假死（Phase 0 和 Phase 2 之间进程变 stuck）
                 return {"status": "running", "lock_pid_alive": False}
-            # Phase 2.5 重检：revive 后恢复 idle
             return {"status": "idle", "lock_pid_alive": False}
 
         spawner._check_session_state = mock_check
@@ -521,7 +393,6 @@ class TestPhase25AndStuck:
 
         spawner._revive_session = mock_revive
 
-        # Phase 2.5 应触发 revive → 重检 → idle → 正常 spawn
         session_id = asyncio.run(
             spawner.spawn_full_agent(
                 "test-agent", "task",
@@ -529,31 +400,11 @@ class TestPhase25AndStuck:
             )
         )
         assert revive_called[0], "Phase 2.5 should have revived stuck session"
-        assert session_id  # spawn 成功
+        assert session_id
 
     def test_session_stuck_after_failed_revive(self, spawner):
-        """Phase 2.5 revive 失败 → session_stuck
-
-        假死 revive 后 status 仍为 running → AgentBusyError(session_stuck)。
-        """
-        call_count = [0]
-
-        def mock_check(agent_id):
-            call_count[0] += 1
-            if call_count[0] <= 1:
-                return {"status": "idle", "lock_pid_alive": False}
-            # Phase 2: 假死
-            return {"status": "running", "lock_pid_alive": False}
-
-        spawner._check_session_state = mock_check
-
-        # revive 后 session 仍 stuck
-        def mock_revive(agent_id):
-            return True  # revive 成功但 mock_check 不变，下次仍返回 running
-
-        # 让 revive 后 check 返回 stuck
+        """Phase 2.5 revive 失败 → session_stuck"""
         revive_and_check_count = [0]
-        original_check = mock_check
 
         def mock_check_v2(agent_id):
             revive_and_check_count[0] += 1
@@ -561,11 +412,10 @@ class TestPhase25AndStuck:
                 return {"status": "idle", "lock_pid_alive": False}
             if revive_and_check_count[0] == 2:
                 return {"status": "running", "lock_pid_alive": False}
-            # revive 后重检：仍 stuck
             return {"status": "running", "lock_pid_alive": False}
 
         spawner._check_session_state = mock_check_v2
-        spawner._revive_session = mock_revive
+        spawner._revive_session = lambda agent_id: True
 
         with pytest.raises(AgentBusyError) as exc_info:
             asyncio.run(