test(e2e): cover all routes/modes, render dashboard in CI, fix issues found (P1 of #31)

.claude/skills/agent-observability/scripts/build_dashboard_data.py

@@ -522,6 +522,12 @@ def build_devflow_runs(project_root: Path, cost_by_slug: dict[str, float] | None
                 "exec": info.get("executor") or "-", "status": status,
                 "retry_count": retry_count, "duration": duration,
             })
+        # 只有所有阶段都已完成或跳过时才算“已完成”；停在两个阶段之间（等待审批、
+        # 下一阶段尚未派发）的运行仍是进行中，不能因为没有 in_progress 就显示完成。
+        if overall == "completed" and any(
+            stage["status"] not in {"completed", "skipped"} for stage in stages_out
+        ):
+            overall = "running"
         # 按真实起止时间跨度算总时长（而不是逐阶段 duration 相加）——阶段之间可能
         # 有重叠（比如 TASK-05 的 started_at 早于 SOLO 的 completed_at），相加会
         # 把重叠部分重复计入，虚高于源数据本身反映的运行时长。

.claude/skills/agent-observability/scripts/core/agent_identity.py

@@ -162,7 +162,8 @@ def infer_identity_from_messages(messages: list[dict[str, Any]]) -> tuple[Option
         next_target = _extract_next_target(payload)
 
         if from_role == "main" and mailbox_role and mailbox_role != "main":
-            if not msg.get("is_shutdown"):
+            # A standby assignment only creates the member; it is not active yet.
+            if not msg.get("is_shutdown") and not msg.get("is_standby"):
                 current = mailbox_role
                 dispatched = mailbox_role
                 evidence = f"dispatch:{mailbox_name}"
@@ -521,6 +522,7 @@ def _normalize_message(item: Any, mailbox_name: str, mailbox_role: str | None, i
         "ts": ts,
         "index": index,
         "is_shutdown": _is_shutdown_message(payload, summary, text),
+        "is_standby": _is_standby_message(payload, text),
     }
 
 
@@ -561,6 +563,14 @@ def _is_shutdown_message(payload: dict[str, Any] | None, summary: str, text: Any
     return "shutdown" in summary_low or "shutdown request" in text_low
 
 
+def _is_standby_message(payload: dict[str, Any] | None, text: Any) -> bool:
+    """识别“先创建成员、待命等待唤醒”的消息；这类消息不代表成员已开始工作。"""
+    if isinstance(payload, dict) and (payload.get("event") or payload.get("dispatch_prompt")):
+        return False
+    text_low = text.lower() if isinstance(text, str) else ""
+    return "待命" in text_low or "standby" in text_low or "stand by" in text_low
+
+
 def _parse_timestamp(value: Any) -> float:
     """把消息时间字段解析成 Unix 时间戳。"""
     if not isinstance(value, str) or not value.strip():

.claude/skills/agent-observability/scripts/core/cls_sink.py

@@ -338,6 +338,10 @@ def load_config(cwd: str | None = None) -> CLSConfig:
 
 def mirror_record(record: dict[str, Any], *, cwd: str | None = None) -> bool:
     config = load_config(cwd)
+    if not (config.endpoint or config.topic_id):
+        # CLS mirroring was never configured: stay silent instead of writing a
+        # debug line into the skill directory for every local event.
+        return False
     if not config.ready:
         _debug_write(
             {

.claude/skills/agent-observability/tests/test_build_dashboard_data.py

@@ -395,6 +395,36 @@ def test_solo_run_includes_task05_and_marks_phase0_completed(self):
             self.assertNotIn("TASK-01", stage_by_key)  # SOLO 路径不该混进 medium/large 的阶段
 
 
+class BuildDevflowRunsOverallStatusTests(unittest.TestCase):
+    """E2E 发现：刚开始或停在两个阶段之间的运行（没有 in_progress、也没有 failed）
+    曾被显示为“已完成”。"""
+
+    def _overall(self, stages):
+        with tempfile.TemporaryDirectory() as td:
+            project_root = Path(td)
+            run_dir = project_root / "artifacts" / "between-stages"
+            run_dir.mkdir(parents=True)
+            (run_dir / "workflow-state.json").write_text(json.dumps({
+                "version": "1.3", "task_slug": "between-stages", "size_class": "medium",
+                "current_stage": "TASK-02", "stages": stages,
+            }), encoding="utf-8")
+            return bdd.build_devflow_runs(project_root)[0]["overall"]
+
+    def test_run_waiting_between_stages_is_not_completed(self):
+        stages = {
+            "TASK-01": {"status": "completed", "executor": "main", "retry_count": 0},
+            "TASK-02": {"status": "completed", "executor": "architect", "retry_count": 0},
+            "TASK-03": {"status": "pending", "executor": "developer", "retry_count": 0},
+        }
+        self.assertEqual(self._overall(stages), "running")
+
+    def test_fully_finished_run_is_completed(self):
+        names = ("TASK-01", "TASK-02", "TASK-03", "CODE-REVIEW", "TASK-04", "TASK-05")
+        stages = {name: {"status": "completed", "executor": "x", "retry_count": 0} for name in names}
+        stages["TASK-05"]["status"] = "skipped"
+        self.assertEqual(self._overall(stages), "completed")
+
+
 class BuildDevflowRunsMediumWithLingeringSoloKeyTests(unittest.TestCase):
     """用户反馈复现：workflow-state.json 模板固定给每次运行都写一份 SOLO stage
     （见 assets/workflow-state-template.json 里 SOLO.description 的说明：

.claude/skills/agent-observability/tests/test_cls_sink.py

@@ -17,6 +17,22 @@
 
 
 class CLSSinkTests(unittest.TestCase):
+    def setUp(self):
+        # Keep the suite hermetic: never write debug logs into the source tree and
+        # do not depend on the optional Node SDK being checked out next to the repo.
+        sandbox = tempfile.TemporaryDirectory()
+        self.addCleanup(sandbox.cleanup)
+        sdk_entry = Path(sandbox.name) / "tencentcloud-cls-sdk-js" / "dist" / "index.js"
+        sdk_entry.parent.mkdir(parents=True)
+        sdk_entry.write_text("module.exports = {};\n", encoding="utf-8")
+        for name, value in (
+            ("sdk_entry_path", sdk_entry),
+            ("debug_log_path", Path(sandbox.name) / "cls-push-debug.ndjson"),
+        ):
+            patcher = mock.patch.object(cls_sink, name, return_value=value)
+            patcher.start()
+            self.addCleanup(patcher.stop)
+
     def _write_cls_env(self, root: Path, *extra_lines: str) -> None:
         lines = [
             "CLS_TOPIC_ID=test-topic-id",
@@ -171,6 +187,18 @@ def test_mirror_record_writes_success_debug_when_uploader_succeeds(self):
         self.assertEqual(entries[-1]["stage"], "uploader_ok")
         self.assertEqual(entries[-1]["method"], "python_api_v3")
 
+    def test_unconfigured_cls_stays_silent(self):
+        with tempfile.TemporaryDirectory() as td:
+            debug_path = Path(td) / "debug.ndjson"
+            with mock.patch.object(cls_sink, "debug_log_path", return_value=debug_path), \
+                 mock.patch.object(cls_sink.subprocess, "run") as run, \
+                 mock.patch.dict(os.environ, {}, clear=True):
+                ok = cls_sink.mirror_record({"event": "start", "sid": "s-off"}, cwd=td)
+
+            self.assertFalse(ok)
+            run.assert_not_called()
+            self.assertFalse(debug_path.exists())
+
     def test_emit_keeps_local_log_when_cls_push_raises(self):
         with tempfile.TemporaryDirectory() as td:
             log_path = Path(td) / "metrics.ndjson"

.codebuddy/skills/agent-observability/scripts/build_dashboard_data.py

@@ -522,6 +522,12 @@ def build_devflow_runs(project_root: Path, cost_by_slug: dict[str, float] | None
                 "exec": info.get("executor") or "-", "status": status,
                 "retry_count": retry_count, "duration": duration,
             })
+        # 只有所有阶段都已完成或跳过时才算“已完成”；停在两个阶段之间（等待审批、
+        # 下一阶段尚未派发）的运行仍是进行中，不能因为没有 in_progress 就显示完成。
+        if overall == "completed" and any(
+            stage["status"] not in {"completed", "skipped"} for stage in stages_out
+        ):
+            overall = "running"
         # 按真实起止时间跨度算总时长（而不是逐阶段 duration 相加）——阶段之间可能
         # 有重叠（比如 TASK-05 的 started_at 早于 SOLO 的 completed_at），相加会
         # 把重叠部分重复计入，虚高于源数据本身反映的运行时长。

.codebuddy/skills/agent-observability/scripts/core/agent_identity.py

@@ -162,7 +162,8 @@ def infer_identity_from_messages(messages: list[dict[str, Any]]) -> tuple[Option
         next_target = _extract_next_target(payload)
 
         if from_role == "main" and mailbox_role and mailbox_role != "main":
-            if not msg.get("is_shutdown"):
+            # A standby assignment only creates the member; it is not active yet.
+            if not msg.get("is_shutdown") and not msg.get("is_standby"):
                 current = mailbox_role
                 dispatched = mailbox_role
                 evidence = f"dispatch:{mailbox_name}"
@@ -521,6 +522,7 @@ def _normalize_message(item: Any, mailbox_name: str, mailbox_role: str | None, i
         "ts": ts,
         "index": index,
         "is_shutdown": _is_shutdown_message(payload, summary, text),
+        "is_standby": _is_standby_message(payload, text),
     }
 
 
@@ -561,6 +563,14 @@ def _is_shutdown_message(payload: dict[str, Any] | None, summary: str, text: Any
     return "shutdown" in summary_low or "shutdown request" in text_low
 
 
+def _is_standby_message(payload: dict[str, Any] | None, text: Any) -> bool:
+    """识别“先创建成员、待命等待唤醒”的消息；这类消息不代表成员已开始工作。"""
+    if isinstance(payload, dict) and (payload.get("event") or payload.get("dispatch_prompt")):
+        return False
+    text_low = text.lower() if isinstance(text, str) else ""
+    return "待命" in text_low or "standby" in text_low or "stand by" in text_low
+
+
 def _parse_timestamp(value: Any) -> float:
     """把消息时间字段解析成 Unix 时间戳。"""
     if not isinstance(value, str) or not value.strip():

.codebuddy/skills/agent-observability/scripts/core/cls_sink.py

@@ -338,6 +338,10 @@ def load_config(cwd: str | None = None) -> CLSConfig:
 
 def mirror_record(record: dict[str, Any], *, cwd: str | None = None) -> bool:
     config = load_config(cwd)
+    if not (config.endpoint or config.topic_id):
+        # CLS mirroring was never configured: stay silent instead of writing a
+        # debug line into the skill directory for every local event.
+        return False
     if not config.ready:
         _debug_write(
             {

.codebuddy/skills/agent-observability/tests/test_build_dashboard_data.py

@@ -395,6 +395,36 @@ def test_solo_run_includes_task05_and_marks_phase0_completed(self):
             self.assertNotIn("TASK-01", stage_by_key)  # SOLO 路径不该混进 medium/large 的阶段
 
 
+class BuildDevflowRunsOverallStatusTests(unittest.TestCase):
+    """E2E 发现：刚开始或停在两个阶段之间的运行（没有 in_progress、也没有 failed）
+    曾被显示为“已完成”。"""
+
+    def _overall(self, stages):
+        with tempfile.TemporaryDirectory() as td:
+            project_root = Path(td)
+            run_dir = project_root / "artifacts" / "between-stages"
+            run_dir.mkdir(parents=True)
+            (run_dir / "workflow-state.json").write_text(json.dumps({
+                "version": "1.3", "task_slug": "between-stages", "size_class": "medium",
+                "current_stage": "TASK-02", "stages": stages,
+            }), encoding="utf-8")
+            return bdd.build_devflow_runs(project_root)[0]["overall"]
+
+    def test_run_waiting_between_stages_is_not_completed(self):
+        stages = {
+            "TASK-01": {"status": "completed", "executor": "main", "retry_count": 0},
+            "TASK-02": {"status": "completed", "executor": "architect", "retry_count": 0},
+            "TASK-03": {"status": "pending", "executor": "developer", "retry_count": 0},
+        }
+        self.assertEqual(self._overall(stages), "running")
+
+    def test_fully_finished_run_is_completed(self):
+        names = ("TASK-01", "TASK-02", "TASK-03", "CODE-REVIEW", "TASK-04", "TASK-05")
+        stages = {name: {"status": "completed", "executor": "x", "retry_count": 0} for name in names}
+        stages["TASK-05"]["status"] = "skipped"
+        self.assertEqual(self._overall(stages), "completed")
+
+
 class BuildDevflowRunsMediumWithLingeringSoloKeyTests(unittest.TestCase):
     """用户反馈复现：workflow-state.json 模板固定给每次运行都写一份 SOLO stage
     （见 assets/workflow-state-template.json 里 SOLO.description 的说明：

.codebuddy/skills/agent-observability/tests/test_cls_sink.py

@@ -17,6 +17,22 @@
 
 
 class CLSSinkTests(unittest.TestCase):
+    def setUp(self):
+        # Keep the suite hermetic: never write debug logs into the source tree and
+        # do not depend on the optional Node SDK being checked out next to the repo.
+        sandbox = tempfile.TemporaryDirectory()
+        self.addCleanup(sandbox.cleanup)
+        sdk_entry = Path(sandbox.name) / "tencentcloud-cls-sdk-js" / "dist" / "index.js"
+        sdk_entry.parent.mkdir(parents=True)
+        sdk_entry.write_text("module.exports = {};\n", encoding="utf-8")
+        for name, value in (
+            ("sdk_entry_path", sdk_entry),
+            ("debug_log_path", Path(sandbox.name) / "cls-push-debug.ndjson"),
+        ):
+            patcher = mock.patch.object(cls_sink, name, return_value=value)
+            patcher.start()
+            self.addCleanup(patcher.stop)
+
     def _write_cls_env(self, root: Path, *extra_lines: str) -> None:
         lines = [
             "CLS_TOPIC_ID=test-topic-id",
@@ -171,6 +187,18 @@ def test_mirror_record_writes_success_debug_when_uploader_succeeds(self):
         self.assertEqual(entries[-1]["stage"], "uploader_ok")
         self.assertEqual(entries[-1]["method"], "python_api_v3")
 
+    def test_unconfigured_cls_stays_silent(self):
+        with tempfile.TemporaryDirectory() as td:
+            debug_path = Path(td) / "debug.ndjson"
+            with mock.patch.object(cls_sink, "debug_log_path", return_value=debug_path), \
+                 mock.patch.object(cls_sink.subprocess, "run") as run, \
+                 mock.patch.dict(os.environ, {}, clear=True):
+                ok = cls_sink.mirror_record({"event": "start", "sid": "s-off"}, cwd=td)
+
+            self.assertFalse(ok)
+            run.assert_not_called()
+            self.assertFalse(debug_path.exists())
+
     def test_emit_keeps_local_log_when_cls_push_raises(self):
         with tempfile.TemporaryDirectory() as td:
             log_path = Path(td) / "metrics.ndjson"

.cursor/skills/agent-observability/scripts/build_dashboard_data.py

@@ -522,6 +522,12 @@ def build_devflow_runs(project_root: Path, cost_by_slug: dict[str, float] | None
                 "exec": info.get("executor") or "-", "status": status,
                 "retry_count": retry_count, "duration": duration,
             })
+        # 只有所有阶段都已完成或跳过时才算“已完成”；停在两个阶段之间（等待审批、
+        # 下一阶段尚未派发）的运行仍是进行中，不能因为没有 in_progress 就显示完成。
+        if overall == "completed" and any(
+            stage["status"] not in {"completed", "skipped"} for stage in stages_out
+        ):
+            overall = "running"
         # 按真实起止时间跨度算总时长（而不是逐阶段 duration 相加）——阶段之间可能
         # 有重叠（比如 TASK-05 的 started_at 早于 SOLO 的 completed_at），相加会
         # 把重叠部分重复计入，虚高于源数据本身反映的运行时长。

.cursor/skills/agent-observability/scripts/core/agent_identity.py

@@ -162,7 +162,8 @@ def infer_identity_from_messages(messages: list[dict[str, Any]]) -> tuple[Option
         next_target = _extract_next_target(payload)
 
         if from_role == "main" and mailbox_role and mailbox_role != "main":
-            if not msg.get("is_shutdown"):
+            # A standby assignment only creates the member; it is not active yet.
+            if not msg.get("is_shutdown") and not msg.get("is_standby"):
                 current = mailbox_role
                 dispatched = mailbox_role
                 evidence = f"dispatch:{mailbox_name}"
@@ -521,6 +522,7 @@ def _normalize_message(item: Any, mailbox_name: str, mailbox_role: str | None, i
         "ts": ts,
         "index": index,
         "is_shutdown": _is_shutdown_message(payload, summary, text),
+        "is_standby": _is_standby_message(payload, text),
     }
 
 
@@ -561,6 +563,14 @@ def _is_shutdown_message(payload: dict[str, Any] | None, summary: str, text: Any
     return "shutdown" in summary_low or "shutdown request" in text_low
 
 
+def _is_standby_message(payload: dict[str, Any] | None, text: Any) -> bool:
+    """识别“先创建成员、待命等待唤醒”的消息；这类消息不代表成员已开始工作。"""
+    if isinstance(payload, dict) and (payload.get("event") or payload.get("dispatch_prompt")):
+        return False
+    text_low = text.lower() if isinstance(text, str) else ""
+    return "待命" in text_low or "standby" in text_low or "stand by" in text_low
+
+
 def _parse_timestamp(value: Any) -> float:
     """把消息时间字段解析成 Unix 时间戳。"""
     if not isinstance(value, str) or not value.strip():

.cursor/skills/agent-observability/scripts/core/cls_sink.py

@@ -338,6 +338,10 @@ def load_config(cwd: str | None = None) -> CLSConfig:
 
 def mirror_record(record: dict[str, Any], *, cwd: str | None = None) -> bool:
     config = load_config(cwd)
+    if not (config.endpoint or config.topic_id):
+        # CLS mirroring was never configured: stay silent instead of writing a
+        # debug line into the skill directory for every local event.
+        return False
     if not config.ready:
         _debug_write(
             {

.cursor/skills/agent-observability/tests/test_build_dashboard_data.py

@@ -395,6 +395,36 @@ def test_solo_run_includes_task05_and_marks_phase0_completed(self):
             self.assertNotIn("TASK-01", stage_by_key)  # SOLO 路径不该混进 medium/large 的阶段
 
 
+class BuildDevflowRunsOverallStatusTests(unittest.TestCase):
+    """E2E 发现：刚开始或停在两个阶段之间的运行（没有 in_progress、也没有 failed）
+    曾被显示为“已完成”。"""
+
+    def _overall(self, stages):
+        with tempfile.TemporaryDirectory() as td:
+            project_root = Path(td)
+            run_dir = project_root / "artifacts" / "between-stages"
+            run_dir.mkdir(parents=True)
+            (run_dir / "workflow-state.json").write_text(json.dumps({
+                "version": "1.3", "task_slug": "between-stages", "size_class": "medium",
+                "current_stage": "TASK-02", "stages": stages,
+            }), encoding="utf-8")
+            return bdd.build_devflow_runs(project_root)[0]["overall"]
+
+    def test_run_waiting_between_stages_is_not_completed(self):
+        stages = {
+            "TASK-01": {"status": "completed", "executor": "main", "retry_count": 0},
+            "TASK-02": {"status": "completed", "executor": "architect", "retry_count": 0},
+            "TASK-03": {"status": "pending", "executor": "developer", "retry_count": 0},
+        }
+        self.assertEqual(self._overall(stages), "running")
+
+    def test_fully_finished_run_is_completed(self):
+        names = ("TASK-01", "TASK-02", "TASK-03", "CODE-REVIEW", "TASK-04", "TASK-05")
+        stages = {name: {"status": "completed", "executor": "x", "retry_count": 0} for name in names}
+        stages["TASK-05"]["status"] = "skipped"
+        self.assertEqual(self._overall(stages), "completed")
+
+
 class BuildDevflowRunsMediumWithLingeringSoloKeyTests(unittest.TestCase):
     """用户反馈复现：workflow-state.json 模板固定给每次运行都写一份 SOLO stage
     （见 assets/workflow-state-template.json 里 SOLO.description 的说明：