feat: improve rpa video timing and narration

Co-authored-by: Cursor <cursoragent@cursor.com>

tests/test_video_session.py

@@ -8,13 +8,16 @@ import sys
 import tempfile
 import unittest
 from pathlib import Path
-from unittest.mock import MagicMock, patch
+from unittest.mock import AsyncMock, MagicMock, patch
 
 from jiangchang_skill_core import config
 from jiangchang_skill_core.rpa import video_session as vs
 from jiangchang_skill_core.rpa.video_session import (
     RpaVideoSession,
     _StepEntry,
+    _INTRO_BUFFER_SECONDS,
+    _INTRO_STABILIZE_SECONDS,
+    _OUTRO_BUFFER_SECONDS,
     _compose_capture_mp4,
     _music_fade_params,
     _select_voiceover_clips,
@@ -225,15 +228,38 @@ class TestVoiceoverTextFilter(unittest.TestCase):
 
     def test_min_gap_two_seconds(self) -> None:
         entries = [
-            _StepEntry(0.0, "开始采集"),
-            _StepEntry(1.0, "启动浏览器"),
+            _StepEntry(0.0, "获取1688账号"),
+            _StepEntry(1.0, "连接CDP"),
             _StepEntry(3.0, "检查登录状态"),
         ]
         clips = _select_voiceover_clips(entries)
         self.assertEqual(len(clips), 2)
-        self.assertEqual(clips[0][1], "开始采集")
+        self.assertEqual(clips[0][1], "获取1688账号")
         self.assertEqual(clips[1][1], "检查登录状态")
 
+    def test_important_bypasses_min_gap(self) -> None:
+        entries = [
+            _StepEntry(0.0, "定位搜索框"),
+            _StepEntry(0.5, "输入关键词：宁德电池"),
+            _StepEntry(1.0, "点击搜索"),
+            _StepEntry(1.5, "等待搜索结果"),
+        ]
+        clips = _select_voiceover_clips(entries)
+        texts = [t for _, t in clips]
+        self.assertEqual(len(clips), 4)
+        self.assertIn("输入关键词：宁德电池", texts)
+        self.assertIn("点击搜索", texts)
+
+    def test_noise_still_skipped_with_important_nearby(self) -> None:
+        entries = [
+            _StepEntry(0.0, "点击搜索"),
+            _StepEntry(0.5, "跳过重复店铺"),
+            _StepEntry(1.0, "写入联系人库"),
+        ]
+        clips = _select_voiceover_clips(entries)
+        texts = [t for _, t in clips]
+        self.assertEqual(texts, ["点击搜索"])
+
 
 class TestComposeCaptureMp4(unittest.TestCase):
     def setUp(self) -> None:
@@ -343,6 +369,134 @@ class TestComposeCaptureMp4(unittest.TestCase):
         self.assertNotIn("-stream_loop", cmd)
 
 
+class TestVideoSessionLifecycle(unittest.TestCase):
+    def setUp(self) -> None:
+        config.reset_cache()
+
+    def tearDown(self) -> None:
+        os.environ.pop("OPENCLAW_RECORD_VIDEO", None)
+        config.reset_cache()
+
+    @patch("jiangchang_skill_core.rpa.video_session.asyncio.sleep", new_callable=AsyncMock)
+    @patch.object(RpaVideoSession, "_start_ffmpeg_capture")
+    @patch.object(RpaVideoSession, "_stop_ffmpeg_capture")
+    @patch.object(RpaVideoSession, "finalize", new_callable=AsyncMock)
+    def test_enter_starts_capture_before_title(
+        self,
+        mock_finalize: AsyncMock,
+        mock_stop: MagicMock,
+        mock_start: MagicMock,
+        mock_sleep: AsyncMock,
+    ) -> None:
+        os.environ["OPENCLAW_RECORD_VIDEO"] = "1"
+        call_order: list[str] = []
+        mock_start.side_effect = lambda: call_order.append("start_capture")
+
+        with tempfile.TemporaryDirectory() as tmp:
+            config.reset_cache()
+            session = RpaVideoSession(
+                skill_slug="test",
+                skill_data_dir=tmp,
+                batch_id="batch_intro",
+                title="开始执行测试",
+            )
+            original_add = session.add_step
+
+            def _track_add(text: str, *, duration: float = 4.0) -> None:
+                call_order.append(f"add_step:{text}")
+                original_add(text, duration=duration)
+
+            with patch.object(session, "add_step", side_effect=_track_add):
+                asyncio.run(session.__aenter__())
+
+        self.assertEqual(call_order[0], "start_capture")
+        self.assertIn("add_step:开始执行测试", call_order)
+        mock_sleep.assert_awaited()
+        total_sleep = sum(c.args[0] for c in mock_sleep.await_args_list if c.args)
+        self.assertAlmostEqual(total_sleep, _INTRO_BUFFER_SECONDS)
+
+    @patch("jiangchang_skill_core.rpa.video_session.asyncio.sleep", new_callable=AsyncMock)
+    @patch.object(RpaVideoSession, "_start_ffmpeg_capture")
+    def test_enter_no_title_still_buffers(
+        self,
+        mock_start: MagicMock,
+        mock_sleep: AsyncMock,
+    ) -> None:
+        os.environ["OPENCLAW_RECORD_VIDEO"] = "1"
+        with tempfile.TemporaryDirectory() as tmp:
+            config.reset_cache()
+            session = RpaVideoSession(
+                skill_slug="test",
+                skill_data_dir=tmp,
+                batch_id="batch_no_title",
+            )
+            with patch.object(session, "add_step") as mock_add:
+                asyncio.run(session.__aenter__())
+            mock_start.assert_called_once()
+            mock_add.assert_not_called()
+            total_sleep = sum(c.args[0] for c in mock_sleep.await_args_list if c.args)
+            self.assertAlmostEqual(total_sleep, _INTRO_BUFFER_SECONDS)
+
+    @patch("jiangchang_skill_core.rpa.video_session.asyncio.sleep", new_callable=AsyncMock)
+    @patch.object(RpaVideoSession, "_stop_ffmpeg_capture")
+    @patch.object(RpaVideoSession, "finalize", new_callable=AsyncMock)
+    def test_exit_closing_title_before_stop(
+        self,
+        mock_finalize: AsyncMock,
+        mock_stop: MagicMock,
+        mock_sleep: AsyncMock,
+    ) -> None:
+        os.environ["OPENCLAW_RECORD_VIDEO"] = "1"
+        call_order: list[str] = []
+        mock_stop.side_effect = lambda: call_order.append("stop_capture")
+
+        with tempfile.TemporaryDirectory() as tmp:
+            config.reset_cache()
+            session = RpaVideoSession(
+                skill_slug="test",
+                skill_data_dir=tmp,
+                batch_id="batch_outro",
+                closing_title="任务已完成",
+            )
+            original_add = session.add_step
+
+            def _track_add(text: str, *, duration: float = 4.0) -> None:
+                call_order.append(f"add_step:{text}")
+                original_add(text, duration=duration)
+
+            with patch.object(session, "add_step", side_effect=_track_add):
+                asyncio.run(session.__aexit__(None, None, None))
+
+        self.assertIn("add_step:任务已完成", call_order)
+        self.assertEqual(call_order[-1], "stop_capture")
+        total_sleep = sum(c.args[0] for c in mock_sleep.await_args_list if c.args)
+        self.assertAlmostEqual(total_sleep, _OUTRO_BUFFER_SECONDS)
+
+    @patch("jiangchang_skill_core.rpa.video_session.asyncio.sleep", new_callable=AsyncMock)
+    @patch.object(RpaVideoSession, "_stop_ffmpeg_capture")
+    @patch.object(RpaVideoSession, "finalize", new_callable=AsyncMock)
+    def test_exit_no_closing_title_still_outro_buffer(
+        self,
+        mock_finalize: AsyncMock,
+        mock_stop: MagicMock,
+        mock_sleep: AsyncMock,
+    ) -> None:
+        os.environ["OPENCLAW_RECORD_VIDEO"] = "1"
+        with tempfile.TemporaryDirectory() as tmp:
+            config.reset_cache()
+            session = RpaVideoSession(
+                skill_slug="test",
+                skill_data_dir=tmp,
+                batch_id="batch_outro_empty",
+            )
+            with patch.object(session, "add_step") as mock_add:
+                asyncio.run(session.__aexit__(None, None, None))
+            mock_add.assert_not_called()
+            mock_stop.assert_called_once()
+            total_sleep = sum(c.args[0] for c in mock_sleep.await_args_list if c.args)
+            self.assertAlmostEqual(total_sleep, _OUTRO_BUFFER_SECONDS)
+
+
 class TestFinalizeDegradation(unittest.TestCase):
     def setUp(self) -> None:
         config.reset_cache()