fix: stabilize native streaming and improve bot controls

src/langbot/libs/wecom_ai_bot_api/api.py

@@ -2123,7 +2123,7 @@ class WecomBotClient:
 
         Args:
             msg_id: 原始企业微信消息 ID。
-            content: 模型产生的片段内容。
+            content: 流水线传入的累计正文快照。
             is_final: 是否为最终片段。
 
         Returns:
@@ -2138,12 +2138,8 @@ class WecomBotClient:
             return False
 
         previous_content = self._stream_last_content.get(msg_id, '')
-        if previous_content and content.startswith(previous_content):
-            next_content = content
-        elif previous_content and not content:
-            next_content = previous_content
-        else:
-            next_content = previous_content + content if previous_content else content
+        # Pipeline delivery provides a snapshot, not a provider delta.
+        next_content = content or previous_content
         if len(next_content) > _MAX_STREAM_CONTENT_CHARS:
             next_content = next_content[-_MAX_STREAM_CONTENT_CHARS:]
 

src/langbot/libs/wecom_ai_bot_api/ws_client.py

@@ -11,6 +11,7 @@ Official Node.js SDK: https://github.com/WecomTeam/aibot-node-sdk
 from __future__ import annotations
 
 import asyncio
+import hashlib
 import json
 import secrets
 import time
@@ -134,6 +135,9 @@ class WecomBotWsClient:
         self._stream_ids: dict[str, str] = {}  # msg_id -> req_id|stream_id
         # Dedup: skip sending when content hasn't changed
         self._stream_last_content: dict[str, str] = {}  # msg_id -> last content sent
+        # Preserve completion identity after removing active stream mappings.
+        # A runner may emit both a final delta and a completed message snapshot.
+        self._completed_stream_content: dict[str, str] = {}
         # Stream session info for feedback tracking
         self._stream_sessions: dict[str, dict] = {}  # msg_id -> session info
         # Feedback tracking: feedback_id -> session info
@@ -661,16 +665,15 @@ class WecomBotWsClient:
         """
         key = self._stream_ids.get(msg_id)
         if not key:
+            if is_final and self._completed_stream_content.get(msg_id) == hashlib.sha256(content.encode()).hexdigest():
+                return True
             return False
         req_id, stream_id = key.split('|', 1)
         try:
             previous_content = self._stream_last_content.get(msg_id, '')
-            if previous_content and content.startswith(previous_content):
-                next_content = content
-            elif previous_content and not content:
-                next_content = previous_content
-            else:
-                next_content = previous_content + content if previous_content else content
+            # Pipeline delivery provides complete snapshots, including edits
+            # that need not start with the previous text (e.g. reasoning tags).
+            next_content = content or previous_content
             if len(next_content) > _MAX_STREAM_CONTENT_CHARS:
                 next_content = next_content[-_MAX_STREAM_CONTENT_CHARS:]
 
@@ -720,6 +723,8 @@ class WecomBotWsClient:
             await self.reply_stream(req_id, stream_id, next_content, finish=is_final, feedback_id=feedback_id)
             self._stream_last_content[msg_id] = next_content
             if is_final:
+                self._completed_stream_content[msg_id] = hashlib.sha256(next_content.encode()).hexdigest()
+                self._cap_mapping(self._completed_stream_content, _STREAM_CACHE_MAX)
                 self._stream_ids.pop(msg_id, None)
                 self._stream_last_content.pop(msg_id, None)
                 self._stream_sessions.pop(msg_id, None)

src/langbot/pkg/box/service.py

@@ -811,6 +811,10 @@ class BoxService:
         to the sandbox (and vice-versa). It is ``None`` / not a local dir for
         E2B and remote runtimes, where we must fall back to the exec channel.
         """
+        # secure_fs relies on POSIX openat/O_NOFOLLOW. On Windows use the
+        # sandbox execution channel rather than misclassifying EACCES as a link.
+        if os.open not in os.supports_dir_fd or not hasattr(os, 'O_NOFOLLOW'):
+            return None
         root = self._tenant_workspace(self._query_execution_context(query))
         if not root or not os.path.isdir(root) or os.path.islink(root):
             return None

src/langbot/pkg/pipeline/process/handlers/chat.py

@@ -6,6 +6,7 @@ import traceback
 
 
 from .. import handler
+from ..stream_results import coalesce_stream_results
 from ... import entities
 from ... import plugin_diagnostics
 
@@ -132,7 +133,10 @@ class ChatMessageHandler(handler.MessageHandler):
 
                 # Use AgentRunOrchestrator to run the agent
                 # This replaces direct runner lookup and PluginRunnerWrapper
-                async for result in self.ap.agent_run_orchestrator.run_from_query(query):
+                results = self.ap.agent_run_orchestrator.run_from_query(query)
+                if is_stream:
+                    results = coalesce_stream_results(results)
+                async for result in results:
                     has_result = True
                     self._check_response_size(result)
 
@@ -149,6 +153,11 @@ class ChatMessageHandler(handler.MessageHandler):
                         )
                         result.attachments = attachments
 
+                    if is_stream and result.tool_calls:
+                        # A model round ending in tool calls does not end the
+                        # reply. Keep the native stream open for the answer.
+                        result = result.model_copy(update={'is_final': False})
+
                     result.resp_message_id = str(resp_message_id)
 
                     # For streaming mode, pop previous response before adding new chunk
@@ -192,6 +201,11 @@ class ChatMessageHandler(handler.MessageHandler):
 
                     if is_stream:
                         yield entities.StageProcessResult(result_type=entities.ResultType.CONTINUE, new_query=query)
+                        if result.is_final and not result.tool_calls:
+                            # Native adapters release their card on completion.
+                            # A distinct later message needs a fresh card identity.
+                            is_create_card = False
+                            resp_message_id = uuid.uuid4()
 
                 # Log final summary after streaming completes
                 if is_stream:

src/langbot/pkg/pipeline/process/stream_results.py

@@ -0,0 +1,44 @@
+"""Coalesce a runner's terminal delta and completed snapshot before delivery."""
+
+from collections.abc import AsyncIterable, AsyncIterator
+
+from langbot_plugin.api.entities.builtin.provider.message import Message, MessageChunk
+
+
+async def coalesce_stream_results(
+    results: AsyncIterable[Message | MessageChunk],
+) -> AsyncIterator[Message | MessageChunk]:
+    pending: MessageChunk | None = None
+    snapshot = ''
+    async for result in results:
+        if pending is not None:
+            # Completion snapshots can add attachments and usage to the final
+            # delta. Deliver that richer result once, before closing the card.
+            content = pending.all_content if pending.all_content is not None else pending.content
+            same_message = (
+                isinstance(result, Message)
+                and result.role == pending.role
+                and result.tool_calls == pending.tool_calls
+                and (not content or result.content == content)
+            )
+            if not same_message:
+                yield pending
+            pending = None
+            snapshot = ''
+
+        if isinstance(result, MessageChunk) and isinstance(result.content, (str, type(None))):
+            # Runner output is already cumulative (unlike raw provider deltas).
+            # Explicit all_content also supports runners forwarding raw deltas.
+            # Never guess delta semantics from a missing all_content field.
+            snapshot = result.all_content if result.all_content is not None else (result.content or snapshot)
+            result = result.model_copy(update={'all_content': snapshot})
+        elif isinstance(result, Message):
+            snapshot = ''
+
+        if isinstance(result, MessageChunk) and result.is_final:
+            pending = result
+        else:
+            yield result
+
+    if pending is not None:
+        yield pending

src/langbot/pkg/platform/adapters/wecom/manifest.yaml

@@ -8,9 +8,9 @@ metadata:
     zh_Hans: 企业微信
     zh_Hant: 企業微信
   description:
-    en_US: WeCom application message adapter with event-driven orchestration support
-    zh_Hans: 企业微信内部应用消息适配器，支持事件驱动编排
-    zh_Hant: 企業微信內部應用訊息適配器，支援事件驅動編排
+    en_US: Serve employees through a custom WeCom app
+    zh_Hans: 通过企业自建应用，为内部员工提供服务
+    zh_Hant: 透過企業自建應用，為內部員工提供服務
   icon: wecom.png
 
 spec:

src/langbot/pkg/platform/adapters/wecombot/adapter.py

@@ -148,6 +148,16 @@ class WecomBotAdapter(WecomBotAPIMixin, abstract_platform_adapter.AbstractPlatfo
         raw = await self.bot.send_message(str(target_id), content)
         return platform_events.MessageResult(raw={'result': raw})
 
+    async def reply_to_event(
+        self, *, context: dict, target_type: str, target_id: str, message: platform_message.MessageChain
+    ) -> platform_events.MessageResult:
+        """Bind Agent replies to the originating callback and its loading stream."""
+        message_id = ((context.get('delivery') or {}).get('reply_target') or {}).get('message_id')
+        if context.get('event_type') == 'message.received' and message_id:
+            source = await self.get_message(target_type, target_id, message_id)
+            return await self.reply_message(source, message)
+        return await self.send_message(target_type, target_id, message)
+
     async def reply_message(
         self,
         message_source: platform_events.MessageEvent,
@@ -160,9 +170,14 @@ class WecomBotAdapter(WecomBotAPIMixin, abstract_platform_adapter.AbstractPlatfo
         items = await WecomBotMessageConverter.yiri2target(message)
         content = self._join_text_components(items)
         raw = None
+        from .stream_text import format_stream_text
+
+        content = format_stream_text(content)
         if not self.config.get('enable-webhook', False) and event.get('req_id'):
             if content:
-                raw = await self.bot.reply_text(event.get('req_id'), content)
+                finished = await self.bot.push_stream_chunk(event.message_id, content, is_final=True)
+                if not finished:
+                    raw = await self.bot.reply_text(event.get('req_id'), content)
             for item in self._iter_media_components(items):
                 await self._send_media(self.bot, event.get('req_id'), item)
         else:
@@ -181,7 +196,9 @@ class WecomBotAdapter(WecomBotAPIMixin, abstract_platform_adapter.AbstractPlatfo
         if not isinstance(event, WecomBotEvent):
             raise ValueError('WeComBot reply_message_chunk requires a WecomBotEvent source object')
         items = await WecomBotMessageConverter.yiri2target(message)
-        content = self._join_text_components(items)
+        from .stream_text import format_stream_text
+
+        content = format_stream_text(self._join_text_components(items))
         success = await self.bot.push_stream_chunk(event.message_id, content, is_final=is_final)
         if not success and is_final and not self.config.get('enable-webhook', False) and event.get('req_id'):
             await self.bot.reply_text(event.get('req_id'), content)

src/langbot/pkg/platform/adapters/wecombot/manifest.yaml

@@ -8,9 +8,9 @@ metadata:
     zh_Hans: 企业微信智能机器人
     zh_Hant: 企業微信智慧機器人
   description:
-    en_US: WeCom AI Bot adapter with event-driven orchestration support
-    zh_Hans: 企业微信智能机器人适配器，支持事件驱动编排、长连接和 Webhook 两种接入方式
-    zh_Hant: 企業微信智慧機器人適配器，支援事件驅動編排、長連線和 Webhook 兩種接入方式
+    en_US: Answer questions and assist teams in WeCom chats and groups
+    zh_Hans: 在企业微信单聊和群聊中提供智能问答与协作
+    zh_Hant: 在企業微信單聊和群聊中提供智慧問答與協作
   icon: wecombot.png
 
 spec:
@@ -27,10 +27,9 @@ spec:
         zh_Hans: 一键创建机器人
         zh_Hant: 一鍵建立機器人
       description:
-        en_US: 'Scan QR code with WeCom to automatically create a bot and fill in BotId and Secret. Note: Robot Name needs to
-          be filled in manually.'
-        zh_Hans: 使用企业微信扫码自动创建机器人并填写 BotId 和 Secret。注意：机器人名称需手动填写。
-        zh_Hant: 使用企業微信掃碼自動建立機器人並填寫 BotId 和 Secret。注意：機器人名稱需手動填寫。
+        en_US: Scan with WeCom to create a bot and fill in BotId and Secret automatically.
+        zh_Hans: 使用企业微信扫码自动创建机器人并填写 BotId 和 Secret。
+        zh_Hant: 使用企業微信掃碼自動建立機器人並填寫 BotId 和 Secret。
       type: qr-code-login
       login_platform: wecombot
       required: false
@@ -60,8 +59,12 @@ spec:
         en_US: Robot Name
         zh_Hans: 机器人名称
         zh_Hant: 機器人名稱
+      description:
+        en_US: Optional. Only removes matching @name text in group messages; bot mentions are identified by BotId.
+        zh_Hans: 可选。仅用于清理群消息中的 @名称 文本，机器人提及通过 BotId 识别。
+        zh_Hant: 選填。僅用於清理群訊息中的 @名稱 文字，機器人提及透過 BotId 識別。
       type: string
-      required: true
+      required: false
       default: ""
     - name: enable-webhook
       label:
@@ -103,6 +106,10 @@ spec:
       type: string
       required: false
       default: ""
+      show_if:
+        field: enable-webhook
+        operator: neq
+        value: true
     - name: Corpid
       label:
         en_US: Corpid
@@ -115,6 +122,10 @@ spec:
       type: string
       required: false
       default: ""
+      show_if:
+        field: enable-webhook
+        operator: eq
+        value: true
     - name: Token
       label:
         en_US: Token
@@ -127,18 +138,26 @@ spec:
       type: string
       required: false
       default: ""
+      show_if:
+        field: enable-webhook
+        operator: eq
+        value: true
     - name: EncodingAESKey
       label:
         en_US: EncodingAESKey
         zh_Hans: 消息加解密密钥 (EncodingAESKey)
         zh_Hant: 訊息加解密密鑰 (EncodingAESKey)
       description:
-        en_US: Required for webhook mode. Optional for WebSocket mode when encrypted files need to be decrypted.
-        zh_Hans: Webhook 模式必填。WebSocket 模式下如需解密文件则填写。
-        zh_Hant: Webhook 模式必填。WebSocket 模式下如需解密檔案則填寫。
+        en_US: Required for webhook mode. Long connections use the decryption key included in each message.
+        zh_Hans: Webhook 模式必填。长连接模式使用消息自带的解密密钥。
+        zh_Hant: Webhook 模式必填。長連線模式使用訊息自帶的解密金鑰。
       type: string
       required: false
       default: ""
+      show_if:
+        field: enable-webhook
+        operator: eq
+        value: true
     - name: enable-stream-reply
       label:
         en_US: Enable Stream Reply

src/langbot/pkg/platform/adapters/wecombot/message_converter.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import datetime
+import re
 
 from langbot.libs.wecom_ai_bot_api.wecombotevent import WecomBotEvent
 import langbot_plugin.api.definition.abstract.platform.adapter as abstract_platform_adapter
@@ -46,12 +47,14 @@ class WecomBotMessageConverter(abstract_platform_adapter.AbstractMessageConverte
             platform_message.Source(id=event.message_id, time=datetime.datetime.now()),
         ]
         if event.type == 'group' and event.ai_bot_id:
+            # Group callbacks identify the mentioned bot by ID, independently
+            # of its display name or the optional text-cleanup configuration.
             components.append(platform_message.At(target=event.ai_bot_id))
 
         if event.content:
             content = event.content
-            if bot_name:
-                content = content.replace(f'@{bot_name}', '').strip()
+            if event.type == 'group' and bot_name:
+                content = re.sub(rf'(?<!\S)@{re.escape(bot_name)}(?=\s|$)', '', content).strip()
             if content:
                 components.append(platform_message.Plain(text=content))
 

src/langbot/pkg/platform/adapters/wecombot/stream_text.py

@@ -0,0 +1,31 @@
+"""Render multi-round reasoning as one WeCom thinking section."""
+
+import re
+
+
+def format_stream_text(content: str) -> str:
+    if content.strip().lower() in {'<', '<t', '<th', '<thi', '<thin', '<think'}:
+        return ''
+    if '<think' not in content.lower() and '</think' not in content.lower():
+        return content
+    # Every invocation receives the full snapshot, so split tags may simply
+    # wait for the next snapshot instead of leaking into the visible answer.
+    content = re.sub(r'<(?:/?t(?:h(?:i(?:n(?:k)?)?)?)?|/?)$', '', content, flags=re.IGNORECASE)
+    tokens = re.split(r'(</?think\s*>)', content, flags=re.IGNORECASE)
+    thoughts: list[str] = []
+    answers: list[str] = []
+    thinking = False
+    for token in tokens:
+        if re.fullmatch(r'<think\s*>', token, flags=re.IGNORECASE):
+            thinking = True
+        elif re.fullmatch(r'</think\s*>', token, flags=re.IGNORECASE):
+            thinking = False
+        elif token.strip():
+            (thoughts if thinking else answers).append(token.strip())
+    if not thoughts:
+        return '\n\n'.join(answers)
+    reasoning = '\n\n'.join(thoughts)
+    answer = '\n\n'.join(answers)
+    if thinking and not answer:
+        return f'<think>\n{reasoning}'
+    return f'<think>\n{reasoning}\n</think>' + (f'\n\n{answer}' if answer else '')

src/langbot/pkg/platform/adapters/wecomcs/manifest.yaml

@@ -8,9 +8,9 @@ metadata:
     zh_Hans: 企业微信客服
     zh_Hant: 企業微信客服
   description:
-    en_US: WeCom Customer Service adapter with event-driven orchestration support
-    zh_Hans: 企业微信客服适配器，支持事件驱动编排，通过统一 Webhook 接收客服会话消息
-    zh_Hant: 企業微信客服適配器，支援事件驅動編排，透過統一 Webhook 接收客服會話訊息
+    en_US: Support customers on WeChat through WeChat Customer Service
+    zh_Hans: 通过微信客服，为微信用户提供咨询与售后服务
+    zh_Hant: 透過微信客服，為微信使用者提供諮詢與售後服務
   icon: wecom.png
 
 spec:

tests/unit_tests/box/test_attachment_host_capabilities.py

@@ -0,0 +1,23 @@
+from unittest.mock import AsyncMock, Mock
+
+import pytest
+
+from langbot.pkg.box.service import BoxService
+from langbot_plugin.box.client import BoxRuntimeClient
+from tests.unit_tests.box.test_box_service import make_app, make_query
+
+
+@pytest.mark.asyncio
+async def test_host_without_openat_uses_sandbox_attachment_channel(monkeypatch, tmp_path):
+    monkeypatch.setattr('os.supports_dir_fd', set())
+    service = BoxService(make_app(Mock()), client=Mock(spec=BoxRuntimeClient))
+    service._available = True
+    service._tenant_workspace = Mock(return_value=str(tmp_path))
+    service._read_outbox_host = Mock(side_effect=AssertionError('POSIX host reader must not run'))
+    service._read_outbox_via_exec = AsyncMock(return_value=[])
+    service._clear_outbox = AsyncMock()
+    query = make_query()
+    assert service._host_query_dir(service.OUTBOX_SUBDIR, query) is None
+    assert await service.collect_outbound_attachments(query) == []
+    service._read_outbox_via_exec.assert_awaited_once_with(query)
+    service._clear_outbox.assert_awaited_once_with(query, None)

tests/unit_tests/pipeline/test_chat_handler.py

@@ -263,6 +263,57 @@ class TestChatMessageHandlerReal:
 class TestChatHandlerStreaming:
     """Tests for streaming behavior."""
 
+    @pytest.mark.asyncio
+    @pytest.mark.parametrize('with_tool_call', [False, True])
+    async def test_terminal_delta_and_snapshot_close_card_once(
+        self, fake_app, mock_event_ctx, set_runner, with_tool_call
+    ):
+        from tests.factories import text_query
+        from langbot_plugin.api.entities.builtin.provider.message import Message, MessageChunk
+        from langbot_plugin.api.entities.builtin.platform.message import File, MessageChain
+
+        fake_app.plugin_connector.emit_event = AsyncMock(return_value=mock_event_ctx)
+        query = text_query('stream test')
+        query.adapter = Mock()
+        query.adapter.is_stream_output_supported = AsyncMock(return_value=True)
+        query.adapter.create_message_card = AsyncMock()
+        attachments = MessageChain([File(url='https://example.com/file.txt', name='file.txt')])
+
+        class StreamRunner:
+            def __init__(self, app, config):
+                pass
+
+            async def run(self, query):
+                if with_tool_call:
+                    yield MessageChunk(
+                        role='assistant',
+                        is_final=True,
+                        tool_calls=[
+                            {'id': 'call-1', 'type': 'function', 'function': {'name': 'search', 'arguments': '{}'}}
+                        ],
+                    )
+                yield MessageChunk(role='assistant', content='hello')
+                yield MessageChunk(role='assistant', content='hello world', is_final=True)
+                completed = Message(role='assistant', content='hello world')
+                completed.attachments = attachments
+                yield completed
+
+        set_runner(StreamRunner)
+        handler = get_chat_handler().ChatMessageHandler(fake_app)
+        handler._ensure_conversation_for_history = AsyncMock()
+        delivered = []
+        async for stage_result in handler.handle(query):
+            assert stage_result.result_type == get_entities().ResultType.CONTINUE
+            delivered.append(query.resp_messages[-1])
+        assert [result.content for result in delivered if result.content] == ['hello', 'hello world']
+        if with_tool_call:
+            assert delivered[0].tool_calls
+            assert not delivered[0].is_final
+        assert sum(result.is_final for result in delivered) == 1
+        assert delivered[-1].attachments is attachments
+        assert len({result.resp_message_id for result in delivered}) == 1
+        query.adapter.create_message_card.assert_awaited_once()
+
     @pytest.mark.asyncio
     async def test_streaming_chunks_collected(self, fake_app, mock_event_ctx, set_runner):
         """Streaming produces multiple results."""

tests/unit_tests/pipeline/test_stream_results.py

@@ -0,0 +1,91 @@
+import pytest
+
+from langbot_plugin.api.entities.builtin.provider.message import Message, MessageChunk
+from langbot_plugin.api.entities.builtin.platform.message import File, MessageChain
+from langbot.pkg.pipeline.process.stream_results import coalesce_stream_results
+
+
+async def source(*items):
+    for item in items:
+        yield item
+
+
+@pytest.mark.asyncio
+async def test_terminal_snapshot_is_delivered_once_with_attachments():
+    text = '<think>reasoning</think>answer'
+    completed = Message(role='assistant', content=text)
+    completed.attachments = MessageChain([File(url='https://example.com/report.txt', name='report.txt')])
+    results = [
+        item
+        async for item in coalesce_stream_results(
+            source(
+                MessageChunk(role='assistant', content='answer', all_content=text, is_final=True),
+                completed,
+            )
+        )
+    ]
+    assert results == [completed]
+    assert results[0].attachments is completed.attachments
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize('with_snapshot', [False, True])
+async def test_progressive_chunks_are_not_buffered(with_snapshot):
+    first = MessageChunk(role='assistant', content='hel', all_content='hel')
+    final = MessageChunk(role='assistant', content='lo', all_content='hello', is_final=True)
+    completed = Message(role='assistant', content='hello')
+    stream = coalesce_stream_results(source(first, final, *([completed] if with_snapshot else [])))
+    assert await anext(stream) == first
+    assert [item async for item in stream] == [completed if with_snapshot else final]
+
+
+@pytest.mark.asyncio
+async def test_completed_only_and_distinct_messages_are_preserved():
+    final = MessageChunk(role='assistant', content='first', is_final=True)
+    other = Message(role='assistant', content='second')
+    results = [item async for item in coalesce_stream_results(source(final, other))]
+    assert [item.content for item in results] == ['first', 'second']
+    assert [item async for item in coalesce_stream_results(source(other))] == [other]
+
+
+@pytest.mark.asyncio
+async def test_blank_terminal_delta_uses_completed_snapshot():
+    completed = Message(role='assistant', content='answer')
+    results = [
+        item
+        async for item in coalesce_stream_results(
+            source(
+                MessageChunk(role='assistant', is_final=True),
+                completed,
+            )
+        )
+    ]
+    assert results == [completed]
+
+
+@pytest.mark.asyncio
+async def test_failures_propagate_without_successful_final():
+    async def failing():
+        yield MessageChunk(role='assistant', content='partial')
+        raise RuntimeError('provider failed')
+
+    stream = coalesce_stream_results(failing())
+    assert (await anext(stream)).content == 'partial'
+    with pytest.raises(RuntimeError, match='provider failed'):
+        await anext(stream)
+
+
+@pytest.mark.asyncio
+async def test_cumulative_stream_ends_with_last_snapshot_even_when_final_is_blank():
+    results = [
+        item
+        async for item in coalesce_stream_results(
+            source(
+                MessageChunk(role='assistant', content='hel'),
+                MessageChunk(role='assistant', content='hello'),
+                MessageChunk(role='assistant', is_final=True),
+            )
+        )
+    ]
+    assert [item.all_content for item in results] == ['hel', 'hello', 'hello']
+    assert results[-1].is_final

tests/unit_tests/platform/test_lark_eba_adapter.py

@@ -301,6 +301,7 @@ async def test_lark_event_converter_maps_group_and_private_message():
 @pytest.mark.asyncio
 async def test_lark_adapter_dispatches_and_caches_message_event():
     adapter = make_adapter()
+    adapter._get_bot_open_id = AsyncMock(return_value='bot-open-id')
     calls: list[platform_events.Event] = []
 
     async def listener(event, adapter):

tests/unit_tests/platform/test_native_stream_completion.py

@@ -0,0 +1,85 @@
+"""Exercise terminal runner snapshots against native adapter delivery methods."""
+
+import datetime
+import time
+from types import SimpleNamespace
+from unittest.mock import AsyncMock
+
+import pytest
+import telegram
+
+from langbot_plugin.api.entities.builtin.provider.message import Message, MessageChunk
+from langbot_plugin.api.entities.builtin.platform.message import MessageChain, Plain
+from langbot.pkg.pipeline.process.stream_results import coalesce_stream_results
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize('platform', ['lark', 'dingtalk', 'telegram', 'discord', 'qqofficial', 'wecombot'])
+async def test_native_stream_receives_one_terminal_snapshot(platform):
+    module = __import__(f'tests.unit_tests.platform.test_{platform}_eba_adapter', fromlist=['make_adapter'])
+    adapter = module.make_adapter()
+    object.__setattr__(adapter, 'reply_message', AsyncMock())
+    source = SimpleNamespace()
+    if platform == 'lark':
+        adapter.card_id_dict['response'] = 'card'
+        adapter.card_sequence_dict['card'] = 0
+        adapter._replace_streaming_card = AsyncMock()
+        source.source_platform_object = None
+    elif platform == 'dingtalk':
+        adapter.card_instance_id_dict['response'] = ('card', 'id')
+    elif platform == 'telegram':
+        source.source_platform_object = telegram.Update(
+            1,
+            message=telegram.Message(
+                1,
+                datetime.datetime.now(datetime.timezone.utc),
+                telegram.Chat(1, 'private'),
+                text='hi',
+            ),
+        )
+        adapter.bot = SimpleNamespace(send_message_draft=AsyncMock(), send_message=AsyncMock())
+        adapter.msg_stream_id['response'] = ('private', 1, False)
+    elif platform == 'discord':
+        adapter._stream_buffer['response'] = {
+            'channel': SimpleNamespace(send=AsyncMock(return_value=SimpleNamespace(content='hello', edit=AsyncMock()))),
+            'sent_message': None,
+            'last_content': '',
+            'chunk_count': 0,
+            'updated_at': time.time(),
+        }
+        channel = adapter._stream_buffer['response']['channel']
+    elif platform == 'qqofficial':
+        source = await module.QQOfficialEventConverter().target2yiri(module.qq_event())
+        await adapter.create_message_card('response', source)
+    else:
+        source = await module.WecomBotEventConverter().target2yiri(module.wecombot_event())
+
+    async def runner():
+        yield MessageChunk(role='assistant', content='hello', all_content='hello', is_final=True)
+        yield Message(role='assistant', content='hello')
+
+    count = 0
+    async for result in coalesce_stream_results(runner()):
+        chunk = MessageChunk(role='assistant', content=result.content, resp_message_id='response', is_final=True)
+        await adapter.reply_message_chunk(source, chunk, MessageChain([Plain(text=result.content)]), is_final=True)
+        count += 1
+    assert count == 1
+    adapter.reply_message.assert_not_awaited()
+    if platform == 'lark':
+        adapter._replace_streaming_card.assert_awaited_once_with('card', 'hello')
+        assert not adapter.card_id_dict
+    elif platform == 'dingtalk':
+        adapter.bot.send_card_message.assert_awaited_once_with('card', 'id', 'hello', True)
+        assert not adapter.card_instance_id_dict
+    elif platform == 'telegram':
+        adapter.bot.send_message.assert_awaited_once()
+        assert not adapter.msg_stream_id
+    elif platform == 'discord':
+        channel.send.assert_awaited_once_with('hello')
+        assert not adapter._stream_buffer
+    elif platform == 'qqofficial':
+        assert len([call for call in adapter.bot.sent if call[0] == 'stream']) == 1
+        assert not adapter._stream_ctx
+    else:
+        adapter.bot.push_stream_chunk.assert_awaited_once_with('msg-1', 'hello', is_final=True)
+        adapter.bot.reply_text.assert_not_awaited()

tests/unit_tests/platform/test_wecombot_eba_adapter.py

@@ -241,7 +241,7 @@ async def test_wecombot_event_converter_maps_private_and_group_messages_to_eba()
     assert private_event.adapter_name == 'wecombot-omni'
     assert private_event.chat_type == platform_entities.ChatType.PRIVATE
     assert private_event.chat_id == 'user-1'
-    assert str(private_event.message_chain) == 'hello'
+    assert str(private_event.message_chain) == '@EBA Bot hello'
 
     assert isinstance(group_event, platform_events.MessageReceivedEvent)
     assert group_event.chat_type == platform_entities.ChatType.GROUP
@@ -302,7 +302,9 @@ async def test_wecombot_send_reply_feedback_and_platform_api_use_underlying_clie
 
     source_event = await WecomBotEventConverter().target2yiri(wecombot_event())
     await adapter.reply_message(source_event, message)
-    adapter.bot.reply_text.assert_awaited_once_with('req-1', 'hello')
+    adapter.bot.push_stream_chunk.assert_awaited_once_with('msg-1', 'hello', is_final=True)
+    adapter.bot.reply_text.assert_not_awaited()
+    adapter.bot.push_stream_chunk.reset_mock()
 
     await adapter.reply_message_chunk(source_event, None, message, is_final=True)
     adapter.bot.push_stream_chunk.assert_awaited_once_with('msg-1', 'hello', is_final=True)

tests/unit_tests/platform/test_wecombot_stream_empty_final.py

@@ -117,3 +117,61 @@ async def test_real_final_chunk_is_sent_and_closes_stream():
     assert frames[-1]['stream']['content'] == 'the answer'
     assert frames[-1]['stream']['finish'] is True
     assert not client.stream_still_open()
+
+
+@pytest.mark.asyncio
+async def test_duplicate_final_is_acknowledged_without_sending_again():
+    client = RecordingClient()
+    client.seed_stream()
+    assert await client.push_stream_chunk('msg-1', 'answer', is_final=True)
+    assert await client.push_stream_chunk('msg-1', 'answer', is_final=True)
+    assert len(client.stream_frames()) == 1
+    assert not await client.push_stream_chunk('different-message', 'answer', is_final=True)
+    assert not await client.push_stream_chunk('msg-1', 'different answer', is_final=True)
+
+
+@pytest.mark.asyncio
+async def test_runner_cumulative_snapshots_reach_wire_once_without_concatenation():
+    from langbot.pkg.pipeline.process.stream_results import coalesce_stream_results
+    from langbot_plugin.api.entities.builtin.provider.message import Message, MessageChunk
+    from tests.unit_tests.platform.test_wecombot_eba_adapter import (
+        make_adapter,
+        wecombot_event,
+        WecomBotEventConverter,
+    )
+
+    client = RecordingClient()
+    client.seed_stream()
+    adapter = make_adapter()
+    adapter.bot = client
+    event = await WecomBotEventConverter().target2yiri(wecombot_event())
+    snapshots = ['<think>test', '<think>test</think>\nOK', '<think>test</think>\nOK!']
+
+    async def runner():
+        # LocalAgent's StreamingModelCaller sets content to the accumulated
+        # text, leaves all_content unset, then emits message.completed.
+        for index, text in enumerate(snapshots):
+            yield MessageChunk(role='assistant', content=text, msg_sequence=index + 1, is_final=index == 2)
+        yield Message(role='assistant', content=snapshots[-1])
+
+    async for result in coalesce_stream_results(runner()):
+        final = isinstance(result, Message) or result.is_final
+        if isinstance(result, MessageChunk):
+            result = result.model_copy(update={'content': result.all_content})
+        await adapter.reply_message_chunk(event, result, result.get_content_platform_message_chain(), is_final=final)
+
+    frames = client.stream_frames()
+    from langbot.pkg.platform.adapters.wecombot.stream_text import format_stream_text
+
+    assert [frame['stream']['content'] for frame in frames] == [format_stream_text(text) for text in snapshots]
+    assert [frame['stream']['finish'] for frame in frames] == [False, False, True]
+    assert len(client.replies) == 3  # No extra plain-text reply after completion.
+
+
+@pytest.mark.asyncio
+async def test_revised_snapshot_replaces_previous_text():
+    client = RecordingClient()
+    client.seed_stream()
+    await client.push_stream_chunk('msg-1', '<think>test</think>\nanswer')
+    await client.push_stream_chunk('msg-1', 'answer', is_final=True)
+    assert client.stream_frames()[-1]['stream']['content'] == 'answer'

tests/unit_tests/platform/test_wecombot_stream_text.py

@@ -0,0 +1,30 @@
+import pytest
+
+from langbot.pkg.platform.adapters.wecombot.stream_text import format_stream_text
+
+
+def test_multiple_rounds_have_one_thinking_section_and_separated_answers():
+    raw = '<think>one</think>Checking.<think>two</think>More checks.<think>three</think>Done.'
+    result = format_stream_text(raw)
+    assert result == '<think>\none\n\ntwo\n\nthree\n</think>\n\nChecking.\n\nMore checks.\n\nDone.'
+    assert format_stream_text(result) == result
+
+
+@pytest.mark.parametrize('suffix', ['<', '<t', '<thi', '<think', '</', '</thi'])
+def test_partial_tags_are_not_shown_in_answer(suffix):
+    result = format_stream_text('<think>one</think>Checking.' + suffix)
+    assert result.endswith('Checking.')
+
+
+def test_later_incomplete_thinking_remains_inside_single_section():
+    result = format_stream_text('<think>one</think>Checking.<think>two')
+    assert result == '<think>\none\n\ntwo\n</think>\n\nChecking.'
+
+
+def test_plain_markdown_is_unchanged():
+    text = '  hello\n\n```python\na < b\n```\n'
+    assert format_stream_text(text) == text
+
+
+def test_initial_partial_thinking_tag_waits_for_next_snapshot():
+    assert format_stream_text('<thi') == ''

tests/unit_tests/platform/test_wecombot_template_card.py

@@ -369,11 +369,13 @@ async def test_ws_stream_sends_cumulative_snapshots_to_wecom():
 
     assert await client.push_stream_chunk('msg-1', '你', is_final=False)
     assert await client.push_stream_chunk('msg-1', '你好', is_final=False)
+    assert await client.push_stream_chunk('msg-1', '<think>test</think>你好', is_final=False)
     assert await client.push_stream_chunk('msg-1', '你好', is_final=True)
 
     assert sent == [
         ('req-1', 'stream-1', '你', False),
         ('req-1', 'stream-1', '你好', False),
+        ('req-1', 'stream-1', '<think>test</think>你好', False),
         ('req-1', 'stream-1', '你好', True),
     ]
 
@@ -385,6 +387,7 @@ async def test_webhook_stream_queues_cumulative_snapshots_for_followups():
 
     assert await client.push_stream_chunk('msg-1', '你', is_final=False)
     assert await client.push_stream_chunk('msg-1', '你好', is_final=False)
+    assert await client.push_stream_chunk('msg-1', '<think>test</think>你好', is_final=False)
     assert await client.push_stream_chunk('msg-1', '你好', is_final=True)
 
     assert session.queue.qsize() == 1

web/src/app/home/bots/BotDetailContent.tsx

@@ -4,7 +4,7 @@ import { useNavigate } from 'react-router-dom';
 import { Tabs, TabsList, TabsTrigger, TabsContent } from '@/components/ui/tabs';
 import { Button } from '@/components/ui/button';
 import { Switch } from '@/components/ui/switch';
-import { Label } from '@/components/ui/label';
+import { Badge } from '@/components/ui/badge';
 import {
   Dialog,
   DialogContent,
@@ -22,7 +22,7 @@ import type { BotSessionMonitorHandle } from '@/app/home/bots/components/bot-ses
 import { httpClient } from '@/app/infra/http/HttpClient';
 import { useSidebarData } from '@/app/home/components/home-sidebar/SidebarDataContext';
 import { useTranslation } from 'react-i18next';
-import { Settings, FileText, Users, RefreshCw, Trash2 } from 'lucide-react';
+import { Settings, FileText, Users, RefreshCw, Trash2, Loader2 } from 'lucide-react';
 import { cn } from '@/lib/utils';
 import { toast } from 'sonner';
 import { showBotError } from './bot-error';
@@ -72,6 +72,7 @@ export default function BotDetailContent({ id }: { id: string }) {
 
   // Enable state managed here so the header switch works
   const [botEnabled, setBotEnabled] = useState(true);
+  const [isTogglingEnable, setIsTogglingEnable] = useState(false);
   const [enableLoaded, setEnableLoaded] = useState(false);
 
   // Fetch bot enable state
@@ -91,20 +92,25 @@ export default function BotDetailContent({ id }: { id: string }) {
 
   const handleEnableToggle = useCallback(
     async (checked: boolean) => {
+      if (isTogglingEnable) return;
       const prev = botEnabled;
+      setIsTogglingEnable(true);
       setBotEnabled(checked);
       try {
         await httpClient.updateBot(id, { enable: checked });
         setBot((current) =>
           current ? { ...current, enable: checked } : current,
         );
+        toast.success(t(checked ? 'bots.enableConfirmed' : 'bots.disableConfirmed'));
         refreshBots();
       } catch (error) {
         setBotEnabled(prev);
         showBotError(error, t('bots.setBotEnableError'), t);
+      } finally {
+        setIsTogglingEnable(false);
       }
     },
-    [id, botEnabled, refreshBots, t],
+    [id, botEnabled, isTogglingEnable, refreshBots, t],
   );
 
   function handleFormSubmit() {
@@ -210,18 +216,27 @@ export default function BotDetailContent({ id }: { id: string }) {
             </div>
             {enableLoaded && (
               <div className="flex items-center gap-2">
+                <Badge
+                  status={isTogglingEnable ? 'pending' : botEnabled ? 'enabled' : 'disabled'}
+                  className="gap-1.5 rounded-full"
+                  role="status"
+                  aria-live="polite"
+                >
+                  {isTogglingEnable
+                    ? <Loader2 className="animate-spin" aria-hidden="true" />
+                    : <span className="size-1.5 rounded-full bg-current" aria-hidden="true" />}
+                  {t(isTogglingEnable
+                    ? botEnabled ? 'bots.enabling' : 'bots.disabling'
+                    : botEnabled ? 'bots.enableConfirmed' : 'bots.disableConfirmed')}
+                </Badge>
                 <Switch
                   id="bot-enable-switch"
                   checked={botEnabled}
                   onCheckedChange={handleEnableToggle}
-                  disabled={!canManage}
+                  disabled={!canManage || isTogglingEnable}
+                  aria-label={t('common.enable')}
+                  aria-busy={isTogglingEnable}
                 />
-                <Label
-                  htmlFor="bot-enable-switch"
-                  className="text-sm text-muted-foreground cursor-pointer"
-                >
-                  {t('common.enable')}
-                </Label>
               </div>
             )}
           </div>
@@ -237,11 +252,11 @@ export default function BotDetailContent({ id }: { id: string }) {
               <Button
                 type="submit"
                 form="bot-form"
-                disabled={!enableLoaded || (!formDirty && botEnabled)}
+                disabled={!enableLoaded || !formDirty || isTogglingEnable}
                 className={activeTab !== 'config' ? 'invisible' : ''}
                 data-guide="bot-config-save"
               >
-                {t(botEnabled ? 'common.save' : 'bots.saveAndEnable')}
+                {t(formDirty && !botEnabled ? 'bots.saveAndEnable' : 'common.save')}
               </Button>
               <Button
                 type="button"

web/src/app/home/components/home-sidebar/HomeSidebar.tsx

@@ -800,7 +800,7 @@ function NavItems({
                       {(isBot || isMCP) && (
                         <span
                           className={cn(
-                            'absolute -bottom-0.5 -right-0.5 size-2 rounded-full border-2 border-popover',
+                            'absolute -bottom-1 -right-1 size-3 rounded-full border-2 border-popover',
                             isMCP
                               ? mcpStatusColor(item)
                               : item.enabled === false
@@ -868,7 +868,7 @@ function NavItems({
                             {(isBot || isMCP) && (
                               <span
                                 className={cn(
-                                  'absolute -bottom-0.5 -right-0.5 size-2 rounded-full border-2 border-sidebar',
+                                  'absolute -bottom-1 -right-1 size-3 rounded-full border-2 border-sidebar',
                                   isMCP
                                     ? mcpStatusColor(item)
                                     : item.enabled === false

web/src/app/home/monitoring/hooks/useMonitoringData.ts

@@ -43,7 +43,8 @@ export function useMonitoringData(filterState: FilterState) {
       requestId === requestIdRef.current &&
       getCurrentWorkspaceSnapshot()?.workspace.uuid === workspaceUuid;
     setLoading(true);
-    setError(null);
+    // Keep the failure view during retries; restore the dashboard only after
+    // a successful response, without briefly restarting its polling timer.
 
     try {
       const { startTime, endTime } = resolveMonitoringWindow(
@@ -385,6 +386,7 @@ export function useMonitoringData(filterState: FilterState) {
       ].sort((a, b) => b.timestamp.getTime() - a.timestamp.getTime());
 
       setData(transformedData);
+      setError(null);
       setRequestScope(scope);
     } catch (err) {
       if (!isCurrent()) return;

web/src/app/home/monitoring/page.tsx

@@ -167,7 +167,7 @@ function MonitoringPageContent() {
     intervalRef.current = refreshInterval;
     countdown.current = refreshInterval;
     setRemainingSeconds(refreshInterval);
-    if (!refreshInterval) return;
+    if (!refreshInterval || error) return;
     const timer = window.setInterval(() => {
       if (document.visibilityState === 'visible' && !refreshState.current.busy) {
         countdown.current = Math.max(0, countdown.current - 1);
@@ -179,7 +179,7 @@ function MonitoringPageContent() {
       }
     }, 1000);
     return () => window.clearInterval(timer);
-  }, [refreshInterval]);
+  }, [refreshInterval, error]);
 
   // State for expanded errors
   const [expandedErrorId, setExpandedErrorId] = useState<string | null>(null);
@@ -204,7 +204,7 @@ function MonitoringPageContent() {
   return (
     <div className="w-full h-full overflow-y-auto overflow-x-hidden">
       {/* Filters and Refresh Button - Sticky */}
-      <div className="sticky top-0 z-10 -mt-1 pb-5 pt-1 bg-background">
+      {!error && <div className="sticky top-0 z-10 -mt-1 pb-5 pt-1 bg-background">
         <div>
           <Card className="flex flex-col gap-4 p-4 xl:flex-row xl:items-end">
             <MonitoringFilters
@@ -264,6 +264,8 @@ function MonitoringPageContent() {
         </div>
       </div>
 
+      }
+
       {/* Content Area */}
       {error ? (
         <LoadErrorState title={t('monitoring.loadError')} onRetry={handleRefresh} busy={refreshing} />

web/src/i18n/locales/en-US.ts

@@ -822,6 +822,10 @@ const enUS = {
     earlier: 'Earlier',
     dateFormat: '{{month}}/{{day}}',
     setBotEnableError: 'Failed to set bot enable status',
+    enabling: 'Enabling…',
+    disabling: 'Disabling…',
+    enableConfirmed: 'Enabled',
+    disableConfirmed: 'Disabled',
     log: 'Log',
     configuration: 'Configuration',
     logs: 'Logs',

web/src/i18n/locales/es-ES.ts

@@ -838,6 +838,10 @@ const esES = {
     earlier: 'Anterior',
     dateFormat: '{{day}}/{{month}}',
     setBotEnableError: 'Error al establecer el estado de activación del Bot',
+    enabling: 'Activando…',
+    disabling: 'Desactivando…',
+    enableConfirmed: 'Activado',
+    disableConfirmed: 'Desactivado',
     log: 'Registro',
     configuration: 'Configuración',
     logs: 'Registros',

web/src/i18n/locales/ja-JP.ts

@@ -825,6 +825,10 @@ const jaJP = {
     earlier: 'それ以前',
     dateFormat: '{{month}}月{{day}}日',
     setBotEnableError: 'ボットの有効状態の設定に失敗しました',
+    enabling: '有効にしています…',
+    disabling: '無効にしています…',
+    enableConfirmed: '有効',
+    disableConfirmed: '無効',
     log: 'ログ',
     configuration: '設定',
     logs: 'ログ',

web/src/i18n/locales/ru-RU.ts

@@ -833,6 +833,10 @@ const ruRU = {
     earlier: 'Ранее',
     dateFormat: '{{day}}.{{month}}',
     setBotEnableError: 'Не удалось изменить статус бота',
+    enabling: 'Включение…',
+    disabling: 'Отключение…',
+    enableConfirmed: 'Включён',
+    disableConfirmed: 'Отключён',
     log: 'Журнал',
     configuration: 'Настройки',
     logs: 'Журналы',

web/src/i18n/locales/th-TH.ts

@@ -809,6 +809,10 @@ const thTH = {
     earlier: 'ก่อนหน้า',
     dateFormat: '{{day}}/{{month}}',
     setBotEnableError: 'ไม่สามารถตั้งค่าสถานะเปิดใช้งาน Bot ได้',
+    enabling: 'กำลังเปิดใช้งาน…',
+    disabling: 'กำลังปิดใช้งาน…',
+    enableConfirmed: 'เปิดใช้งานแล้ว',
+    disableConfirmed: 'ปิดใช้งานแล้ว',
     log: 'บันทึก',
     configuration: 'การกำหนดค่า',
     logs: 'บันทึก',

web/src/i18n/locales/vi-VN.ts

@@ -823,6 +823,10 @@ const viVN = {
     earlier: 'Trước đó',
     dateFormat: '{{day}}/{{month}}',
     setBotEnableError: 'Thiết lập trạng thái bật Bot thất bại',
+    enabling: 'Đang bật…',
+    disabling: 'Đang tắt…',
+    enableConfirmed: 'Đã bật',
+    disableConfirmed: 'Đã tắt',
     log: 'Nhật ký',
     configuration: 'Cấu hình',
     logs: 'Nhật ký',

web/src/i18n/locales/zh-Hans.ts

@@ -775,6 +775,10 @@ const zhHans = {
     earlier: '更久之前',
     dateFormat: '{{month}}月{{day}}日',
     setBotEnableError: '设置机器人启用状态失败',
+    enabling: '正在启用…',
+    disabling: '正在禁用…',
+    enableConfirmed: '已启用',
+    disableConfirmed: '已禁用',
     log: '日志',
     configuration: '配置',
     logs: '日志',

web/src/i18n/locales/zh-Hant.ts

@@ -773,6 +773,10 @@ const zhHant = {
     earlier: '更久之前',
     dateFormat: '{{month}}月{{day}}日',
     setBotEnableError: '設定機器人啟用狀態失敗',
+    enabling: '正在啟用…',
+    disabling: '正在停用…',
+    enableConfirmed: '已啟用',
+    disableConfirmed: '已停用',
     log: '日誌',
     configuration: '設定',
     logs: '日誌',