fix(runtime): preserve explicit replies and report bot configuration errors

pyproject.toml

@@ -232,4 +232,4 @@ line-ending = "auto"
 
 [tool.uv.sources]
 # Development contract: update to the matching SDK release before publishing.
-langbot-plugin = { git = "https://github.com/langbot-app/langbot-plugin-sdk", rev = "f82b3ce935f9a33afee389fb39fe8dc29a45b615" }
+langbot-plugin = { git = "https://github.com/langbot-app/langbot-plugin-sdk", rev = "ca0671b81db5b2ed937d01158a3982de3fd8500f" }

src/langbot/pkg/agent/runner/orchestrator.py

@@ -8,6 +8,9 @@ import typing
 from langbot_plugin.api.entities.builtin.provider import message as provider_message
 from langbot_plugin.api.entities.builtin.pipeline import query as pipeline_query
 
+from langbot_plugin.entities.io.actions.enums import PluginToRuntimeAction
+
+from .reply_stream import ReplyStreamSession
 from ...core import app
 from ...api.http.context import ExecutionContext
 from ...pipeline.pool import get_query_execution_context
@@ -155,6 +158,16 @@ class AgentRunOrchestrator:
 
         state_context = build_state_context(event, binding, descriptor)
         run_id = context['run_id']
+        context['context']['available_apis']['reply_stream'] = hasattr(PluginToRuntimeAction, 'REPLY_STREAM') and any(
+            tool.get('tool_name') == 'event_reply' and tool.get('tool_type') == 'platform'
+            for tool in resources.get('tools', [])
+        )
+        reply_streams = ReplyStreamSession(
+            event,
+            adapter=(adapter_context or {}).get('_delivery_adapter') or getattr(execution_query, 'adapter', None),
+            source=(adapter_context or {}).get('_platform_event')
+            or getattr((adapter_context or {}).get('_query'), 'message_event', None),
+        )
         available_apis = context.get('context', {}).get('available_apis')
         run_authorization = {
             'runner_id': descriptor.id,
@@ -208,6 +221,7 @@ class AgentRunOrchestrator:
                 state_context=state_context,
                 execution_query=execution_query,
                 platform_context=freeze_platform_context(event),
+                reply_streams=reply_streams,
             )
 
             event_log_id = await self.journal.write_event_log(
@@ -358,6 +372,7 @@ class AgentRunOrchestrator:
             raise
         finally:
             session = await self._session_registry.unregister(run_id)
+            await reply_streams.close()
             pending_steering = session.get('steering_queue', []) if session else []
             if pending_steering:
                 try:

src/langbot/pkg/agent/runner/platform_tools.py

@@ -54,8 +54,10 @@ PLATFORM_TOOL_DEFINITIONS: tuple[PlatformToolDefinition, ...] = (
         'write',
         {'zh_Hans': '回复当前会话', 'en_US': 'Reply to current conversation'},
         {
-            'zh_Hans': '向触发当前事件的会话发送文本消息。目标由 LangBot 固定，Agent 无法改写。',
+            'zh_Hans': '向触发当前事件的会话发送回复或任务进度。目标由 LangBot 固定，Agent 无法改写。'
-            'en_US': 'Send text to the conversation that triggered this run. LangBot fixes the target.',
+            '需要发送消息时请调用此工具，Agent 的输出文本不会自动发送。',
+            'en_US': 'Send a reply or task progress to the conversation that triggered this run. LangBot fixes the target. '
+            'Call this tool to send a message; Agent output text is not sent automatically.',
         },
         _object_schema({'text': {**_TEXT, 'description': 'Reply text'}}, ['text']),
         ('message.*', 'friend.*', 'group.*', 'feedback.*'),
@@ -209,8 +211,9 @@ PLATFORM_TOOL_DEFINITIONS: tuple[PlatformToolDefinition, ...] = (
         'write',
         {'zh_Hans': '发送消息', 'en_US': 'Send message'},
         {
-            'zh_Hans': '使用当前机器人向指定用户或群组发送文本消息。',
+            'zh_Hans': '使用当前机器人向指定用户或群组发送回复或任务进度。Agent 的输出文本不会自动发送。',
-            'en_US': 'Send text to a specified person or group using the current bot.',
+            'en_US': 'Send a reply or task progress to a specified person or group using the current bot. '
+            'Agent output text is not sent automatically.',
         },
         _object_schema(
             {'target_type': _TARGET_TYPE, 'target_id': _ID, 'text': _TEXT}, ['target_type', 'target_id', 'text']

src/langbot/pkg/agent/runner/reply_stream.py

@@ -0,0 +1,162 @@
+"""Host-owned explicit replies using the existing adapter streaming lifecycle."""
+
+from __future__ import annotations
+
+import asyncio
+from dataclasses import dataclass, field
+from typing import Literal
+from uuid import UUID, uuid4
+
+import pydantic
+from langbot_plugin.api.entities.builtin.platform import events, message
+from langbot_plugin.api.entities.builtin.provider import message as provider_message
+
+
+class ReplyStreamRequest(pydantic.BaseModel):
+    model_config = pydantic.ConfigDict(extra='forbid')
+
+    stream_id: UUID
+    operation: Literal['update', 'finish', 'abort']
+    text: str = pydantic.Field(max_length=200_000, strict=True)
+
+
+@dataclass
+class _Reply:
+    delivery_id: str = field(default_factory=lambda: str(uuid4()))
+    text: str = ''
+    started: bool = False
+    native: bool = False
+    final_attempted: bool = False
+    sequence: int = 0
+    status: str = 'open'
+    result: dict | None = None
+
+
+class ReplyStreamSession:
+    """Ephemeral delivery state owned by one authorized run, never by plugin input."""
+
+    def __init__(self, event, adapter=None, source=None):
+        self.adapter = adapter
+        self.source = source.to_legacy_event() if isinstance(source, events.MessageReceivedEvent) else source
+        if not isinstance(self.source, events.MessageEvent):
+            self.source = None
+        self.target = dict(event.delivery.reply_target or {})
+        self.mock = (
+            event.delivery.surface == 'webui' and (event.delivery.platform_capabilities or {}).get('debug_mock') is True
+        )
+        self.mock_error = (
+            ((event.delivery.platform_capabilities or {}).get('mock_options') or {})
+            .get('errors', {})
+            .get('event_reply')
+        )
+        self._replies: dict[str, _Reply] = {}
+        self._lock = asyncio.Lock()
+        self._closed = False
+        self._active: set[asyncio.Task] = set()
+
+    async def apply(self, request: ReplyStreamRequest) -> dict:
+        task = asyncio.create_task(self._apply(request))
+        self._active.add(task)
+        try:
+            return await task
+        finally:
+            self._active.discard(task)
+
+    async def _apply(self, request: ReplyStreamRequest) -> dict:
+        async with self._lock:
+            if self._closed:
+                raise ValueError('Reply stream run has ended')
+            key = str(request.stream_id)
+            reply = self._replies.get(key)
+            if reply is None:
+                if len(self._replies) >= 16:
+                    raise ValueError('A run may create at most 16 reply streams')
+                reply = self._replies[key] = _Reply()
+            if reply.status != 'open':
+                if request.operation == 'update' or reply.status == 'failed':
+                    raise ValueError('Reply stream is closed')
+                return reply.result
+            try:
+                if request.operation != 'abort':
+                    reply.text = request.text
+                if self.mock_error:
+                    raise ValueError(str(self.mock_error))
+                if not self.mock and reply.text and not reply.started and request.operation != 'abort':
+                    if self.adapter is None:
+                        raise ValueError('This run has no platform delivery adapter')
+                    # Set before I/O: an uncertain create must never be retried as another send.
+                    reply.started = True
+                    reply.native = self.source is not None and await self.adapter.is_stream_output_supported()
+                    if reply.native:
+                        await self.adapter.create_message_card(reply.delivery_id, self.source)
+                if reply.started and reply.native:
+                    await self._update_native(key, reply, final=request.operation != 'update')
+                elif request.operation == 'finish' and reply.text and not self.mock:
+                    if self.source is not None:
+                        await self.adapter.reply_message(
+                            message_source=self.source, message=self._message(reply.text), quote_origin=False
+                        )
+                    else:
+                        target_type, target_id = self.target.get('target_type'), self.target.get('target_id')
+                        if not target_type or not target_id:
+                            raise ValueError('This event has no reply target')
+                        await self.adapter.send_message(str(target_type), str(target_id), self._message(reply.text))
+                if request.operation != 'update':
+                    reply.status = 'completed' if request.operation == 'finish' else 'cancelled'
+                reply.result = {
+                    'stream_id': key,
+                    'status': reply.status,
+                    'delivery': 'simulated' if self.mock else 'streaming' if reply.native else 'buffered',
+                    'mock': self.mock,
+                    **({'text': reply.text} if request.operation == 'finish' else {}),
+                }
+                return reply.result
+            except BaseException:
+                reply.status = 'failed'
+                raise
+
+    @staticmethod
+    def _message(text):
+        return message.MessageChain([message.Plain(text=text)])
+
+    async def _update_native(self, key, reply, *, final):
+        if final:
+            reply.final_attempted = True
+        reply.sequence += 1
+        chunk = provider_message.MessageChunk(
+            role='assistant',
+            content=reply.text,
+            all_content=reply.text,
+            resp_message_id=reply.delivery_id,
+            msg_sequence=reply.sequence,
+            is_final=final,
+        )
+        await self.adapter.reply_message_chunk(
+            message_source=self.source,
+            bot_message=chunk,
+            message=self._message(reply.text),
+            quote_origin=False,
+            is_final=final,
+        )
+
+    async def close(self):
+        """Close visible streams on cancellation; never send a buffered partial reply."""
+        self._closed = True
+        pending_requests = list(self._active)
+        for task in pending_requests:
+            task.cancel()
+        if pending_requests:
+            await asyncio.gather(*pending_requests, return_exceptions=True)
+        async with self._lock:
+            pending = [
+                self._update_native(key, reply, final=True)
+                for key, reply in self._replies.items()
+                if reply.started and reply.native and not reply.final_attempted
+            ]
+            try:
+                if pending:
+                    await asyncio.wait_for(asyncio.gather(*pending, return_exceptions=True), timeout=5)
+            except TimeoutError:
+                pass
+            finally:
+                self._replies.clear()

src/langbot/pkg/agent/runner/session_registry.py

@@ -77,6 +77,7 @@ class AgentRunSession(typing.TypedDict):
     plugin_identity: str  # author/name
     authorization: RunAuthorizationSnapshot
     status: AgentRunSessionStatus
+    reply_streams: typing.Any
     steering_queue: list[SteeringQueueItem]
 
 
@@ -115,6 +116,7 @@ class AgentRunSessionRegistry:
         state_context: dict[str, typing.Any] | None = None,
         execution_query: pipeline_query.Query | None = None,
         platform_context: dict[str, typing.Any] | None = None,
+        reply_streams: typing.Any = None,
     ) -> None:
         """Register a new agent run session.
 
@@ -167,6 +169,7 @@ class AgentRunSessionRegistry:
             'runner_id': runner_id,
             'query_id': query_id,
             'execution_query': execution_query,
+            'reply_streams': reply_streams,
             'plugin_identity': plugin_identity,
             'authorization': authorization,
             'status': {

src/langbot/pkg/api/http/controller/groups/platform/bots.py

@@ -3,11 +3,20 @@ from sqlalchemy.exc import IntegrityError
 
 from ....authz import Permission, has_permission
 from ....context import RequestContext
+from ....service.bot_errors import BotApplyError, bot_error_message
 from ... import group
 
 
 @group.group_class('bots', '/api/v1/platform/bots')
 class BotsRouterGroup(group.RouterGroup):
+    def _apply_error_response(self, exc: BotApplyError):
+        request_id = self.request_id()
+        logger = getattr(self.ap, 'logger', self.quart_app.logger)
+        logger.error(f'Bot configuration apply failed request_id={request_id} bot_uuid={exc.bot_uuid}', exc_info=True)
+        return quart.jsonify(
+            code='bot_apply_failed', msg=str(exc), data={'uuid': exc.bot_uuid}, request_id=request_id
+        ), 400
+
     async def initialize(self) -> None:
         @self.route(
             '',
@@ -34,7 +43,14 @@ class BotsRouterGroup(group.RouterGroup):
         )
         async def _(request_context: RequestContext) -> str:
             json_data = await quart.request.json
-            bot_uuid = await self.ap.bot_service.create_bot(request_context, json_data)
+            if not isinstance(json_data, dict):
+                return self.http_status(400, 'invalid_bot_config', 'Bot configuration must be an object')
+            try:
+                bot_uuid = await self.ap.bot_service.create_bot(request_context, json_data)
+            except BotApplyError as exc:
+                return self._apply_error_response(exc)
+            except ValueError as exc:
+                return self.http_status(400, 'invalid_bot_config', bot_error_message(exc, json_data))
             return self.success(data={'uuid': bot_uuid})
 
         @self.route(
@@ -63,7 +79,14 @@ class BotsRouterGroup(group.RouterGroup):
         async def _(bot_uuid: str, request_context: RequestContext) -> str:
             if quart.request.method == 'PUT':
                 json_data = await quart.request.json
-                await self.ap.bot_service.update_bot(request_context, bot_uuid, json_data)
+                if not isinstance(json_data, dict):
+                    return self.http_status(400, 'invalid_bot_config', 'Bot configuration must be an object')
+                try:
+                    await self.ap.bot_service.update_bot(request_context, bot_uuid, json_data)
+                except BotApplyError as exc:
+                    return self._apply_error_response(exc)
+                except ValueError as exc:
+                    return self.http_status(400, 'invalid_bot_config', bot_error_message(exc, json_data))
             else:
                 await self.ap.bot_service.delete_bot(request_context, bot_uuid)
             return self.success()

src/langbot/pkg/api/http/service/bot.py

@@ -11,6 +11,7 @@ from ....entity.persistence import agent as persistence_agent
 from ....entity.persistence import bot as persistence_bot
 from ....entity.persistence import pipeline as persistence_pipeline
 from ....workspace.errors import WorkspaceNotFoundError
+from .bot_errors import BotApplyError, bot_error_message
 from .tenant import TenantContext, require_workspace_uuid, scope_statement
 from ....utils import httpclient
 from ....platform.sources import http_bot_signing
@@ -664,7 +665,10 @@ class BotService:
 
         bot = await self.get_bot(context, bot_data['uuid'], include_secret=True)
 
-        await self.ap.platform_mgr.load_bot(context, bot)
+        try:
+            await self.ap.platform_mgr.load_bot(context, bot)
+        except Exception as exc:
+            raise BotApplyError(bot_error_message(exc, bot), bot['uuid']) from exc
 
         return bot_data['uuid']
 
@@ -694,17 +698,17 @@ class BotService:
                     runtime_bot.bot_entity.description = update_data['description']
             return
 
-        await self.ap.platform_mgr.remove_bot(context, bot_uuid)
+        # Persisted configuration is distinct from applying it to the running adapter.
-
-        # select from db
         bot = await self.get_bot(context, bot_uuid, include_secret=True)
+        try:
+            await self.ap.platform_mgr.remove_bot(context, bot_uuid)
+            runtime_bot = await self.ap.platform_mgr.load_bot(context, bot)
+            if runtime_bot.enable:
+                await runtime_bot.run()
+        except Exception as exc:
+            raise BotApplyError(bot_error_message(exc, bot), bot['uuid']) from exc
 
-        runtime_bot = await self.ap.platform_mgr.load_bot(context, bot)
+        # Reset conversations using this bot after its configuration is applied.
-
-        if runtime_bot.enable:
-            await runtime_bot.run()
-
-        # update all conversation that use this bot
         for session in self.ap.sess_mgr.session_list:
             if (
                 session.using_conversation is not None

src/langbot/pkg/api/http/service/bot_errors.py

@@ -0,0 +1,43 @@
+"""User-facing bot configuration errors without credentials or validation inputs."""
+
+import json
+
+import pydantic
+
+from .secrets import redact_secrets
+
+
+class BotApplyError(Exception):
+    """Configuration was persisted, but the runtime could not apply it."""
+
+    def __init__(self, message: str, bot_uuid: str | None = None):
+        super().__init__(message)
+        self.bot_uuid = bot_uuid
+
+
+def bot_error_message(error: Exception, configuration: dict) -> str:
+    if isinstance(error, pydantic.ValidationError):
+        text = '; '.join(
+            f'{".".join(map(str, item["loc"]))}: {item["msg"]}'
+            for item in error.errors(include_input=False, include_context=False, include_url=False)
+        )
+    else:
+        text = str(error).strip() or type(error).__name__
+
+    replacements = []
+
+    def collect(original, masked):
+        if isinstance(original, dict) and isinstance(masked, dict):
+            for key, value in original.items():
+                collect(value, masked.get(key))
+        elif isinstance(original, (list, tuple)) and isinstance(masked, (list, tuple)):
+            for value, replacement in zip(original, masked):
+                collect(value, replacement)
+        elif isinstance(original, str) and original and original != masked:
+            for representation in {original, repr(original)[1:-1], json.dumps(original, ensure_ascii=False)[1:-1]}:
+                replacements.append((representation, str(masked)))
+
+    collect(configuration, redact_secrets(configuration))
+    for original, masked in sorted(replacements, key=lambda item: len(item[0]), reverse=True):
+        text = text.replace(original, masked)
+    return text[:2000]

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

@@ -103,9 +103,9 @@ spec:
         zh_Hans: 启用钉钉卡片流式回复模式
         zh_Hant: 啟用釘釘卡片串流回覆模式
       description:
-        en_US: If enabled, the bot will use DingTalk card streaming replies.
+        en_US: Used for streaming replies from pipelines or plugins. If enabled, the bot will use DingTalk card streaming replies.
-        zh_Hans: 如果启用，将使用钉钉卡片流式方式来回复内容
+        zh_Hans: 用于流水线和插件主动发起的流式回复。如果启用，将使用钉钉卡片流式方式来回复内容
-        zh_Hant: 如果啟用，將使用釘釘卡片串流方式來回覆內容
+        zh_Hant: 用於流水線和插件主動發起的串流回覆。如果啟用，將使用釘釘卡片串流方式來回覆內容
       type: boolean
       required: true
       default: false

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

@@ -38,6 +38,10 @@ spec:
         en_US: Enable stream reply
         zh_Hans: 启用流式回复
         zh_Hant: 啟用串流回覆
+      description:
+        en_US: Used for streaming replies from pipelines or plugins.
+        zh_Hans: 用于流水线和插件主动发起的流式回复。
+        zh_Hant: 用於流水線和插件主動發起的串流回覆。
       type: boolean
       required: true
       default: false

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

@@ -174,10 +174,10 @@ spec:
         zh_Hant: 啟用飛書串流回覆模式
         ja_JP: ストリーミング返信モードを有効化
       description:
-        en_US: If enabled, replies are rendered through an updating Lark card.
+        en_US: Used for streaming replies from pipelines or plugins. If enabled, replies are rendered through an updating Lark card.
-        zh_Hans: 如果启用，将使用可更新的飞书卡片进行流式回复。
+        zh_Hans: 用于流水线和插件主动发起的流式回复。如果启用，将使用可更新的飞书卡片进行流式回复。
-        zh_Hant: 如果啟用，將使用可更新的飛書卡片進行串流回覆。
+        zh_Hant: 用於流水線和插件主動發起的串流回覆。如果啟用，將使用可更新的飛書卡片進行串流回覆。
-        ja_JP: 有効にすると、更新可能な Lark カードでストリーミング返信します。
+        ja_JP: パイプラインやプラグインのストリーミング返信に適用されます。有効にすると、更新可能な Lark カードでストリーミング返信します。
       type: boolean
       required: true
       default: false

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

@@ -100,9 +100,9 @@ spec:
         zh_Hans: 启用流式回复模式
         zh_Hant: 啟用串流回覆模式
       description:
-        en_US: If enabled, the adapter uses QQ Official streaming replies for C2C private messages.
+        en_US: Used for streaming replies from pipelines or plugins. If enabled, the adapter uses QQ Official streaming replies for C2C private messages.
-        zh_Hans: 启用后，适配器会对 C2C 私聊使用 QQ 官方流式回复。
+        zh_Hans: 用于流水线和插件主动发起的流式回复。启用后，适配器会对 C2C 私聊使用 QQ 官方流式回复。
-        zh_Hant: 啟用後，適配器會對 C2C 私聊使用 QQ 官方串流回覆。
+        zh_Hant: 用於流水線和插件主動發起的串流回覆。啟用後，適配器會對 C2C 私聊使用 QQ 官方串流回覆。
       type: boolean
       required: true
       default: false

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

@@ -39,8 +39,8 @@ spec:
         en_US: Enable Stream Reply Mode
         zh_Hans: 启用电报流式回复模式
       description:
-        en_US: If enabled, the bot will use the stream of telegram reply mode
+        en_US: Used for streaming replies from pipelines or plugins. If enabled, the bot will use the stream of telegram reply mode
-        zh_Hans: 如果启用，将使用电报流式方式来回复内容
+        zh_Hans: 用于流水线和插件主动发起的流式回复。如果启用，将使用电报流式方式来回复内容
       type: boolean
       required: true
       default: false

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

@@ -145,9 +145,9 @@ spec:
         zh_Hans: 启用流式回复
         zh_Hant: 啟用串流回覆
       description:
-        en_US: If enabled, the bot will use WeComBot streaming replies.
+        en_US: Used for streaming replies from pipelines or plugins. If enabled, the bot will use WeComBot streaming replies.
-        zh_Hans: 如果启用，机器人将使用企业微信智能机器人流式回复。
+        zh_Hans: 用于流水线和插件主动发起的流式回复。如果启用，机器人将使用企业微信智能机器人流式回复。
-        zh_Hant: 如果啟用，機器人將使用企業微信智慧機器人串流回覆。
+        zh_Hant: 用於流水線和插件主動發起的串流回覆。如果啟用，機器人將使用企業微信智慧機器人串流回覆。
       type: boolean
       required: false
       default: true

src/langbot/pkg/platform/botmgr.py

@@ -38,7 +38,6 @@ from .logger import EventLogger
 from .adapter_names import canonical_adapter_name
 
 import langbot_plugin.api.entities.builtin.provider.session as provider_session
-import langbot_plugin.api.entities.builtin.provider.message as provider_message
 import langbot_plugin.api.entities.events as plugin_events
 import langbot_plugin.api.entities.builtin.platform.events as platform_events
 import langbot_plugin.api.entities.builtin.platform.entities as platform_entities
@@ -826,7 +825,7 @@ class RuntimeBot:
             state_policy=StatePolicy(state_scopes=['conversation', 'actor', 'subject', 'runner']),
             delivery_policy=DeliveryPolicy(
                 enable_streaming=False,
-                enable_reply=True,
+                enable_reply=False,
                 enable_interactions=agent.get('kind') != 'event_processor',
             ),
             agent_id=agent.get('uuid'),
@@ -834,65 +833,6 @@ class RuntimeBot:
             processor_id=agent.get('uuid'),
         )
 
-    @staticmethod
-    def _provider_content_to_text(content: typing.Any) -> str:
-        if content is None:
-            return ''
-        if isinstance(content, str):
-            return content
-        if isinstance(content, list):
-            parts: list[str] = []
-            for item in content:
-                item_data = item.model_dump(mode='json') if hasattr(item, 'model_dump') else item
-                if isinstance(item_data, dict):
-                    if item_data.get('type') == 'text' and item_data.get('text') is not None:
-                        parts.append(str(item_data.get('text')))
-                    elif item_data.get('text') is not None:
-                        parts.append(str(item_data.get('text')))
-                elif item_data is not None:
-                    parts.append(str(item_data))
-            return ''.join(parts)
-        return str(content)
-
-    @classmethod
-    def _provider_output_to_text(cls, result: provider_message.Message | provider_message.MessageChunk) -> str:
-        if getattr(result, 'all_content', None):
-            return str(getattr(result, 'all_content'))
-        return cls._provider_content_to_text(getattr(result, 'content', None))
-
-    async def _deliver_agent_outputs(
-        self,
-        envelope: AgentEventEnvelope,
-        outputs: list[provider_message.Message | provider_message.MessageChunk],
-        adapter: abstract_platform_adapter.AbstractMessagePlatformAdapter | None = None,
-    ) -> None:
-        if not outputs or not envelope.delivery.reply_target:
-            return
-
-        reply_target = envelope.delivery.reply_target
-        target_type = reply_target.get('target_type')
-        target_id = reply_target.get('target_id')
-        if not target_type or not target_id:
-            return
-
-        final_text = ''
-        for output in outputs:
-            output_text = self._provider_output_to_text(output)
-            if isinstance(output, provider_message.Message):
-                final_text = output_text or final_text
-            elif output_text:
-                final_text = output_text
-
-        if not final_text:
-            return
-
-        delivery_adapter = adapter or self.adapter
-        await delivery_adapter.send_message(
-            str(target_type),
-            str(target_id),
-            platform_message.MessageChain([platform_message.Plain(text=final_text)]),
-        )
-
     async def _handle_platform_event(
         self,
         event: platform_events.EBAEvent,
@@ -1053,17 +993,18 @@ class RuntimeBot:
         envelope = self._eba_event_to_agent_envelope(event, adapter)
         if target_type == 'event_processor':
             envelope.data = event.model_dump(mode='json', exclude={'source_platform_object', 'legacy_event'})
-        outputs: list[provider_message.Message | provider_message.MessageChunk] = []
         try:
-            async for output in self.ap.agent_run_orchestrator.run(
+            async for _ in self.ap.agent_run_orchestrator.run(
                 envelope,
                 binding,
                 adapter_context={
                     '_delivery_adapter': adapter,
+                    '_platform_event': event,
                     '_execution_context': self.execution_context,
                 },
             ):
-                outputs.append(output)
+                # Results are journaled by the orchestrator; platform sends require explicit actions.
+                pass
         except Exception:
             return await self._record_event_route_trace(
                 event_type=event_type,
@@ -1076,21 +1017,6 @@ class RuntimeBot:
                 reason='Agent runner failed',
                 text=f'Failed to run Agent for EBA event {event_type}: {traceback.format_exc()}',
             )
-
-        try:
-            await self._deliver_agent_outputs(envelope, outputs, adapter=adapter)
-        except Exception:
-            return await self._record_event_route_trace(
-                event_type=event_type,
-                status='failed',
-                level='error',
-                binding=event_binding,
-                target_type=target_type,
-                target_uuid=target_uuid,
-                failure_code='delivery_failed',
-                reason='Agent output delivery failed',
-                text=f'Failed to deliver Agent output for EBA event {event_type}: {traceback.format_exc()}',
-            )
         return await self._record_event_route_trace(
             event_type=event_type,
             status='delivered',
@@ -1458,14 +1384,13 @@ class RuntimeBot:
             data={'interaction': submission},
         )
 
-        outputs: list[provider_message.Message | provider_message.MessageChunk] = []
+        async for _ in self.ap.agent_run_orchestrator.run(
-        async for output in self.ap.agent_run_orchestrator.run(
             envelope,
             binding,
             adapter_context={'_delivery_adapter': adapter},
         ):
-            outputs.append(output)
+            # Resuming an interaction follows the same explicit-action delivery policy.
-        await self._deliver_agent_outputs(envelope, outputs, adapter=adapter)
+            pass
 
     async def _dispatch_eba_message_to_pipeline(
         self,

src/langbot/pkg/platform/sources/dingtalk.yaml

@@ -82,9 +82,9 @@ spec:
         zh_Hans: 启用钉钉卡片流式回复模式
         zh_Hant: 啟用釘釘卡片串流回覆模式
       description:
-        en_US: If enabled, the bot will use the stream of lark reply mode
+        en_US: Used for streaming replies from pipelines or plugins. If enabled, the bot will use the stream of lark reply mode
-        zh_Hans: 如果启用，将使用钉钉卡片流式方式来回复内容
+        zh_Hans: 用于流水线和插件主动发起的流式回复。如果启用，将使用钉钉卡片流式方式来回复内容
-        zh_Hant: 如果啟用，將使用釘釘卡片串流方式來回覆內容
+        zh_Hant: 用於流水線和插件主動發起的串流回覆。如果啟用，將使用釘釘卡片串流方式來回覆內容
       type: boolean
       required: true
       default: false

src/langbot/pkg/platform/sources/lark.yaml

@@ -178,10 +178,10 @@ spec:
         zh_Hant: 啟用飛書串流回覆模式
         ja_JP: ストリーミング返信モードを有効化
       description:
-        en_US: If enabled, the bot will use the stream of lark reply mode
+        en_US: Used for streaming replies from pipelines or plugins. If enabled, the bot will use the stream of lark reply mode
-        zh_Hans: 如果启用，将使用飞书流式方式来回复内容
+        zh_Hans: 用于流水线和插件主动发起的流式回复。如果启用，将使用飞书流式方式来回复内容
-        zh_Hant: 如果啟用，將使用飛書串流方式來回覆內容
+        zh_Hant: 用於流水線和插件主動發起的串流回覆。如果啟用，將使用飛書串流方式來回覆內容
-        ja_JP: 有効にすると、ボットはストリーミングモードでメッセージに返信します
+        ja_JP: パイプラインやプラグインのストリーミング返信に適用されます。有効にすると、ボットはストリーミングモードでメッセージに返信します
       type: boolean
       required: true
       default: false

src/langbot/pkg/platform/sources/mattermost.yaml

@@ -64,8 +64,8 @@ spec:
         vi_VN: Bật phản hồi luồng
         es_ES: Activar respuesta en streaming
       description:
-        en_US: Update a Mattermost post while LangBot generates a response
+        en_US: Used for streaming replies from pipelines or plugins. Update a Mattermost post while LangBot generates a response
-        zh_Hans: 在 LangBot 生成回复时持续更新同一条 Mattermost 消息
+        zh_Hans: 用于流水线和插件主动发起的流式回复。在 LangBot 生成回复时持续更新同一条 Mattermost 消息
       type: boolean
       required: false
       default: true

src/langbot/pkg/platform/sources/qqofficial.yaml

@@ -90,9 +90,9 @@ spec:
         zh_Hans: 启用流式回复模式
         zh_Hant: 啟用串流回覆模式
       description:
-        en_US: If enabled, the bot will use streaming mode to reply messages (C2C only)
+        en_US: Used for streaming replies from pipelines or plugins. If enabled, the bot will use streaming mode to reply messages (C2C only)
-        zh_Hans: 如果启用，机器人将使用流式方式回复消息（仅私聊）
+        zh_Hans: 用于流水线和插件主动发起的流式回复。如果启用，机器人将使用流式方式回复消息（仅私聊）
-        zh_Hant: 如果啟用，機器人將使用串流方式回覆訊息（僅私聊）
+        zh_Hant: 用於流水線和插件主動發起的串流回覆。如果啟用，機器人將使用串流方式回覆訊息（僅私聊）
       type: boolean
       required: true
       default: false

src/langbot/pkg/platform/sources/telegram.yaml

@@ -63,13 +63,13 @@ spec:
         vi_VN: Bật chế độ trả lời trực tuyến
         es_ES: Habilitar modo de respuesta en streaming
       description:
-        en_US: If enabled, the bot will use the stream of telegram reply mode
+        en_US: Used for streaming replies from pipelines or plugins. If enabled, the bot will use the stream of telegram reply mode
-        zh_Hans: 如果启用，将使用电报流式方式来回复内容
+        zh_Hans: 用于流水线和插件主动发起的流式回复。如果启用，将使用电报流式方式来回复内容
-        zh_Hant: 如果啟用，將使用 Telegram 串流方式來回覆內容
+        zh_Hant: 用於流水線和插件主動發起的串流回覆。如果啟用，將使用 Telegram 串流方式來回覆內容
-        ja_JP: 有効にすると、ボットはストリーミングモードでメッセージに返信します
+        ja_JP: パイプラインやプラグインのストリーミング返信に適用されます。有効にすると、ボットはストリーミングモードでメッセージに返信します
-        th_TH: หากเปิดใช้งาน บอทจะใช้โหมดสตรีมของ Telegram ในการตอบกลับ
+        th_TH: ใช้กับการตอบกลับแบบสตรีมจากไปป์ไลน์หรือปลั๊กอิน หากเปิดใช้งาน บอทจะใช้โหมดสตรีมของ Telegram ในการตอบกลับ
-        vi_VN: Nếu bật, bot sẽ sử dụng chế độ trả lời trực tuyến của Telegram
+        vi_VN: Áp dụng cho phản hồi streaming từ pipeline hoặc plugin. Nếu bật, bot sẽ sử dụng chế độ trả lời trực tuyến của Telegram
-        es_ES: Si está habilitado, el bot usará el modo de respuesta en streaming de Telegram
+        es_ES: Se aplica a respuestas en streaming de pipelines o plugins. Si está habilitado, el bot usará el modo de respuesta en streaming de Telegram
       type: boolean
       required: true
       default: false

src/langbot/pkg/platform/sources/wecombot.yaml

@@ -142,9 +142,9 @@ spec:
         zh_Hans: 启用流式回复
         zh_Hant: 啟用串流回覆
       description:
-        en_US: If enabled, the bot will use streaming mode to reply messages
+        en_US: Used for streaming replies from pipelines or plugins. If enabled, the bot will use streaming mode to reply messages
-        zh_Hans: 如果启用，机器人将使用流式模式回复消息
+        zh_Hans: 用于流水线和插件主动发起的流式回复。如果启用，机器人将使用流式模式回复消息
-        zh_Hant: 如果啟用，機器人將使用串流模式回覆訊息
+        zh_Hant: 用於流水線和插件主動發起的串流回覆。如果啟用，機器人將使用串流模式回覆訊息
       type: boolean
       required: false
       default: true

src/langbot/pkg/plugin/handler.py

@@ -1511,6 +1511,56 @@ class RuntimeConnectionHandler(handler.Handler):
                     },
                 )
 
+        async def reply_stream(data: dict[str, Any]) -> handler.ActionResponse:
+            """Explicit reply delivery authorized by the frozen event_reply permission."""
+            from ..agent.runner.reply_stream import ReplyStreamRequest
+
+            action_context = self._require_runtime_action_context()
+            session, error = await _validate_run_authorization(
+                data.get('run_id'),
+                'tool',
+                'event_reply',
+                self.ap,
+                data.get('caller_plugin_identity'),
+                operation='call',
+            )
+            if error:
+                return error
+            source_ref, error = _validate_frozen_tool_source_identity(session, 'event_reply', self.ap)
+            if error:
+                return error
+            if source_ref is None or source_ref['source'] != 'platform':
+                return handler.ActionResponse.error('event_reply must be a Host platform tool')
+            query = session.get('execution_query')
+            context = self._execution_context(action_context)
+            if query is None or any(
+                getattr(query, field, None) != getattr(context, field, None)
+                for field in ('instance_uuid', 'workspace_uuid', 'placement_generation')
+            ):
+                return handler.ActionResponse.error('Reply stream run belongs to another execution scope')
+            streams = session.get('reply_streams')
+            if streams is None:
+                return handler.ActionResponse.error('Streaming replies are unavailable for this run')
+            try:
+                if not streams.mock:
+                    bot_id = session['authorization'].get('bot_id')
+                    bot = await self.ap.platform_mgr.get_bot_by_uuid(context, bot_id)
+                    if bot is None or bot.adapter is not streams.adapter:
+                        return handler.ActionResponse.error('The reply adapter is no longer active')
+                    if 'send_message' not in bot.adapter.get_supported_apis():
+                        return handler.ActionResponse.error('The reply adapter no longer supports sending messages')
+                request = ReplyStreamRequest.model_validate(
+                    {key: data[key] for key in ('stream_id', 'operation', 'text') if key in data}
+                )
+                result = await streams.apply(request)
+                return handler.ActionResponse.success(data={'result': result})
+            except Exception as exc:
+                return handler.ActionResponse.error(f'Streaming reply failed: {exc}')
+
+        reply_stream_action = getattr(PluginToRuntimeAction, 'REPLY_STREAM', None)
+        if reply_stream_action is not None:
+            self.action(reply_stream_action)(reply_stream)
+
         @self.action(PluginToRuntimeAction.CALL_TOOL)
         async def call_tool(data: dict[str, Any]) -> handler.ActionResponse:
             """Call a tool

tests/unit_tests/agent/test_reply_stream.py

@@ -0,0 +1,270 @@
+"""Explicit streaming delivery across SDK, Host lifecycle, and adapter boundaries."""
+
+import asyncio
+from types import SimpleNamespace
+from unittest.mock import AsyncMock
+from uuid import uuid4
+
+import pytest
+from langbot_plugin.api.entities.builtin.platform import events, entities, message
+from langbot_plugin.api.entities.builtin.agent_runner.context_access import ContextAPICapabilities
+from langbot_plugin.api.proxies.agent_run import AgentRunAPIProxy
+from langbot_plugin.api.proxies.agent_run.common import PermissionDeniedError
+from langbot_plugin.entities.io.actions.enums import PluginToRuntimeAction
+
+from langbot.pkg.agent.runner.reply_stream import ReplyStreamRequest, ReplyStreamSession
+
+
+def make_session(*, native=True, source=True, mock=False):
+    event = SimpleNamespace(
+        delivery=SimpleNamespace(
+            surface='webui' if mock else 'platform',
+            platform_capabilities={'debug_mock': mock},
+            reply_target={'target_type': 'person', 'target_id': 'user-1'},
+        )
+    )
+    incoming = (
+        events.MessageReceivedEvent(
+            message_id='source-1',
+            sender=entities.User(id='user-1'),
+            chat_id='user-1',
+            chat_type=entities.ChatType.PRIVATE,
+            message_chain=message.MessageChain([message.Plain(text='hello')]),
+            source_platform_object=object(),
+        )
+        if source
+        else None
+    )
+    adapter = SimpleNamespace(
+        is_stream_output_supported=AsyncMock(return_value=native),
+        create_message_card=AsyncMock(return_value=True),
+        reply_message_chunk=AsyncMock(),
+        send_message=AsyncMock(),
+        reply_message=AsyncMock(),
+    )
+    return ReplyStreamSession(event, adapter, incoming), adapter, incoming
+
+
+def request(key, operation='update', text='hello'):
+    return ReplyStreamRequest(stream_id=key, operation=operation, text=text)
+
+
+def proxy_for(session, *, allowed=True, advertised=True):
+    if not hasattr(AgentRunAPIProxy, 'reply_stream'):
+        pytest.skip('SDK does not provide the optional streaming reply API')
+    context = SimpleNamespace(
+        run_id='run-1',
+        runtime=SimpleNamespace(deadline_at=None),
+        context=SimpleNamespace(available_apis=ContextAPICapabilities(reply_stream=advertised)),
+        resources=SimpleNamespace(
+            models=[],
+            knowledge_bases=[],
+            tools=[SimpleNamespace(tool_name='event_reply', operations=['call'])] if allowed else [],
+        ),
+    )
+
+    async def action(action, data, timeout):
+        assert action == PluginToRuntimeAction.REPLY_STREAM
+        assert data['run_id'] == 'run-1'
+        return {
+            'result': await session.apply(
+                ReplyStreamRequest.model_validate({k: v for k, v in data.items() if k != 'run_id'})
+            )
+        }
+
+    transport = SimpleNamespace(call_action=AsyncMock(side_effect=action))
+    return AgentRunAPIProxy(context, transport), transport
+
+
+@pytest.mark.parametrize(
+    'native,source,mock',
+    [
+        (True, True, False),
+        (False, True, False),
+        (True, False, False),
+        (True, True, True),
+    ],
+)
+async def test_sdk_stream_reuses_adapter_or_sends_one_final_message(native, source, mock):
+    session, adapter, incoming = make_session(native=native, source=source, mock=mock)
+    api, transport = proxy_for(session)
+    async with api.reply_stream() as stream:
+        await stream.update('hello')
+        await stream.update('hello world')
+        adapter.send_message.assert_not_awaited()
+        adapter.reply_message.assert_not_awaited()
+    assert stream.result['status'] == 'completed'
+    assert stream.result['text'] == 'hello world'
+    assert transport.call_action.await_count == 3
+    if mock:
+        adapter.create_message_card.assert_not_awaited()
+        adapter.reply_message_chunk.assert_not_awaited()
+        adapter.send_message.assert_not_awaited()
+        assert stream.result['mock'] is True
+    elif native and source:
+        adapter.create_message_card.assert_awaited_once()
+        delivered_source = adapter.create_message_card.await_args.args[1]
+        assert delivered_source.source_platform_object is incoming.source_platform_object
+        chunks = adapter.reply_message_chunk.await_args_list
+        assert [c.kwargs['bot_message'].all_content for c in chunks] == ['hello', 'hello world', 'hello world']
+        assert [c.kwargs['is_final'] for c in chunks] == [False, False, True]
+        assert chunks[-1].kwargs['bot_message'].tool_calls is None
+    else:
+        adapter.reply_message_chunk.assert_not_awaited()
+        if source:
+            adapter.reply_message.assert_awaited_once()
+            assert adapter.reply_message.await_args.kwargs['message'][0].text == 'hello world'
+        else:
+            adapter.send_message.assert_awaited_once()
+            assert adapter.send_message.await_args.args[2][0].text == 'hello world'
+    await session.close()
+
+
+@pytest.mark.parametrize('native', [True, False])
+@pytest.mark.parametrize('error', [RuntimeError, asyncio.CancelledError])
+async def test_exception_finalizes_visible_card_without_sending_buffered_partial(native, error):
+    session, adapter, _ = make_session(native=native)
+    api, _ = proxy_for(session)
+    with pytest.raises(error):
+        async with api.reply_stream() as stream:
+            await stream.update('partial')
+            raise error()
+    if native:
+        assert adapter.reply_message_chunk.await_args.kwargs['is_final'] is True
+    adapter.reply_message.assert_not_awaited()
+    adapter.send_message.assert_not_awaited()
+    count = adapter.reply_message_chunk.await_count
+    await session.close()
+    assert adapter.reply_message_chunk.await_count == count
+
+
+@pytest.mark.parametrize('allowed,advertised', [(False, True), (True, False)])
+async def test_missing_permission_or_old_host_fails_before_delivery(allowed, advertised):
+    session, _, _ = make_session()
+    api, transport = proxy_for(session, allowed=allowed, advertised=advertised)
+    with pytest.raises(PermissionDeniedError):
+        async with api.reply_stream():
+            pytest.fail('Not authorized')
+    transport.call_action.assert_not_awaited()
+
+
+async def test_empty_stream_and_duplicate_finish_do_not_send_twice():
+    session, adapter, _ = make_session(native=False)
+    empty = uuid4()
+    await session.apply(request(empty, 'finish', ''))
+    adapter.reply_message.assert_not_awaited()
+    key = uuid4()
+    await session.apply(request(key))
+    first = await session.apply(request(key, 'finish'))
+    assert await session.apply(request(key, 'finish')) == first
+    adapter.reply_message.assert_awaited_once()
+    with pytest.raises(ValueError, match='closed'):
+        await session.apply(request(key))
+    await session.close()
+    with pytest.raises(ValueError, match='ended'):
+        await session.apply(request(uuid4()))
+
+
+async def test_host_cleanup_closes_stream_when_plugin_disappears():
+    session, adapter, _ = make_session()
+    await session.apply(request(uuid4()))
+    await session.close()
+    assert adapter.reply_message_chunk.await_args.kwargs['is_final'] is True
+
+
+async def test_uncertain_final_send_is_not_retried_by_finish_or_cleanup():
+    session, adapter, _ = make_session(native=False)
+    adapter.reply_message.side_effect = TimeoutError('Response lost')
+    key = uuid4()
+    with pytest.raises(TimeoutError):
+        await session.apply(request(key, 'finish'))
+    with pytest.raises(ValueError, match='closed'):
+        await session.apply(request(key, 'finish'))
+    await session.close()
+    adapter.reply_message.assert_awaited_once()
+
+
+async def test_failed_update_closes_existing_card_during_run_cleanup():
+    session, adapter, _ = make_session()
+    adapter.reply_message_chunk.side_effect = [RuntimeError('update failed'), None]
+    with pytest.raises(RuntimeError):
+        await session.apply(request(uuid4()))
+    await session.close()
+    assert adapter.reply_message_chunk.await_args.kwargs['is_final'] is True
+
+
+async def test_streams_are_isolated_by_run_and_bounded():
+    first, a, _ = make_session(native=False)
+    second, b, _ = make_session(native=False)
+    key = uuid4()
+    await first.apply(request(key, 'update', 'first'))
+    await second.apply(request(key, 'finish', 'second'))
+    assert b.reply_message.await_args.kwargs['message'][0].text == 'second'
+    a.reply_message.assert_not_awaited()
+    for _ in range(15):
+        await first.apply(request(uuid4(), 'finish', ''))
+    with pytest.raises(ValueError, match='at most'):
+        await first.apply(request(uuid4()))
+    await first.close()
+    a.reply_message.assert_not_awaited()
+
+
+async def test_event_processor_uses_shared_sdk_api_and_emits_one_trace_for_the_stream():
+    from unittest.mock import Mock
+    from langbot_plugin.api.definition.components.event_processor import EventProcessor
+
+    session, adapter, incoming = make_session()
+    api, _ = proxy_for(session)
+    processor = EventProcessor()
+    processor.get_run_api = Mock(return_value=api)
+
+    @processor.handler(events.MessageReceivedEvent)
+    async def handle(ctx):
+        async with ctx.reply_stream() as stream:
+            await stream.update('one')
+            await stream.update('one two')
+
+    context = SimpleNamespace(
+        run_id='run-1',
+        config={},
+        event=SimpleNamespace(
+            data=incoming.model_dump(mode='json', exclude={'source_platform_object', 'legacy_event'})
+        ),
+    )
+    results = [result async for result in processor.run(context)]
+    assert [r.type for r in results] == ['tool.call.started', 'tool.call.completed', 'run.completed']
+    assert results[1].data['result']['text'] == 'one two'
+    assert adapter.reply_message_chunk.await_count == 3
+    await session.close()
+
+
+async def test_shared_adapter_uses_host_ids_to_isolate_identical_plugin_stream_ids():
+    first, adapter, _ = make_session()
+    second, _, _ = make_session()
+    second.adapter = adapter
+    key = uuid4()
+    await first.apply(request(key, text='first'))
+    await second.apply(request(key, text='second'))
+    ids = [c.args[0] for c in adapter.create_message_card.await_args_list]
+    assert len(set(ids)) == 2
+    assert str(key) not in ids
+    await first.close()
+    await second.close()
+
+
+async def test_run_cleanup_cancels_inflight_update_and_finalizes_card():
+    session, adapter, _ = make_session()
+    started = asyncio.Event()
+
+    async def update(**kwargs):
+        if not kwargs['is_final']:
+            started.set()
+            await asyncio.Event().wait()
+
+    adapter.reply_message_chunk.side_effect = update
+    task = asyncio.create_task(session.apply(request(uuid4())))
+    await asyncio.wait_for(started.wait(), 1)
+    await asyncio.wait_for(session.close(), 1)
+    with pytest.raises(asyncio.CancelledError):
+        await task
+    assert adapter.reply_message_chunk.await_args.kwargs['is_final'] is True

tests/unit_tests/api/service/test_agent_service.py

@@ -151,7 +151,18 @@ class TestAgentServiceMetadata:
 
 class TestAgentServiceDebug:
     @pytest.mark.parametrize('streaming', [False, True])
-    async def test_debug_agent_runs_configured_runner_with_synthetic_event(self, streaming):
+    @pytest.mark.parametrize(
+        'result_type',
+        [
+            'tool.call.started',
+            'tool.call.completed',
+            'message.delta',
+            'message.completed',
+            'processor.log',
+            'run.completed',
+        ],
+    )
+    async def test_debug_agent_runs_configured_runner_with_synthetic_event(self, streaming, result_type):
         app = _make_app()
         agent_config = _agent_row().config
         agent_config['allowed_platform_tools'] = ['platform_get_user_info']
@@ -162,7 +173,7 @@ class TestAgentServiceDebug:
         agent_config['allowed_tools'] = ['exec', 'weather']
 
         visible_event = {
-            'type': 'tool.call.started',
+            'type': result_type,
             'data': {'tool_name': 'exec', 'parameters': {'command': 'echo hi'}},
         }
         observer = AsyncMock() if streaming else None
@@ -171,6 +182,9 @@ class TestAgentServiceDebug:
             assert binding.delivery_policy.enable_streaming is streaming
             await adapter_context['_result_observer']({**visible_event, 'private_context': 'must not leak'})
             await adapter_context['_result_observer']({'type': 'state.updated', 'data': {'private': True}})
+            if streaming:
+                # Debug events reach the client before the runner returns its final output.
+                observer.assert_awaited_once_with(visible_event)
             yield SimpleNamespace(
                 role='assistant',
                 content='debug result',

tests/unit_tests/api/test_bots_controller_errors.py

@@ -0,0 +1,138 @@
+from __future__ import annotations
+
+import sys
+import types
+from importlib import import_module
+from types import SimpleNamespace
+from unittest.mock import AsyncMock
+
+import pytest
+import quart
+
+
+core_app_module = types.ModuleType('langbot.pkg.core.app')
+core_app_module.Application = object
+sys.modules.setdefault('langbot.pkg.core.app', core_app_module)
+
+
+pytestmark = pytest.mark.asyncio
+
+
+async def _create_test_client(bot_service: SimpleNamespace):
+    app = quart.Quart(__name__)
+    account = SimpleNamespace(
+        uuid='account-test',
+        user='test@example.com',
+    )
+    user_service = SimpleNamespace(
+        get_authenticated_account=AsyncMock(return_value=account),
+    )
+    access = SimpleNamespace(
+        workspace=SimpleNamespace(uuid='workspace-test'),
+        membership=SimpleNamespace(
+            uuid='membership-test',
+            role='developer',
+            projection_revision=1,
+        ),
+        execution=SimpleNamespace(
+            instance_uuid='instance-test',
+            placement_generation=1,
+        ),
+    )
+    ap = SimpleNamespace(
+        bot_service=bot_service,
+        user_service=user_service,
+        apikey_service=SimpleNamespace(authenticate_api_key=AsyncMock(return_value=None)),
+        workspace_collaboration_service=SimpleNamespace(resolve_account_workspace=AsyncMock(return_value=access)),
+    )
+    BotsRouterGroup = import_module('langbot.pkg.api.http.controller.groups.platform.bots').BotsRouterGroup
+    group = BotsRouterGroup(ap, app)
+    await group.initialize()
+    return app.test_client()
+
+
+@pytest.mark.parametrize('method,path', [('post', '/api/v1/platform/bots'), ('put', '/api/v1/platform/bots/bot-1')])
+async def test_bot_config_error_preserves_details(method, path):
+    error = ValueError('Lark missing required config: app_id, app_secret, bot_name')
+    service = SimpleNamespace(create_bot=AsyncMock(side_effect=error), update_bot=AsyncMock(side_effect=error))
+    client = await _create_test_client(service)
+    response = await getattr(client, method)(
+        path, json={'adapter_config': {}}, headers={'Authorization': 'Bearer token'}
+    )
+    assert response.status_code == 400
+    body = await response.get_json()
+    assert body['code'] == 'invalid_bot_config'
+    assert body['msg'] == str(error)
+
+
+async def test_bot_apply_failure_identifies_persisted_record():
+    from langbot.pkg.api.http.service.bot_errors import BotApplyError
+
+    service = SimpleNamespace(create_bot=AsyncMock(side_effect=BotApplyError('Missing app_id', 'saved-bot')))
+    client = await _create_test_client(service)
+    response = await client.post('/api/v1/platform/bots', json={}, headers={'Authorization': 'Bearer token'})
+    assert response.status_code == 400
+    body = await response.get_json()
+    assert body.pop('request_id')
+    assert body == {
+        'code': 'bot_apply_failed',
+        'msg': 'Missing app_id',
+        'data': {'uuid': 'saved-bot'},
+    }
+
+
+async def test_unexpected_failure_keeps_request_reference_without_exception_details():
+    client = await _create_test_client(
+        SimpleNamespace(update_bot=AsyncMock(side_effect=RuntimeError('database password')))
+    )
+    response = await client.put('/api/v1/platform/bots/bot-1', json={}, headers={'Authorization': 'Bearer token'})
+    body = await response.get_json()
+    assert response.status_code == 500
+    assert body['request_id']
+    assert 'database password' not in str(body)
+
+
+async def test_invalid_request_body_is_actionable():
+    service = SimpleNamespace(update_bot=AsyncMock())
+    client = await _create_test_client(service)
+    response = await client.put('/api/v1/platform/bots/bot-1', json=[], headers={'Authorization': 'Bearer token'})
+    assert response.status_code == 400
+    service.update_bot.assert_not_awaited()
+
+
+async def test_runtime_error_redacts_persisted_secrets_on_partial_update():
+    from langbot.pkg.api.http.service.bot import BotService
+    from langbot.pkg.api.http.service.bot_errors import BotApplyError
+
+    ap = SimpleNamespace(
+        persistence_mgr=SimpleNamespace(execute_async=AsyncMock(return_value=SimpleNamespace(rowcount=1))),
+        platform_mgr=SimpleNamespace(
+            remove_bot=AsyncMock(),
+            load_bot=AsyncMock(side_effect=ValueError('Invalid app_secret: persisted-secret-value')),
+        ),
+    )
+    service = BotService(ap)
+    service.get_bot = AsyncMock(
+        return_value={'uuid': 'bot-1', 'adapter_config': {'app_secret': 'persisted-secret-value'}}
+    )
+    with pytest.raises(BotApplyError) as captured:
+        await service.update_bot('workspace-test', 'bot-1', {'enable': True})
+    assert str(captured.value) == 'Invalid app_secret: ***'
+    assert captured.value.bot_uuid == 'bot-1'
+    ap.persistence_mgr.execute_async.assert_awaited_once()
+
+
+async def test_validation_error_does_not_include_input_values():
+    import pydantic
+    from langbot.pkg.api.http.service.bot_errors import bot_error_message
+
+    class Config(pydantic.BaseModel):
+        app_secret: int
+
+    try:
+        Config(app_secret='private-value')
+    except pydantic.ValidationError as error:
+        result = bot_error_message(error, {'app_secret': 'private-value'})
+    assert 'app_secret' in result
+    assert 'private-value' not in result
+    assert 'input_value' not in result

tests/unit_tests/platform/test_routing_rules.py

@@ -447,7 +447,7 @@ class TestEBAEventBindings:
             'event_get_actor',
         ]
         assert binding.delivery_policy.enable_streaming is False
-        assert binding.delivery_policy.enable_reply is True
+        assert binding.delivery_policy.enable_reply is False
         assert binding.delivery_policy.enable_interactions is True
         assert binding.state_policy.state_scopes == ['conversation', 'actor', 'subject', 'runner']
         assert binding.agent_id == 'agent-1'
@@ -651,7 +651,9 @@ class TestInteractionResumeRouting:
         assert binding.processor_type == 'agent'
         assert binding.processor_id == 'agent-1'
         assert adapter_context == {'_delivery_adapter': adapter}
-        adapter.send_message.assert_awaited_once()
+        assert envelope.delivery.supports_streaming is False
+        assert binding.delivery_policy.enable_reply is False
+        adapter.send_message.assert_not_awaited()
 
     def test_agent_product_to_binding_does_not_fallback_to_component_ref(self):
         """An empty config runner stays unconfigured even if component_ref is stale."""
@@ -762,3 +764,89 @@ async def test_bound_event_processor_receives_one_complete_typed_event():
     assert envelope.data['type'] == 'group.member_joined'
     assert 'source_platform_object' not in envelope.data
     bot.ap.plugin_connector.emit_event.assert_not_called()
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize('kind', ['agent', 'event_processor'])
+@pytest.mark.parametrize('output_kind', ['message', 'chunks', 'tool_rounds'])
+@pytest.mark.parametrize('explicit_reply', [False, True])
+@pytest.mark.parametrize('runner_fails', [False, True])
+async def test_processor_outputs_require_explicit_platform_actions(kind, output_kind, explicit_reply, runner_fails):
+    """Draining runner results must not send text, duplicate replies, or stream cards."""
+    from langbot_plugin.api.entities.builtin.platform import entities, events, message
+    from langbot_plugin.api.entities.builtin.provider import message as provider_message
+
+    from langbot.pkg.agent.runner.platform_tools import execute_platform_tool, freeze_platform_context
+
+    bot = TestEventRouteTrace._make_bot(
+        [{'id': 'binding-1', 'event_pattern': 'message.received', 'target_type': kind, 'target_uuid': 'agent-1'}]
+    )
+    adapter = SimpleNamespace(
+        get_supported_apis=lambda: ['send_message'],
+        is_stream_output_supported=AsyncMock(return_value=True),
+        send_message=AsyncMock(return_value='message-2'),
+        create_message_card=AsyncMock(),
+        reply_message_chunk=AsyncMock(),
+    )
+    completed = []
+
+    async def run(envelope, binding, adapter_context):
+        assert envelope.delivery.supports_streaming is False
+        assert binding.delivery_policy.enable_streaming is False
+        assert binding.delivery_policy.enable_reply is False
+        assert adapter_context['_delivery_adapter'] is adapter
+        if explicit_reply:
+            session = {'authorization': {'bot_id': 'bot-1', 'platform_context': freeze_platform_context(envelope)}}
+            await execute_platform_tool(bot.ap, TEST_CONTEXT, session, 'event_reply', {'text': 'Working on it'})
+            # Progress arrives while the runner is still working.
+            adapter.send_message.assert_awaited_once()
+            assert completed == []
+        if output_kind == 'chunks':
+            yield provider_message.MessageChunk(role='assistant', content='Done', all_content='Done')
+            yield provider_message.MessageChunk(role='assistant', content='.', all_content='Done.', is_final=True)
+        elif output_kind == 'tool_rounds':
+            yield provider_message.Message(role='assistant', content='Checking the request')
+            yield provider_message.Message(role='assistant', content='Done.')
+        else:
+            yield provider_message.Message(role='assistant', content='Done.')
+        if runner_fails:
+            raise RuntimeError('Runner failed after producing text')
+        completed.append(True)
+
+    bot.ap = SimpleNamespace(
+        workspace_service=active_workspace_service(),
+        platform_mgr=SimpleNamespace(get_bot_by_uuid=AsyncMock(return_value=SimpleNamespace(adapter=adapter))),
+        agent_service=SimpleNamespace(
+            get_agent=AsyncMock(
+                return_value={
+                    'uuid': 'agent-1',
+                    'kind': kind,
+                    'enabled': True,
+                    'supported_event_patterns': ['message.received'],
+                    'config': {'runner': {'id': 'runner-1'}, 'runner_config': {'runner-1': {}}},
+                }
+            )
+        ),
+        agent_run_orchestrator=SimpleNamespace(run=run),
+    )
+    event = events.MessageReceivedEvent(
+        message_id='message-1',
+        message_chain=message.MessageChain([message.Plain(text='hello')]),
+        sender=entities.User(id='user-1', nickname='QA'),
+        chat_type=entities.ChatType.PRIVATE,
+        chat_id='user-1',
+    )
+    trace = await bot._dispatch_eba_event_to_processor(event, adapter)
+
+    assert trace['status'] == ('failed' if runner_fails else 'delivered')
+    if runner_fails:
+        assert trace['failure_code'] == 'runner_failed'
+    assert completed == ([] if runner_fails else [True])
+    assert adapter.send_message.await_count == int(explicit_reply)
+    if explicit_reply:
+        kwargs = adapter.send_message.await_args.kwargs
+        assert kwargs['target_type'] == 'person'
+        assert kwargs['target_id'] == 'user-1'
+        assert kwargs['message'][0].text == 'Working on it'
+    adapter.create_message_card.assert_not_awaited()
+    adapter.reply_message_chunk.assert_not_awaited()

tests/unit_tests/plugin/test_handler_actions.py

@@ -1575,9 +1575,7 @@ class TestAgentRunProxyActions:
         query = build_execution_query(event, [])
         app.box_service = SimpleNamespace(
             available=True,
-            get_backend_status=AsyncMock(
+            get_backend_status=AsyncMock(return_value={'backend': {'available': True}}),
-                return_value={'backend': {'available': True}}
-            ),
             execute_tool=AsyncMock(
                 return_value={
                     'ok': True,
@@ -1683,3 +1681,88 @@ class TestAgentRunProxyActions:
         provider.invoke_rerank.assert_awaited_once()
         kwargs = provider.invoke_rerank.await_args.kwargs
         assert kwargs['extra_args'] == {'top_n': 2, 'return_documents': False}
+
+
+@pytest.mark.parametrize(
+    'case',
+    [
+        'valid',
+        'expired',
+        'plugin',
+        'workspace',
+        'permission',
+        'shadowed',
+        'native',
+        'bot_removed',
+        'adapter_replaced',
+        'api_revoked',
+    ],
+)
+@pytest.mark.skipif(not hasattr(PluginToRuntimeAction, 'REPLY_STREAM'), reason='SDK does not support streaming replies')
+async def test_reply_stream_authorization_is_run_and_workspace_scoped(case):
+    from uuid import uuid4
+    from langbot.pkg.agent.runner.session_registry import get_session_registry
+
+    app = SimpleNamespace(logger=Mock(), _test_plugin_identity='test/runner')
+    runtime_handler = make_handler(app)
+    query = SimpleNamespace(
+        **{
+            field: getattr(TEST_EXECUTION_CONTEXT, field)
+            for field in ('instance_uuid', 'workspace_uuid', 'placement_generation')
+        }
+    )
+    if case == 'workspace':
+        query.workspace_uuid = 'another-workspace'
+    streams = SimpleNamespace(mock=True, apply=AsyncMock(return_value={'status': 'completed'}))
+    if case in {'native', 'bot_removed', 'adapter_replaced', 'api_revoked'}:
+        streams.mock = False
+        streams.adapter = SimpleNamespace(get_supported_apis=lambda: [] if case == 'api_revoked' else ['send_message'])
+        bot = (
+            None
+            if case == 'bot_removed'
+            else SimpleNamespace(adapter=object() if case == 'adapter_replaced' else streams.adapter)
+        )
+        app.platform_mgr = SimpleNamespace(get_bot_by_uuid=AsyncMock(return_value=bot))
+    registry = get_session_registry()
+    run_id = str(uuid4())
+    resources = make_agent_resources(
+        tools=[]
+        if case == 'permission'
+        else [
+            {
+                'tool_name': 'event_reply',
+                'operations': ['call'],
+                'source': 'mcp' if case == 'shadowed' else 'platform',
+                'source_id': 'event_reply',
+            }
+        ]
+    )
+    await registry.register(
+        run_id=run_id,
+        runner_id='plugin:test/runner/default',
+        query_id=None,
+        plugin_identity='other/plugin' if case == 'plugin' else 'test/runner',
+        resources=resources,
+        execution_query=query,
+        reply_streams=streams,
+    )
+    try:
+        if case == 'expired':
+            await registry.unregister(run_id)
+        response = await runtime_handler.actions[PluginToRuntimeAction.REPLY_STREAM.value](
+            {
+                'run_id': run_id,
+                'caller_plugin_identity': 'test/runner',
+                'stream_id': str(uuid4()),
+                'operation': 'finish',
+                'text': 'hello',
+            }
+        )
+        if case in {'valid', 'native'}:
+            assert response.code == 0
+            streams.apply.assert_awaited_once()
+        else:
+            assert response.code != 0
+            streams.apply.assert_not_awaited()
+    finally:
+        await registry.unregister(run_id)

uv.lock

@@ -2119,7 +2119,7 @@ requires-dist = [
     { name = "ebooklib", specifier = ">=0.18" },
     { name = "gewechat-client", specifier = ">=0.1.5" },
     { name = "html2text", specifier = ">=2024.2.26" },
-    { name = "langbot-plugin", git = "https://github.com/langbot-app/langbot-plugin-sdk?rev=f82b3ce935f9a33afee389fb39fe8dc29a45b615" },
+    { name = "langbot-plugin", git = "https://github.com/langbot-app/langbot-plugin-sdk?rev=ca0671b81db5b2ed937d01158a3982de3fd8500f" },
     { name = "langchain", specifier = ">=1.3.9" },
     { name = "langchain-core", specifier = ">=1.3.3" },
     { name = "langchain-text-splitters", specifier = ">=1.1.2" },
@@ -2186,7 +2186,7 @@ dev = [
 [[package]]
 name = "langbot-plugin"
 version = "0.5.5"
-source = { git = "https://github.com/langbot-app/langbot-plugin-sdk?rev=f82b3ce935f9a33afee389fb39fe8dc29a45b615#f82b3ce935f9a33afee389fb39fe8dc29a45b615" }
+source = { git = "https://github.com/langbot-app/langbot-plugin-sdk?rev=ca0671b81db5b2ed937d01158a3982de3fd8500f#ca0671b81db5b2ed937d01158a3982de3fd8500f" }
 dependencies = [
     { name = "aiofiles" },
     { name = "aiohttp" },

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

@@ -24,6 +24,7 @@ import { useTranslation } from 'react-i18next';
 import { Settings, FileText, Users, RefreshCw, Trash2 } from 'lucide-react';
 import { cn } from '@/lib/utils';
 import { toast } from 'sonner';
+import { showBotError } from './bot-error';
 import { useCurrentWorkspace } from '@/app/infra/http';
 import { Bot } from '@/app/infra/entities/api';
 import EntityBasicInfoDialog, {
@@ -91,9 +92,9 @@ export default function BotDetailContent({ id }: { id: string }) {
           current ? { ...current, enable: checked } : current,
         );
         refreshBots();
-      } catch {
+      } catch (error) {
         setBotEnabled(prev);
-        toast.error(t('bots.setBotEnableError'));
+        showBotError(error, t('bots.setBotEnableError'), t);
       }
     },
     [id, botEnabled, refreshBots, t],
@@ -129,11 +130,7 @@ export default function BotDetailContent({ id }: { id: string }) {
       await refreshBots();
       toast.success(t('bots.saveSuccess'));
     } catch (error) {
-      const message =
+      showBotError(error, t('bots.saveError'), t);
-        typeof error === 'object' && error && 'msg' in error
-          ? String((error as { msg?: string }).msg || '')
-          : '';
-      toast.error(t('bots.saveError') + message);
       throw error;
     }
   }

web/src/app/home/bots/bot-error.ts

@@ -0,0 +1,31 @@
+import type { TFunction } from 'i18next';
+import { toast } from 'sonner';
+
+export function showBotError(error: unknown, title: string, t: TFunction) {
+  const detail =
+    error && typeof error === 'object'
+      ? (error as {
+          code?: string;
+          msg?: string;
+          message?: string;
+          request_id?: string;
+        })
+      : {};
+  const message = detail.msg || detail.message || '';
+  const description =
+    detail.code === 'internal_error'
+      ? [
+          t('bots.internalErrorHint'),
+          detail.request_id &&
+            t('bots.errorReference', { id: detail.request_id }),
+        ]
+          .filter(Boolean)
+          .join('\n')
+      : message;
+  toast.error(
+    detail.code === 'bot_apply_failed'
+      ? t('bots.applyFailed')
+      : title.replace(/[:：]\s*$/, ''),
+    { description, duration: 10000 },
+  );
+}

web/src/app/home/bots/components/bot-form/BotForm.tsx

@@ -1,3 +1,4 @@
+import { showBotError } from '../../bot-error';
 import React, {
   forwardRef,
   useEffect,
@@ -413,7 +414,7 @@ const BotForm = forwardRef<BotFormHandle, BotFormProps>(function BotForm(
           toast.success(t('bots.saveSuccess'));
         })
         .catch((err) => {
-          toast.error(t('bots.saveError') + err.msg);
+          showBotError(err, t('bots.saveError'), t);
         })
         .finally(() => {
           setIsLoading(false);
@@ -438,7 +439,10 @@ const BotForm = forwardRef<BotFormHandle, BotFormProps>(function BotForm(
           onNewBotCreated(res.uuid);
         })
         .catch((err) => {
-          toast.error(t('bots.createError') + err.msg);
+          showBotError(err, t('bots.createError'), t);
+          if (err.code === 'bot_apply_failed' && err.data?.uuid) {
+            onNewBotCreated(err.data.uuid);
+          }
         })
         .finally(() => {
           setIsLoading(false);

web/src/app/infra/http/BaseHttpClient.ts

@@ -139,6 +139,9 @@ export abstract class BaseHttpClient {
             code: data?.code || status,
             msg: errMsg,
             data: data?.data || null,
+            request_id:
+              (data as { request_id?: string })?.request_id ||
+              error.response.headers['x-request-id'],
           });
         }
 

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

@@ -340,6 +340,11 @@ const enUS = {
     },
   },
   bots: {
+    applyFailed: 'Configuration saved, but could not be applied',
+    internalErrorHint:
+      'An unexpected error occurred. Check the backend logs using the reference below.',
+    errorReference: 'Error reference: {{id}}',
+
     title: 'Bots',
     description:
       'Create and manage bots, which are the entry points for LangBot to connect with various platforms',

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

@@ -348,6 +348,11 @@ const esES = {
       "Uses the Workspace owner's LangBot Account billing and credits.",
   },
   bots: {
+    applyFailed: 'Configuración guardada, pero no se pudo aplicar',
+    internalErrorHint:
+      'Se produjo un error interno. Consulta los registros del servidor con esta referencia.',
+    errorReference: 'Referencia del error: {{id}}',
+
     adapterEventDebugAction: 'Probar escucha',
     title: 'Bots',
     description:

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

@@ -346,6 +346,11 @@ const jaJP = {
     },
   },
   bots: {
+    applyFailed: '設定を保存しましたが、適用に失敗しました',
+    internalErrorHint:
+      '内部エラーが発生しました。エラー番号でバックエンドのログを確認してください。',
+    errorReference: 'エラー番号: {{id}}',
+
     title: 'ボット',
     description:
       'ボットの作成と管理を行います。LangBotと各プラットフォームを接続するためのエントリーポイントです',

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

@@ -346,6 +346,11 @@ const ruRU = {
       "Uses the Workspace owner's LangBot Account billing and credits.",
   },
   bots: {
+    applyFailed: 'Настройки сохранены, но не применены',
+    internalErrorHint:
+      'Внутренняя ошибка. Проверьте журналы сервера по указанному идентификатору.',
+    errorReference: 'Идентификатор ошибки: {{id}}',
+
     adapterEventDebugAction: 'Тест прослушивания',
     title: 'Боты',
     description:

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

@@ -333,6 +333,11 @@ const thTH = {
       "Uses the Workspace owner's LangBot Account billing and credits.",
   },
   bots: {
+    applyFailed: 'บันทึกการตั้งค่าแล้ว แต่ไม่สามารถนำไปใช้ได้',
+    internalErrorHint:
+      'เกิดข้อผิดพลาดภายใน โปรดตรวจสอบบันทึกของเซิร์ฟเวอร์ด้วยหมายเลขอ้างอิง',
+    errorReference: 'หมายเลขข้อผิดพลาด: {{id}}',
+
     adapterEventDebugAction: 'ทดสอบการรับเหตุการณ์',
     title: 'บอท',
     description:

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

@@ -342,6 +342,11 @@ const viVN = {
       "Uses the Workspace owner's LangBot Account billing and credits.",
   },
   bots: {
+    applyFailed: 'Đã lưu cấu hình nhưng không thể áp dụng',
+    internalErrorHint:
+      'Đã xảy ra lỗi nội bộ. Hãy kiểm tra nhật ký máy chủ bằng mã lỗi.',
+    errorReference: 'Mã lỗi: {{id}}',
+
     adapterEventDebugAction: 'Kiểm tra lắng nghe',
     title: 'Bot',
     description:

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

@@ -325,6 +325,10 @@ const zhHans = {
     },
   },
   bots: {
+    applyFailed: '配置已保存，但应用失败',
+    internalErrorHint: '发生内部错误，请通过错误编号查看后端日志。',
+    errorReference: '错误编号：{{id}}',
+
     title: '机器人',
     description: '创建和管理机器人，这是 LangBot 与各个平台连接的入口',
     createBot: '创建机器人',

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

@@ -322,6 +322,10 @@ const zhHant = {
       "Uses the Workspace owner's LangBot Account billing and credits.",
   },
   bots: {
+    applyFailed: '設定已儲存，但套用失敗',
+    internalErrorHint: '發生內部錯誤，請透過錯誤編號查看後端日誌。',
+    errorReference: '錯誤編號：{{id}}',
+
     adapterEventDebugAction: '測試監聽',
     title: '機器人',
     description: '建立和管理機器人，這是 LangBot 與各個平台連接的入口',

web/tests/e2e/bot-save-errors.spec.ts

@@ -0,0 +1,55 @@
+import { expect, test } from '@playwright/test';
+import { installLangBotApiMocks } from './fixtures/langbot-api';
+
+for (const failure of [
+  {
+    status: 400,
+    code: 'invalid_bot_config',
+    msg: 'Lark missing required config: app_id, app_secret, bot_name',
+  },
+  {
+    status: 400,
+    code: 'bot_apply_failed',
+    msg: 'Lark missing required config: app_id, app_secret, bot_name',
+  },
+  {
+    status: 500,
+    code: 'internal_error',
+    msg: 'Internal server error',
+    request_id: 'bot-save-test-reference',
+  },
+]) {
+  test(`bot save displays actionable ${failure.code}`, async ({ page }) => {
+    await installLangBotApiMocks(page, { authenticated: true });
+    await page.goto('/home/bots?id=new');
+    await page.getByRole('combobox').click();
+    await page.getByRole('option', { name: 'Playwright Adapter' }).click();
+    await page.locator('input[name="name"]').fill('Error Test Bot');
+    await page.getByRole('button', { name: /^Submit$/ }).click();
+    await expect(page).toHaveURL(/\/home\/bots\?id=bot-1$/);
+    await page.route('**/api/v1/platform/bots/bot-1', async (route) => {
+      if (route.request().method() !== 'PUT') return route.fallback();
+      await route.fulfill({
+        status: failure.status,
+        contentType: 'application/json',
+        body: JSON.stringify(failure),
+      });
+    });
+    await page.getByRole('button', { name: 'Edit basic information' }).click();
+    const dialog = page.getByRole('dialog');
+    await dialog.getByLabel('Name', { exact: true }).fill('Edited Bot');
+    await dialog.getByRole('button', { name: /^Save$/ }).click();
+    if (failure.code === 'internal_error') {
+      await expect(
+        page.getByText('Error reference: bot-save-test-reference'),
+      ).toBeVisible();
+    } else {
+      await expect(page.getByText(failure.msg, { exact: true })).toBeVisible();
+    }
+    if (failure.code === 'bot_apply_failed') {
+      await expect(
+        page.getByText('Configuration saved, but could not be applied'),
+      ).toBeVisible();
+    }
+  });
+}