feat: Enhance error handling and caching for Dify and WeCom interactions

src/langbot/libs/dify_service_api/v1/client.py

@@ -156,9 +156,10 @@ class AsyncDifyServiceClient:
                 headers={'Authorization': f'Bearer {self.api_key}'},
                 params={'user': user},
             ) as r:
+                if r.status_code != 200:
+                    body = (await r.aread()).decode(errors='replace')
+                    raise DifyAPIError(f'{r.status_code} {body}')
                 async for chunk in r.aiter_lines():
-                    if r.status_code != 200:
-                        raise DifyAPIError(f'{r.status_code} {chunk}')
                     if chunk.strip() == '':
                         continue
                     if chunk.startswith('data:'):

src/langbot/libs/wecom_ai_bot_api/api.py

@@ -2042,19 +2042,18 @@ class WecomBotClient:
 
         previous_content = self._stream_last_content.get(msg_id, '')
         if previous_content and content.startswith(previous_content):
-            delta_content = content[len(previous_content) :]
             next_content = content
         elif previous_content and not content:
-            delta_content = ''
             next_content = previous_content
         else:
-            delta_content = content
             next_content = previous_content + content if previous_content else content
 
-        if not is_final and not delta_content:
+        if not is_final and next_content == previous_content:
             return True
 
-        chunk = StreamChunk(content=delta_content, is_final=is_final)
+        # Follow-up responses replace the displayed stream body in WeCom.
+        # Publish the complete snapshot so earlier chunks remain visible.
+        chunk = StreamChunk(content=next_content, is_final=is_final)
         await self.stream_sessions.publish(stream_id, chunk)
         self._stream_last_content[msg_id] = next_content
         if is_final:

src/langbot/libs/wecom_ai_bot_api/ws_client.py

@@ -453,20 +453,17 @@ class WecomBotWsClient:
         try:
             previous_content = self._stream_last_content.get(msg_id, '')
             if previous_content and content.startswith(previous_content):
-                delta_content = content[len(previous_content) :]
                 next_content = content
             elif previous_content and not content:
-                delta_content = ''
                 next_content = previous_content
             else:
-                delta_content = content
                 next_content = previous_content + content if previous_content else content
 
             # Skip sending if content hasn't changed (e.g. during tool call argument streaming)
-            if not is_final and not delta_content:
+            if not is_final and next_content == previous_content:
                 return True
 
-            # Skip empty/whitespace-only chunks — the runner injects a
+            # Skip empty/whitespace-only snapshots — the runner injects a
             # zero-width space ('​') as a pass-through when workflow_paused
             # fires without any preceding LLM output.  WeCom renders that
             # as an empty bubble that sits before the form card; skip it.
@@ -476,7 +473,7 @@ class WecomBotWsClient:
             if not is_final:
                 import re as _re
 
-                if not _re.sub(r'[\s​‌‍]', '', delta_content):
+                if not _re.sub(r'[\s​‌‍]', '', next_content):
                     return True
 
             # Generate feedback_id for final chunk
@@ -489,7 +486,9 @@ class WecomBotWsClient:
                 if session_info:
                     self._feedback_sessions[feedback_id] = session_info
 
-            await self.reply_stream(req_id, stream_id, delta_content, finish=is_final, feedback_id=feedback_id)
+            # WeCom replaces the displayed stream content on each refresh, so
+            # every frame must contain the complete snapshot, not only a delta.
+            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._stream_ids.pop(msg_id, None)

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

@@ -10,6 +10,8 @@ import typing
 import traceback
 import json
 import base64
+import time
+import uuid
 import pydantic
 
 from langbot.pkg.utils import httpclient
@@ -260,7 +262,38 @@ class TelegramAdapter(abstract_platform_adapter.AbstractMessagePlatformAdapter):
         typing.Callable[[platform_events.Event, abstract_platform_adapter.AbstractMessagePlatformAdapter], None],
     ] = {}
 
-    _form_action_titles: typing.Dict[str, str] = {}  # callback_data -> display title
+    _FORM_ACTION_CACHE_TTL = 30 * 60
+    # callback_data -> (display title, expiration monotonic time, form group id)
+    _form_action_titles: typing.Dict[str, tuple[str, float, str]] = {}
+
+    def _prune_form_action_titles(self, now: float | None = None) -> None:
+        now = time.monotonic() if now is None else now
+        expired = [key for key, (_, expires_at, _) in self._form_action_titles.items() if expires_at <= now]
+        for key in expired:
+            self._form_action_titles.pop(key, None)
+
+    def _cache_form_action_titles(self, mappings: dict[str, str], now: float | None = None) -> None:
+        now = time.monotonic() if now is None else now
+        self._prune_form_action_titles(now)
+        group_id = uuid.uuid4().hex
+        expires_at = now + self._FORM_ACTION_CACHE_TTL
+        self._form_action_titles.update(
+            {callback_data: (title, expires_at, group_id) for callback_data, title in mappings.items()}
+        )
+
+    def _take_form_action_title(self, callback_data: str, now: float | None = None) -> str | None:
+        """Consume a callback and invalidate every button from the same form."""
+        self._prune_form_action_titles(now)
+        entry = self._form_action_titles.get(callback_data)
+        if entry is None:
+            return None
+        title, _, group_id = entry
+        group_keys = [
+            key for key, (_, _, cached_group_id) in self._form_action_titles.items() if cached_group_id == group_id
+        ]
+        for key in group_keys:
+            self._form_action_titles.pop(key, None)
+        return title
 
     def __init__(self, config: dict, logger: abstract_platform_logger.AbstractEventLogger):
         async def telegram_callback(update: Update, context: ContextTypes.DEFAULT_TYPE):
@@ -297,11 +330,11 @@ class TelegramAdapter(abstract_platform_adapter.AbstractMessagePlatformAdapter):
                     w_suffix = data.get('w', '')
                     session_key = data.get('session_key') or data.get('s', '')
                     callback_action = _telegram_form_action_from_callback(data)
-                    if callback_action is None or query.data not in self._form_action_titles:
+                    action_title = self._take_form_action_title(query.data) if callback_action is not None else None
+                    if callback_action is None or action_title is None:
                         await self.logger.warning(f'Invalid or stale Telegram form callback: {query.data!r}')
                         return
                     # Show selected action feedback by editing the original message
-                    action_title = self._form_action_titles[query.data]
                     try:
                         original_text = query.message.text or ''
                         selected_text = f'{original_text}\n\n✅ {action_title}'
@@ -309,9 +342,6 @@ class TelegramAdapter(abstract_platform_adapter.AbstractMessagePlatformAdapter):
                     except Exception:
                         # If edit fails (e.g. message too long), just pass
                         pass
-                    finally:
-                        # Clean up the stored title
-                        self._form_action_titles.pop(query.data, None)
 
                     if session_key.startswith('group_') or session_key.startswith('g:'):
                         launcher_type = provider_session.LauncherTypes.GROUP
@@ -702,7 +732,7 @@ class TelegramAdapter(abstract_platform_adapter.AbstractMessagePlatformAdapter):
                 'text': '\n\n'.join(text_lines),
             }
         elif keyboard:
-            self._form_action_titles.update(pending_title_mappings)
+            self._cache_form_action_titles(pending_title_mappings)
             reply_markup = InlineKeyboardMarkup(keyboard)
             args = {
                 'chat_id': chat_id,

src/langbot/pkg/provider/runners/difysvapi.py

@@ -1235,9 +1235,7 @@ class DifyServiceAPIRunner(runner.RequestRunner):
 
         if not saw_event:
             raise errors.DifyAPIError('Dify API did not return any workflow resume events')
-        if pending_content:
+        raise errors.DifyAPIError('Dify workflow resume stream ended before a terminal event')
-            content, _ = self._process_thinking_content(pending_content)
-            yield provider_message.Message(role='assistant', content=content)
 
     def _resolve_pending_form(self, session_key: PendingFormKey, form_action: dict) -> dict | None:
         """Locate the pending form this action targets.
@@ -2001,6 +1999,8 @@ class DifyServiceAPIRunner(runner.RequestRunner):
                 if is_final:
                     return
 
+        raise errors.DifyAPIError('Dify workflow resume stream ended before a terminal event')
+
     async def _workflow_messages_chunk(
         self, query: pipeline_query.Query
     ) -> typing.AsyncGenerator[provider_message.MessageChunk, None]:

tests/unit_tests/platform/test_telegram_adapter.py

@@ -53,6 +53,26 @@ def test_telegram_invalid_select_callback_is_rejected():
     assert _telegram_form_action_from_callback({'f': 1, 'x': 'invalid'}) is None
 
 
+def test_telegram_form_callback_cache_consumes_the_whole_form_group():
+    adapter = TelegramAdapter.model_construct()
+    adapter._form_action_titles = {}
+    adapter._cache_form_action_titles({'callback-a': 'A', 'callback-b': 'B'}, now=100.0)
+
+    assert adapter._take_form_action_title('callback-a', now=101.0) == 'A'
+    assert adapter._take_form_action_title('callback-a', now=101.0) is None
+    assert adapter._take_form_action_title('callback-b', now=101.0) is None
+    assert adapter._form_action_titles == {}
+
+
+def test_telegram_form_callback_cache_prunes_expired_entries():
+    adapter = TelegramAdapter.model_construct()
+    adapter._form_action_titles = {}
+    adapter._cache_form_action_titles({'callback-a': 'A'}, now=100.0)
+
+    assert adapter._take_form_action_title('callback-a', now=100.0 + adapter._FORM_ACTION_CACHE_TTL) is None
+    assert adapter._form_action_titles == {}
+
+
 @pytest.mark.asyncio
 async def test_telegram_select_field_sends_two_column_inline_keyboard():
     bot = MagicMock()

tests/unit_tests/platform/test_wecombot_template_card.py

@@ -9,6 +9,7 @@ logger_module.EventLogger = object
 sys.modules.setdefault('langbot.pkg.platform.logger', logger_module)
 
 from langbot.libs.wecom_ai_bot_api.api import (  # noqa: E402
+    WecomBotClient,
     build_button_interaction_payload,
     build_human_input_template_card_payload,
     build_button_interaction_update_card,
@@ -254,7 +255,7 @@ async def test_ws_push_form_pause_sends_text_prompt_without_empty_card():
 
 
 @pytest.mark.asyncio
-async def test_ws_stream_sends_delta_chunks_to_wecom():
+async def test_ws_stream_sends_cumulative_snapshots_to_wecom():
     client = WecomBotWsClient('bot-id', 'secret', object())
     client._stream_ids['msg-1'] = 'req-1|stream-1'
     client._stream_sessions['msg-1'] = {}
@@ -272,8 +273,29 @@ async def test_ws_stream_sends_delta_chunks_to_wecom():
 
     assert sent == [
         ('req-1', 'stream-1', '你', False),
-        ('req-1', 'stream-1', '好', False),
+        ('req-1', 'stream-1', '你好', False),
-        ('req-1', 'stream-1', '', True),
+        ('req-1', 'stream-1', '你好', True),
+    ]
+
+
+@pytest.mark.asyncio
+async def test_webhook_stream_queues_cumulative_snapshots_for_followups():
+    client = WecomBotClient('', '', '', object(), unified_mode=True)
+    session, _ = client.stream_sessions.create_or_get({'msgid': 'msg-1', 'chatid': '', 'from': {'userid': 'user-1'}})
+
+    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', '你好', is_final=True)
+
+    chunks = [
+        await client.stream_sessions.consume(session.stream_id),
+        await client.stream_sessions.consume(session.stream_id),
+        await client.stream_sessions.consume(session.stream_id),
+    ]
+    assert [(chunk.content, chunk.is_final) for chunk in chunks] == [
+        ('你', False),
+        ('你好', False),
+        ('你好', True),
     ]
 
 

tests/unit_tests/provider/runners/test_difysvapi_runner.py

@@ -11,6 +11,62 @@ from unittest.mock import AsyncMock, MagicMock
 import langbot_plugin.api.entities.builtin.platform.message as platform_message
 
 
+class TestDifyWorkflowSubmitClient:
+    @pytest.mark.asyncio
+    async def test_rejects_empty_error_response_before_iterating_sse(self, monkeypatch):
+        from langbot.libs.dify_service_api.v1 import client, errors
+
+        class FakeResponse:
+            status_code = 503
+
+            async def aread(self):
+                return b''
+
+            async def aiter_lines(self):
+                raise AssertionError('error responses must not enter the SSE loop')
+                yield
+
+        class FakeStreamContext:
+            async def __aenter__(self):
+                return FakeResponse()
+
+            async def __aexit__(self, exc_type, exc, traceback):
+                return False
+
+        class FakeClient:
+            def __init__(self, **kwargs):
+                del kwargs
+
+            async def __aenter__(self):
+                return self
+
+            async def __aexit__(self, exc_type, exc, traceback):
+                return False
+
+            async def post(self, *args, **kwargs):
+                del args, kwargs
+                response = MagicMock()
+                response.status_code = 200
+                return response
+
+            def stream(self, *args, **kwargs):
+                del args, kwargs
+                return FakeStreamContext()
+
+        monkeypatch.setattr(client.httpx, 'AsyncClient', FakeClient)
+        dify_client = client.AsyncDifyServiceClient('test-key', 'https://dify.example/v1')
+
+        with pytest.raises(errors.DifyAPIError, match='503'):
+            await anext(
+                dify_client.workflow_submit(
+                    form_token='token-1',
+                    workflow_run_id='run-1',
+                    inputs={},
+                    user='person_user-1',
+                )
+            )
+
+
 class TestDifyExtractTextOutput:
     """Tests for _extract_dify_text_output method."""
 
@@ -834,7 +890,7 @@ class TestDifyHumanInputForms:
         assert difysvapi._get_pending_form_by_token(session_key, 'token-1') is None
 
     @pytest.mark.asyncio
-    async def test_failed_blocking_resume_keeps_pending_form_for_retry(self):
+    async def test_incomplete_blocking_resume_keeps_pending_form_for_retry(self):
         from langbot.libs.dify_service_api.v1.errors import DifyAPIError
         from langbot.pkg.provider.runners import difysvapi
 
@@ -843,8 +899,7 @@ class TestDifyHumanInputForms:
 
         async def workflow_submit(**kwargs):
             del kwargs
-            raise DifyAPIError('temporary failure')
+            yield {'event': 'message', 'answer': 'partial answer'}
-            yield
 
         runner.dify_client.workflow_submit = workflow_submit
         query = MagicMock()
@@ -873,12 +928,57 @@ class TestDifyHumanInputForms:
             }
         }
 
-        with pytest.raises(DifyAPIError, match='temporary failure'):
+        with pytest.raises(DifyAPIError, match='before a terminal event'):
             _ = [message async for message in runner._workflow_messages(query)]
 
         assert difysvapi._get_pending_form_by_token(session_key, 'token-1') is not None
         difysvapi._PENDING_FORMS.clear()
 
+    @pytest.mark.asyncio
+    async def test_incomplete_streaming_resume_keeps_pending_form_for_retry(self):
+        from langbot.libs.dify_service_api.v1.errors import DifyAPIError
+        from langbot.pkg.provider.runners import difysvapi
+
+        runner = self._create_runner()
+        query = MagicMock()
+        query.session.launcher_type.value = 'person'
+        query.session.launcher_id = 'user-1'
+        query.bot_uuid = 'bot-1'
+        query.pipeline_uuid = 'pipeline-1'
+        query.variables = {
+            '_dify_form_action': {
+                'form_token': 'token-1',
+                'workflow_run_id': 'run-1',
+                'action_id': 'yes',
+                'user': 'person_user-1',
+                'inputs': {},
+            }
+        }
+        session_key = difysvapi._session_key_from_query(query)
+        difysvapi._PENDING_FORMS.clear()
+        difysvapi._set_pending_form(
+            session_key,
+            {
+                'form_token': 'token-1',
+                'workflow_run_id': 'run-1',
+                'actions': [{'id': 'yes', 'title': 'Yes'}],
+                'inputs': {},
+                'user': 'person_user-1',
+            },
+        )
+
+        async def workflow_submit(**kwargs):
+            del kwargs
+            yield {'event': 'text_chunk', 'data': {'text': 'partial answer'}}
+
+        runner.dify_client.workflow_submit = workflow_submit
+
+        with pytest.raises(DifyAPIError, match='before a terminal event'):
+            _ = [message async for message in runner._workflow_messages_chunk(query)]
+
+        assert difysvapi._get_pending_form_by_token(session_key, 'token-1') is not None
+        difysvapi._PENDING_FORMS.clear()
+
 
 class TestDifyCumulativeStreaming:
     def _create_runner(self, *, remove_think: bool = False):