Feat/chat summary compact phase

backend/app/api/ws/chat_namespace.py

@@ -72,7 +72,11 @@
 from app.services.chat.rag import process_context_and_rag
 from app.services.chat.storage import session_manager
 from app.services.chat.storage.db import get_db_session, run_sync_in_executor
-from app.services.chat.trigger import trigger_ai_response_unified
+from app.services.chat.trigger import (
+    collect_completed_result,
+    persist_completed_result,
+    trigger_ai_response_unified,
+)
 from app.services.chat.wework_task_defaults import apply_wework_task_defaults
 from app.services.task_fork_history import task_fork_history_resolver
 from app.utils.prompt_utils import extract_display_prompt
@@ -84,6 +88,29 @@
 logger = logging.getLogger(__name__)
 
 
+async def _finalize_failed_ai_trigger(
+    *,
+    task_id: int,
+    assistant_subtask_id: int,
+    error_message: str,
+    error_code: str,
+) -> None:
+    """Persist a failed async AI trigger so follow-up messages are not blocked."""
+    final_result = await collect_completed_result(
+        assistant_subtask_id,
+        status="FAILED",
+        error_message=error_message,
+        error_code=error_code,
+    )
+    await persist_completed_result(
+        subtask_id=assistant_subtask_id,
+        task_id=task_id,
+        status="FAILED",
+        result=final_result,
+        error=error_message,
+    )
+
+
 class ChatNamespace(socketio.AsyncNamespace):
     """
     Socket.IO namespace for chat functionality.
@@ -496,6 +523,7 @@ async def on_task_join(self, sid: str, data: dict) -> dict:
                     "started_at": streaming_info.get("started_at"),
                     "last_activity_at": streaming_info.get("last_activity_at"),
                 },
+                "status_updated": cached_context_metrics,
                 "subtasks": subtasks_dict,
             }
 
@@ -989,12 +1017,31 @@ async def _trigger_ai():
                         )
 
                         error_code = classify_error(e)
+                        error_message = format_error_message(e)
+
+                        try:
+                            await _finalize_failed_ai_trigger(
+                                task_id=task.id,
+                                assistant_subtask_id=assistant_subtask.id,
+                                error_message=error_message,
+                                error_code=error_code,
+                            )
+                        except Exception as finalize_error:
+                            logger.error(
+                                "[WS] chat:send failed to persist async trigger error: "
+                                "task_id=%s, subtask_id=%s, error=%s",
+                                task.id,
+                                assistant_subtask.id,
+                                finalize_error,
+                                exc_info=True,
+                            )
+
                         # Emit error to frontend so user sees the failure
                         await self.emit(
                             ServerEvents.CHAT_ERROR,
                             ChatErrorPayload(
                                 subtask_id=assistant_subtask.id,
-                                error=format_error_message(e),
+                                error=error_message,
                                 type=error_code,
                                 message_id=assistant_subtask.message_id,
                                 task_id=task.id,

backend/app/api/ws/events.py

@@ -762,6 +762,9 @@ class TaskJoinAck(BaseModel):
     subtasks: Optional[List[Dict[str, Any]]] = Field(
         None, description="Subtasks data for immediate message sync"
     )
+    status_updated: Optional[Dict[str, Any]] = Field(
+        None, description="Latest cached chat:status_updated snapshot"
+    )
     error: Optional[str] = None
 
 

backend/app/core/config.py

@@ -178,9 +178,6 @@ class Settings(BaseSettings):
     CHAT_TOOL_MAX_TIME_SECONDS: float = (
         60.0  # Maximum time for tool calling flow (5 minutes)
     )
-    # Source-level tool output compaction default for users without an explicit
-    # preference override.
-    TOOL_OUTPUT_GUARD_ENABLED: bool = False
     # Group chat history configuration
     # In group chat mode, AI-bot sees: first N messages + last M messages (no duplicates)
     # If total messages < N + M, all messages are kept

backend/app/services/chat/storage/session.py

@@ -48,6 +48,11 @@
 # Internal field stored in Redis block metadata. It is stripped before returning
 # blocks to callers so API consumers still see the existing block shape.
 BLOCK_CONTENT_KEY_FIELD = "_content_key"
+UNRESOLVED_PREVIEW_TOOL_BLOCK_STATUSES = {"pending", "generating_arguments"}
+UNRESOLVED_PREVIEW_TOOL_BLOCK_MESSAGE = "Tool call was not executed before the turn completed. The turn may have hit the tool-call limit."
+UNRESOLVED_PREVIEW_TOOL_BLOCK_GENERIC_MESSAGE = (
+    "Tool call preview did not complete before the turn ended."
+)
 
 
 class StreamContentType(str, Enum):
@@ -1317,7 +1322,12 @@ async def get_blocks(self, subtask_id: int) -> List[Dict[str, Any]]:
             )
             return []
 
-    async def finalize_and_get_blocks(self, subtask_id: int) -> List[Dict[str, Any]]:
+    async def finalize_and_get_blocks(
+        self,
+        subtask_id: int,
+        *,
+        termination_reason: Optional[str] = None,
+    ) -> List[Dict[str, Any]]:
         """Finalize any pending text block and return all blocks.
 
         This should be called when the subtask completes (DONE event).
@@ -1338,6 +1348,10 @@ async def finalize_and_get_blocks(self, subtask_id: int) -> List[Dict[str, Any]]
             redis_client = await self._cache._get_client()
             try:
                 blocks = await self._load_blocks_from_client(redis_client, blocks_key)
+                blocks = self._finalize_unresolved_preview_tool_blocks(
+                    blocks,
+                    termination_reason=termination_reason,
+                )
                 logger.info(
                     f"[SessionManager] Finalized blocks for subtask {subtask_id}: "
                     f"count={len(blocks)}"
@@ -1351,6 +1365,54 @@ async def finalize_and_get_blocks(self, subtask_id: int) -> List[Dict[str, Any]]
             )
             return []
 
+    def _finalize_unresolved_preview_tool_blocks(
+        self,
+        blocks: List[Dict[str, Any]],
+        *,
+        termination_reason: Optional[str] = None,
+    ) -> List[Dict[str, Any]]:
+        """Convert unresolved preview-only tool blocks into terminal error blocks.
+
+        Preview tool blocks are created while tool arguments stream, before real tool
+        execution begins. If the turn ends without a matching TOOL_RESULT, those
+        blocks would otherwise stay stuck in a pending state in the final result.
+
+        Only preview-only blocks without tool_output are rewritten. Legitimate
+        pending blocks that already carry a semantic output payload (for example
+        deferred interactive forms waiting for user input) are preserved.
+        """
+        inferred_tool_limit = (
+            termination_reason == "completed_with_unexecuted_tool_calls"
+        )
+        has_explicit_tool_error = any(
+            block.get("type") == "tool"
+            and block.get("status") == BlockStatus.ERROR.value
+            and block.get("tool_output")
+            for block in blocks
+        )
+        unresolved_message = (
+            UNRESOLVED_PREVIEW_TOOL_BLOCK_MESSAGE
+            if inferred_tool_limit and not has_explicit_tool_error
+            else UNRESOLVED_PREVIEW_TOOL_BLOCK_GENERIC_MESSAGE
+        )
+        finalized_blocks: List[Dict[str, Any]] = []
+        for block in blocks:
+            if (
+                block.get("type") == "tool"
+                and block.get("status") in UNRESOLVED_PREVIEW_TOOL_BLOCK_STATUSES
+                and not block.get("tool_output")
+            ):
+                finalized_blocks.append(
+                    {
+                        **block,
+                        "status": BlockStatus.ERROR.value,
+                        "tool_output": unresolved_message,
+                    }
+                )
+                continue
+            finalized_blocks.append(block)
+        return finalized_blocks
+
     async def get_accumulated_content(self, subtask_id: int) -> str:
         """Get accumulated content for a subtask.
 

backend/app/services/chat/trigger/lifecycle.py

@@ -408,7 +408,6 @@ async def collect_completed_result(
     accumulated_content = await chat_storage.session_manager.get_accumulated_content(
         subtask_id
     )
-    blocks = await chat_storage.session_manager.finalize_and_get_blocks(subtask_id)
     existing_result = await _get_existing_subtask_result(subtask_id)
 
     if result is not None and not isinstance(result, dict):
@@ -419,6 +418,23 @@ async def collect_completed_result(
         )
 
     runtime_result = dict(result) if isinstance(result, dict) else {}
+    termination_reason = runtime_result.get("termination_reason")
+    if not isinstance(termination_reason, str) or not termination_reason:
+        existing_termination_reason = existing_result.get("termination_reason")
+        termination_reason = (
+            existing_termination_reason
+            if isinstance(existing_termination_reason, str)
+            and existing_termination_reason
+            else None
+        )
+    if termination_reason:
+        runtime_result["termination_reason"] = termination_reason
+    else:
+        runtime_result.pop("termination_reason", None)
+    blocks = await chat_storage.session_manager.finalize_and_get_blocks(
+        subtask_id,
+        termination_reason=termination_reason,
+    )
 
     has_payload = bool(
         runtime_result

backend/app/services/chat/webpage_ws_chat_emitter.py

@@ -248,16 +248,20 @@ async def emit_chat_status_updated(
         subtask_id: int,
         phase: str,
         context_metrics: Dict[str, Any],
+        context_compaction: Dict[str, Any] | None = None,
     ) -> None:
         """Emit chat:status_updated event to task room."""
+        payload: Dict[str, Any] = {
+            "task_id": task_id,
+            "subtask_id": subtask_id,
+            "phase": phase,
+            "context_metrics": context_metrics,
+        }
+        if context_compaction is not None:
+            payload["context_compaction"] = context_compaction
         await self.sio.emit(
             ServerEvents.CHAT_STATUS_UPDATED,
-            {
-                "task_id": task_id,
-                "subtask_id": subtask_id,
-                "phase": phase,
-                "context_metrics": context_metrics,
-            },
+            payload,
             room=f"task:{task_id}",
             namespace=self.namespace,
         )

backend/app/services/execution/dispatcher.py

@@ -171,8 +171,10 @@ def extract_completed_result(response_data: dict) -> dict:
         ),
         "loaded_skills": response_data.get("loaded_skills"),
         "stop_reason": response_data.get("stop_reason"),
+        "termination_reason": response_data.get("termination_reason"),
         "messages_chain": response_data.get("messages_chain"),
         "context_metrics": response_data.get("context_metrics"),
+        "context_compactions": response_data.get("context_compactions"),
         "standalone_chat_workspace_path": response_data.get(
             "standalone_chat_workspace_path"
         ),
@@ -249,6 +251,7 @@ def parse(
                 data={
                     "phase": data.get("phase"),
                     "context_metrics": data.get("context_metrics") or {},
+                    "context_compaction": data.get("context_compaction"),
                 },
                 message_id=message_id,
             )