fix(soul): keep agent loop alive while background tasks are running

CHANGELOG.md

@@ -11,6 +11,7 @@ Only write entries that are worth mentioning to users.
 
 ## Unreleased
 
+- Print: Wait for background tasks before exiting — in one-shot `--print` mode, the process now waits for running background agents to finish and lets the model process their results, instead of exiting and killing them
 - Core: Fix agent loop silently stopping when model response contains only thinking content — detect think-only responses (reasoning content with no text or tool calls) as an incomplete response error and retry automatically
 - Core: Fix crash on streaming mid-flight network disconnection — when the OpenAI SDK raises a base `APIError` (instead of `APIConnectionError`) during long-running streams, the error is now correctly classified as retryable, enabling automatic retry and connection recovery instead of an unrecoverable crash
 - Shell: Exclude empty current session from `/sessions` picker — completely empty sessions (no conversation history and no custom title) are no longer shown in the session list; sessions with a custom title are still displayed

docs/en/release-notes/changelog.md

@@ -4,6 +4,7 @@ This page documents the changes in each Kimi Code CLI release.
 
 ## Unreleased
 
+- Print: Wait for background tasks before exiting — in one-shot `--print` mode, the process now waits for running background agents to finish and lets the model process their results, instead of exiting and killing them
 - Core: Fix agent loop silently stopping when model response contains only thinking content — detect think-only responses (reasoning content with no text or tool calls) as an incomplete response error and retry automatically
 - Core: Fix crash on streaming mid-flight network disconnection — when the OpenAI SDK raises a base `APIError` (instead of `APIConnectionError`) during long-running streams, the error is now correctly classified as retryable, enabling automatic retry and connection recovery instead of an unrecoverable crash
 - Shell: Exclude empty current session from `/sessions` picker — completely empty sessions (no conversation history and no custom title) are no longer shown in the session list; sessions with a custom title are still displayed

docs/zh/release-notes/changelog.md

@@ -4,6 +4,7 @@
 
 ## 未发布
 
+- Print：退出前等待后台任务完成——在单次 `--print` 模式下，进程现在会等待仍在运行的后台 Agent 完成并让模型处理它们的结果，而不是直接退出并杀死它们
 - Core：修复模型响应仅包含思考内容时 agent loop 静默停止的问题——将仅含思考内容（无文本或工具调用）的响应检测为不完整响应错误并自动重试
 - Core：修复长时间 streaming 过程中网络断连导致崩溃的问题——当 OpenAI SDK 在流式传输中途抛出基类 `APIError`（而非 `APIConnectionError`）时，现在能正确识别为可重试错误，自动触发重试和连接恢复，而不再直接崩溃退出
 - Shell：从 `/sessions` 选择器中排除空的当前会话——完全为空的会话（既无对话记录也无自定义标题）不再显示在会话列表中；有自定义标题的会话仍然正常显示

src/kimi_cli/background/manager.py

@@ -93,6 +93,14 @@ def _active_task_count(self) -> int:
             1 for view in self._store.list_views() if not is_terminal_status(view.runtime.status)
         )
 
+    def has_active_tasks(self) -> bool:
+        """Return True if any background tasks are in a non-terminal status.
+
+        This includes ``running``, ``awaiting_approval``, and any other
+        non-terminal state — not just actively executing tasks.
+        """
+        return self._active_task_count() > 0
+
     def _worker_command(self, task_dir: Path) -> list[str]:
         if getattr(sys, "frozen", False):
             return [

src/kimi_cli/soul/kimisoul.py

@@ -758,6 +758,7 @@ async def _agent_loop(self) -> TurnOutcome:
                 has_steers = await self._consume_pending_steers()
                 if has_steers:
                     continue  # steers injected, force another LLM step
+
                 final_message = (
                     step_outcome.assistant_message
                     if step_outcome.stop_reason == "no_tool_calls"

src/kimi_cli/ui/print/__init__.py

@@ -96,6 +96,74 @@ def _handler():
                         runtime.session.wire_file if runtime else None,
                         runtime,
                     )
+
+                    # In one-shot text mode the process exits after this
+                    # function returns, which would kill still-running
+                    # background agents.  Poll until they finish, calling
+                    # reconcile() each iteration (the notification pump
+                    # inside run_soul is no longer running, so we must
+                    # drive reconcile ourselves to recover lost workers
+                    # and publish terminal notifications).  Only re-enter
+                    # the soul when there are pending LLM notifications.
+                    #
+                    # stream-json mode is multi-turn: background tasks
+                    # from one command must not block the next command.
+                    #
+                    # keep_alive_on_exit opts into "fire and forget"
+                    # semantics: background tasks are meant to outlive
+                    # the CLI process, so Print must not block waiting
+                    # for them.
+                    if (
+                        runtime
+                        and runtime.role == "root"
+                        and self.input_format == "text"
+                        and not runtime.config.background.keep_alive_on_exit
+                    ):
+                        manager = runtime.background_tasks
+                        notifications = runtime.notifications
+                        while not cancel_event.is_set():
+                            # Drive reconcile() ourselves: the notification
+                            # pump inside run_soul is no longer running, so
+                            # we must recover lost workers and publish
+                            # terminal notifications here.
+                            manager.reconcile()
+                            if notifications.has_pending_for_sink("llm"):
+                                # Re-enter soul so the LLM can process the
+                                # completion notification.  Do this even if
+                                # other tasks are still active — progress on
+                                # completed tasks should not wait on siblings.
+                                bg_prompt = (
+                                    "<system-reminder>"
+                                    "Background tasks have completed."
+                                    " Process their results."
+                                    "</system-reminder>"
+                                )
+                                await run_soul(
+                                    self.soul,
+                                    bg_prompt,
+                                    partial(visualize, self.output_format, self.final_only),
+                                    cancel_event,
+                                    runtime.session.wire_file,
+                                    runtime,
+                                )
+                                continue
+                            if not manager.has_active_tasks():
+                                # Re-check once after noticing no active
+                                # tasks: a worker may have finished between
+                                # the reconcile above and this snapshot,
+                                # leaving a terminal state on disk that we
+                                # haven't published yet.  Without this
+                                # second reconcile+pending check, that
+                                # final completion notification would be
+                                # lost when the process exits.
+                                manager.reconcile()
+                                if notifications.has_pending_for_sink("llm"):
+                                    continue
+                                break
+                            # Still waiting for tasks to finish.
+                            await asyncio.sleep(1.0)
+                        if cancel_event.is_set():
+                            raise RunCancelled
                 else:
                     logger.info("Empty command, skipping")