fix: every /bg entry point joins its background delegations in the run

The gateway /bg fix above covered messaging chats only. The shared-owner
/bg (prompt.background, used by the CLI, TUI and Desktop) and the classic
local CLI /bg run the same detached bg_* side agent, and on main a
background delegate_task there also leaves a completed ledger row pending
forever: nothing wakes or drains a bg_* session. Their side agents now
declare the channel stateless, so the batch runs inline and the result is
in the one answer /bg reports. The sync-fallback note names /bg.

Refs #140569.

Co-authored-by: milnerrad <302432023+milnerrad@users.noreply.github.com>

gateway/session_tui_side.py

@@ -66,7 +66,10 @@ def _side_scope(authority, policy, task_id):
     home = authority.profile_id
     profile = served_profile_name(home) if _is_routed_home(home) else ''
     with owner_scope(authority, hydrate_secrets=True), policy_scope(policy, authority=authority):
-        tokens = set_session_vars(session_key=task_id, session_id=task_id, cwd=policy.cwd, profile=profile)
+        # async_delivery=False: the side agent's answer is one ``*.complete`` event, and its bg_* session
+        # has no later turn to wake, so a background delegate_task joins in this run (#140569).
+        tokens = set_session_vars(session_key=task_id, session_id=task_id, cwd=policy.cwd, profile=profile,
+                                  async_delivery=False)
         try:
             # The session's workspace, as its own turns get it (#140665): mounted at /workspace in a
             # per-task container, or the session's own persistent workspace container (same digest).

hermes_cli/cli_commands_mixin.py

@@ -1362,6 +1362,8 @@ class CLICommandsMixin(CLICommandsSessionToolsMixin):
 
         def produce():
             from agent.vault_backends.unlock import set_code_prompt_callback, set_save_login_prompt_callback, set_unlock_prompt_callback
+            from gateway.session_context import declare_stateless_channel
+            declare_stateless_channel()  # nothing drains a bg_* session's completions later: delegations join (#140569)
             set_sudo_password_callback(self._sudo_password_callback)
             set_approval_callback(self._approval_callback)
             set_unlock_prompt_callback(self._vault_unlock_callback)
@@ -1402,10 +1404,8 @@ class CLICommandsMixin(CLICommandsSessionToolsMixin):
                         response = _gt("model.error_prefix", error=result["error"])
                     return response
                 finally:
-                    # One agent per /bg task in a long-lived CLI process: close()
-                    # is the owner boundary (memory shutdown, tool subprocesses,
-                    # httpx clients); an unclosed side agent leaks all of them
-                    # until the CLI exits (#50197).
+                    # One agent per /bg task in a long-lived CLI: close() is the owner boundary (memory,
+                    # tool subprocesses, httpx clients); unclosed, they leak until the CLI exits (#50197).
                     with suppress(Exception):
                         bg_agent.close()
             finally:

tests/hermes_cli/test_cli_background_delegation.py

@@ -0,0 +1,51 @@
+"""Classic-CLI /bg: a background delegate_task joins inside the bg_* run (#140569).
+
+The /bg answer is printed once and nothing drains completions for its bg_* session, so a detached
+child's result would sit in the ledger forever.
+"""
+import json
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import cli as cli_module
+from tests.hermes_cli.test_cli_approval_ui import _make_background_cli_stub
+
+
+def test_background_task_delegation_runs_inline(monkeypatch):
+    import tools.delegate_tool as dt
+
+    monkeypatch.setattr(dt, "_build_child_agent", lambda **kw: SimpleNamespace(_subagent_id="child"))
+    monkeypatch.setattr(dt, "_resolve_delegation_credentials", lambda *a, **kw: {
+        "model": "test", "provider": None, "base_url": None, "api_key": None,
+        "api_mode": None, "command": None, "args": None,
+    })
+    monkeypatch.setattr(dt, "_run_single_child", lambda **kw: {
+        "task_index": 0, "status": "completed", "summary": "child work product",
+        "api_calls": 1, "duration_seconds": 0, "model": "test", "exit_reason": "completed",
+    })
+    detached = MagicMock(return_value={"delegation_id": "detached"})
+    monkeypatch.setattr("tools.async_delegation.dispatch_async_delegation_batch", detached)
+    seen = {}
+
+    class FakeAgent:
+        def __init__(self, **kwargs):
+            self._print_fn = None
+            self.thinking_callback = None
+
+        def run_conversation(self, **kwargs):
+            parent = SimpleNamespace(_delegate_depth=0, _subagent_id=None)
+            seen["tool"] = dt.delegate_task(goal="inspect workspace", background=True, parent_agent=parent)
+            return {"final_response": "done", "messages": []}
+
+        def close(self):
+            pass
+
+    cli = _make_background_cli_stub()
+    with patch("run_agent.AIAgent", FakeAgent), patch.object(cli_module, "_cprint"), \
+            patch.object(cli_module, "ChatConsole"):
+        cli._handle_background_command("/bg inspect workspace")
+        for thread in list(cli._background_tasks.values()):
+            thread.join(timeout=10)
+
+    detached.assert_not_called()
+    assert "child work product" in json.dumps(seen["tool"])

tools/delegate_tool_dispatch.py

@@ -229,8 +229,8 @@ _SYNC_FALLBACK_NOTES = {
     "no_async": (
         "background=true is not available in this session — it cannot "
         "receive a detached subagent result after the turn ends (a "
-        "finite chat using -Q, --oneshot, or non-TTY stdio, `hermes -z`, a cron job, a Kanban "
-        "worker, or a stateless HTTP endpoint). The subagent(s) ran SYNCHRONOUSLY and the result is included above."
+        "finite chat using -Q, --oneshot, or non-TTY stdio, `hermes -z`, a /bg background task, a cron "
+        "job, a Kanban worker, or a stateless HTTP endpoint). The subagent(s) ran SYNCHRONOUSLY and the result is included above."
     ),
     "at_capacity": (
         "The background delegation pool was at capacity (delegation.max_concurrent_children), so the subagent(s) ran "