fix: 4 small surgical bugs

Salvages #23302 by @Bartok9. Four independent one-area fixes:

1. kanban boards delete alias now hard-deletes (not archives) — the
   alias didn't carry --delete, so getattr(args, 'delete', False)
   returned False. Detect boards_action=='delete' explicitly.
2. Gateway auto-title failures no longer leak as user-visible
   warnings — debug-log only since they're not actionable.
3. Background process completion notification snaps truncation to
   the next newline boundary, prepends a marker when content is
   dropped.
4. _cprint() schedules the run_in_terminal coroutine via
   asyncio.ensure_future so output isn't silently dropped from
   background threads (fixes #23185 Bug A). Skips the
   double-print fallback that would fire for mock paths.

cli.py

@@ -1835,13 +1835,26 @@ def _cprint(text: str):
     # prompt, prints, and redraws.  Fire-and-forget — if scheduling
     # fails we fall back to a direct print so the line isn't lost.
     def _schedule():
+        # run_in_terminal() may return either:
+        #   • a coroutine / Future (prompt_toolkit ≥ 3.0) — must be scheduled
+        #     via ensure_future so the coroutine is actually awaited; calling
+        #     it bare would leave it unawaited and silently drop the output
+        #     (fixes #23185 Bug A).
+        #   • None (some mocks / older PT builds) — just call the inner
+        #     function directly since PT already executed it synchronously.
+        # Do NOT fall back to a bare _pt_print when ensure_future raises,
+        # because run_in_terminal already invoked the lambda in that case
+        # (the mock path), which would double-print the line.
         try:
-            run_in_terminal(lambda: _pt_print(_PT_ANSI(text)))
+            import asyncio as _aio
+            import inspect as _inspect
+            coro = run_in_terminal(lambda: _pt_print(_PT_ANSI(text)))
+            if coro is not None and (_inspect.isawaitable(coro) or _inspect.iscoroutine(coro)):
+                _aio.ensure_future(coro)
+            # else: run_in_terminal ran the lambda synchronously; nothing more
+            # to do (double-scheduling would print twice).
         except Exception:
-            try:
+            pass  # best-effort; the line may already have been printed
-                _pt_print(_PT_ANSI(text))
-            except Exception:
-                pass
 
     try:
         loop.call_soon_threadsafe(_schedule)

gateway/run.py

@@ -13956,7 +13956,19 @@ class GatewayRunner:
                 from tools.process_registry import process_registry as _pr_check
                 if agent_notify and not _pr_check.is_completion_consumed(session_id):
                     from tools.ansi_strip import strip_ansi
-                    _out = strip_ansi(session.output_buffer[-2000:]) if session.output_buffer else ""
+                    _raw = strip_ansi(session.output_buffer) if session.output_buffer else ""
+                    # Truncate at line boundaries so notifications never start
+                    # mid-line (fixes #23284). Keep the last ~2000 chars but
+                    # snap to the nearest preceding newline, then prepend a
+                    # truncation marker when output was cut.
+                    _LIMIT = 2000
+                    if len(_raw) > _LIMIT:
+                        _tail = _raw[-_LIMIT:]
+                        _nl = _tail.find("\n")
+                        _tail = _tail[_nl + 1:] if _nl != -1 else _tail
+                        _out = f"[… output truncated — showing last {len(_tail)} chars]\n{_tail}"
+                    else:
+                        _out = _raw
                     synth_text = (
                         f"[IMPORTANT: Background process {session_id} completed "
                         f"(exit code {session.exit_code}).\n"
@@ -16156,13 +16168,16 @@ class GatewayRunner:
                 try:
                     from agent.title_generator import maybe_auto_title
                     all_msgs = result_holder[0].get("messages", []) if result_holder[0] else []
-                    # Route title-generation failures through the agent's
+                    # In Gateway mode, auto-title failures must NOT be
-                    # user-visible warning channel so a depleted auxiliary
+                    # surfaced as user-visible messages (fixes #23246).
-                    # provider doesn't silently leave sessions untitled
+                    # Log them at debug level only — they are not actionable
-                    # (issue #15775).
+                    # to the end user. CLI mode keeps the existing behaviour
-                    _title_failure_cb = getattr(
+                    # via the agent's _emit_auxiliary_failure path.
-                        agent, "_emit_auxiliary_failure", None
+                    def _title_failure_cb(task: str, exc: BaseException) -> None:
-                    )
+                        logger.debug(
+                            "Gateway auto-title failure suppressed (not user-visible): %s: %s",
+                            task, exc,
+                        )
                     maybe_auto_title_kwargs = {
                         "failure_callback": _title_failure_cb,
                         "main_runtime": {

hermes_cli/kanban.py

@@ -951,8 +951,13 @@ def _cmd_boards_create(args: argparse.Namespace) -> int:
 
 
 def _cmd_boards_rm(args: argparse.Namespace) -> int:
+    # When the user runs `hermes kanban boards delete <slug>` (alias), the
+    # boards_action is 'delete' but args.delete is never set to True because
+    # the --delete flag belongs to the 'rm' subparser only.  Detect the alias
+    # and treat it identically to `boards rm --delete` (fixes #23139).
+    force_delete = getattr(args, "delete", False) or getattr(args, "boards_action", "") == "delete"
     try:
-        res = kb.remove_board(args.slug, archive=not getattr(args, "delete", False))
+        res = kb.remove_board(args.slug, archive=not force_delete)
     except ValueError as exc:
         print(f"kanban boards rm: {exc}", file=sys.stderr)
         return 1