refactor(agent): leave the API-interrupt tail close to the finalizer

handle_api_interrupt exits via break into finalize_turn, whose
_close_transcript_tail already closes the tail with the same final_response, so
its own close was a duplicate; the scaffold strip stays (the partial-text row
must follow the tool row). The scaffolding drop no longer returns a flag nobody
reads, and the give-up test relies on the autouse backoff stub in
tests/agent/conftest.py instead of re-stubbing it.

agent/session_persistence.py

@@ -364,19 +364,16 @@ class SessionPersistenceMixin:
                 self._session_db.flush_token_counts()
             note_turn_persisted(self)
 
-    def _drop_trailing_empty_response_scaffolding(self, messages: List[Dict]) -> bool:
-        """Pop empty-response retry scaffolding from the tail; True when any was present. The
+    def _drop_trailing_empty_response_scaffolding(self, messages: List[Dict]) -> None:
+        """Pop empty-response retry scaffolding from the tail. The
         assistant(tool_calls) / tool rows before it stay: they were saved before the tools ran, so
         dropping them from the live history only makes the model repeat a side effect the durable
         transcript already records."""
         def tail(*keys: str) -> bool:
             return bool(messages) and isinstance(messages[-1], dict) and any(messages[-1].get(k) for k in keys)
 
-        dropped_scaffolding = False
         while tail("_empty_recovery_synthetic", "_empty_terminal_sentinel"):
             messages.pop()
-            dropped_scaffolding = True
-        return dropped_scaffolding
 
     _repair_message_sequence = _forward("agent.agent_runtime_helpers", "repair_message_sequence")
 

agent/turn_api_call.py

@@ -17,7 +17,6 @@ from typing import Any, Dict, Optional
 from agent.error_classifier import FailoverReason
 from agent.agent_runtime_helpers import _INTERRUPTED_PLACEHOLDER
 from agent.message_metadata import append_message
-from agent.message_sanitization import close_interrupted_tool_sequence
 from agent.repetition_guard import REPETITION_LOOP_INTERRUPTED, is_runaway_repetition
 from agent.turn_failure_copy import site_copy, stamp_failure
 
@@ -189,8 +188,8 @@ def handle_api_interrupt(
     agent._vprint(f"{agent.log_prefix}⚡ Interrupted during API call.", force=True)
     interrupted = True
     # A Stop during the empty-response nudge request leaves the synthetic assistant+nudge
-    # pair after an executed tool result; strip it so the row appended below (or the close)
-    # follows the tool row, and this exit owner records its own reason.
+    # pair after an executed tool result; strip it so the row appended below follows the tool
+    # row (the finalizer then closes the tail with this exit's own reason).
     agent._drop_trailing_empty_response_scaffolding(messages)
     _partial = agent._strip_think_blocks(
         getattr(agent, "_current_streamed_assistant_text", "") or ""
@@ -209,7 +208,6 @@ def handle_api_interrupt(
         final_response = _partial
     else:
         final_response = f"{INTERRUPT_WAITING_FOR_MODEL_PREFIX}{api_elapsed:.1f}s elapsed)."
-        close_interrupted_tool_sequence(messages, final_response)
     agent._persist_session(messages, conversation_history)
     return ApiInterruptVerdict("break", thinking_spinner, interrupted, final_response)
 

tests/agent/test_empty_response_recovery_persistence.py

@@ -197,9 +197,6 @@ def real_loop(tmp_path, monkeypatch):
     monkeypatch.setenv("no_proxy", "")
     monkeypatch.setattr("agent.title_generator.maybe_auto_title", lambda *a, **k: None)
     monkeypatch.setattr("agent.title_generator.start_title_upgrade", lambda *a, **k: None)
-    # The first empty response backs off 5-7.5 s for real; the give-up path is reached via the
-    # deterministic-empty guard, not the wait (``turn_empty_response`` imports this lazily).
-    monkeypatch.setattr("agent.retry_utils.jittered_backoff", lambda *a, **k: 0.0)
     monkeypatch.chdir(tmp_path)
     db = SessionDB(db_path=tmp_path / "state.db")
     sid = "sess-empty-exit"

tests/agent/test_message_sequence_repair.py

@@ -33,7 +33,7 @@ def test_drop_scaffolding_keeps_executed_tool_pair():
          "_empty_terminal_sentinel": True},
     ]
 
-    assert AIAgent._drop_trailing_empty_response_scaffolding(agent, messages) is True
+    AIAgent._drop_trailing_empty_response_scaffolding(agent, messages)
     assert messages == executed