feat(engine): finalize after tool iteration limit

app-instance/backend/beaver/engine/loop.py

@@ -708,8 +708,19 @@ class AgentLoop:
                     break
 
                 if iterations >= resolved_max_tool_iterations:
-                    final_text = response.content or "Tool loop stopped after reaching the configured iteration limit."
+                    finalized = await self._finalize_after_tool_limit(
-                    final_finish_reason = "max_tool_iterations"
+                        provider=provider,
+                        messages=messages,
+                        model=final_model,
+                        max_tokens=resolved_max_tokens,
+                        temperature=resolved_temperature,
+                        thinking_enabled=thinking_enabled,
+                    )
+                    final_text = finalized or (
+                        "Tool loop stopped after reaching the configured iteration limit, "
+                        "and no final answer was produced."
+                    )
+                    final_finish_reason = "max_tool_iterations_finalized" if finalized else "max_tool_iterations"
                     session_manager.append_message(
                         resolved_session_id,
                         run_id=resolved_run_id,
@@ -853,6 +864,39 @@ class AgentLoop:
             raise RuntimeError(f"Engine loader did not provide required dependency {field_name!r}")
         return value
 
+    @staticmethod
+    async def _finalize_after_tool_limit(
+        *,
+        provider: Any,
+        messages: list[dict[str, Any]],
+        model: str,
+        max_tokens: int,
+        temperature: float,
+        thinking_enabled: bool | None,
+    ) -> str:
+        final_messages = [
+            *messages,
+            {
+                "role": "system",
+                "content": (
+                    "The configured tool iteration budget is exhausted. Do not call tools. "
+                    "Produce the best final answer from the existing conversation and tool results. "
+                    "State uncertainty explicitly."
+                ),
+            },
+        ]
+        kwargs: dict[str, Any] = {
+            "messages": final_messages,
+            "tools": None,
+            "model": model,
+            "max_tokens": max_tokens,
+            "temperature": temperature,
+        }
+        if thinking_enabled is not None:
+            kwargs["thinking_enabled"] = thinking_enabled
+        response = await provider.chat(**kwargs)
+        return (response.content or "").strip()
+
     @staticmethod
     def _load_pinned_skill_contexts(skills_loader: Any, skill_names: list[str]) -> list[SkillContext]:
         contexts: list[SkillContext] = []

app-instance/backend/tests/unit/test_phase5_skills_runtime.py

@@ -608,6 +608,12 @@ def test_agent_loop_records_max_tool_iterations_as_failed_skill_effect(tmp_path:
                     provider_name="stub",
                     model="stub-model",
                 ),
+                LLMResponse(
+                    content="Based on the available tool result, the container likely failed during startup.",
+                    finish_reason="stop",
+                    provider_name="stub",
+                    model="stub-model",
+                ),
             ]
         ),
     )
@@ -621,7 +627,9 @@ def test_agent_loop_records_max_tool_iterations_as_failed_skill_effect(tmp_path:
     )
     loaded = loop.boot()
 
-    assert result.finish_reason == "max_tool_iterations"
+    assert result.finish_reason == "max_tool_iterations_finalized"
+    assert "Based on the available tool result" in result.output_text
+    assert "Tool loop stopped" not in result.output_text
     effect_records = loaded.run_memory_store.list_skill_effects("docker-debug", version="v0007")
     assert effect_records[-1].run_id == result.run_id
     assert effect_records[-1].success is False