diff --git a/backend/routes/test_runs.py b/backend/routes/test_runs.py
index 64caa3b..2dbc237 100644
--- a/backend/routes/test_runs.py
+++ b/backend/routes/test_runs.py
@@ -154,6 +154,14 @@ async def create_test_run(
             raise HTTPException(422, f"用例“{row.name}”校验失败: {exc}") from exc
         definitions_by_id[row.id] = definition
         has_image_input = has_image_input or bool(image_asset_ids(definition))
+        for simulation in definition.tool_simulations:
+            tool = tools_by_id.get(simulation.tool_id)
+            if tool is None or tool.function_name != simulation.function_name:
+                raise HTTPException(
+                    422,
+                    f"用例“{row.name}”引用的工具 {simulation.function_name} "
+                    "不属于当前被测助手",
+                )
         for turn in definition.turns:
             for behavior in turn.behaviors:
                 if behavior.type != "tool_call":
diff --git a/backend/services/config_resolver.py b/backend/services/config_resolver.py
index 219b2c9..43ebab2 100644
--- a/backend/services/config_resolver.py
+++ b/backend/services/config_resolver.py
@@ -22,7 +22,7 @@ from models import (
     RuntimeToolBinding,
 )
 from services.node_specs import copy_graph, graph_references
-from sqlalchemy import select
+from sqlalchemy import or_, select
 from sqlalchemy.ext.asyncio import AsyncSession
 
 
@@ -135,24 +135,56 @@ def _runtime_tool_ids(
     )
 
 
+def _implicit_startup_tool_names(startup: dict | None) -> list[str]:
+    """Return platform client tools used by built-in startup experiences.
+
+    A confirmation opening is configured directly on the assistant instead of
+    through an AssistantToolBinding, but text tests still need to simulate the
+    underlying client interaction explicitly.
+    """
+
+    value = startup or {}
+    opening_mode = value.get("opening_mode") or value.get("openingMode")
+    opening_message = value.get("opening_message") or value.get("openingMessage")
+    if opening_mode == "confirmation" and isinstance(opening_message, dict):
+        return ["show_message"]
+    return []
+
+
 async def _tools_for(
     session: AsyncSession,
     assistant: Assistant,
     tool_ids: list[str],
+    implicit_function_names: list[str] | None = None,
 ) -> list[RuntimeTool]:
-    if assistant.type not in {"prompt", "workflow"} or not tool_ids:
+    function_names = list(dict.fromkeys(implicit_function_names or []))
+    if assistant.type not in {"prompt", "workflow"} or not (
+        tool_ids or function_names
+    ):
         return []
+    identity_filters = []
+    if tool_ids:
+        identity_filters.append(Tool.id.in_(tool_ids))
+    if function_names:
+        identity_filters.append(Tool.function_name.in_(function_names))
     rows = (
         await session.execute(
             select(Tool)
             .where(
-                Tool.id.in_(tool_ids),
+                or_(*identity_filters),
                 Tool.status == "active",
             )
         )
     ).scalars().all()
     tool_by_id = {tool.id: tool for tool in rows}
+    tool_by_function = {tool.function_name: tool for tool in rows}
     tools = [tool_by_id[tool_id] for tool_id in tool_ids if tool_id in tool_by_id]
+    seen_tool_ids = {tool.id for tool in tools}
+    for function_name in function_names:
+        tool = tool_by_function.get(function_name)
+        if tool is not None and tool.id not in seen_tool_ids:
+            tools.append(tool)
+            seen_tool_ids.add(tool.id)
     server_ids = {
         str(tool.mcp_server_id)
         for tool in tools
@@ -268,7 +300,12 @@ async def resolve_runtime_config(
 
     prompt_tool_bindings = await _bound_tools(session, assistant.id)
     runtime_tool_ids = _runtime_tool_ids(prompt_tool_bindings, assistant.startup or {})
-    runtime_tools = await _tools_for(session, assistant, runtime_tool_ids)
+    runtime_tools = await _tools_for(
+        session,
+        assistant,
+        runtime_tool_ids,
+        _implicit_startup_tool_names(assistant.startup or {}),
+    )
 
     return AssistantConfig(
         name=assistant.name,
diff --git a/backend/services/test_runs/mock_tools.py b/backend/services/test_runs/mock_tools.py
index c20911d..b3dda60 100644
--- a/backend/services/test_runs/mock_tools.py
+++ b/backend/services/test_runs/mock_tools.py
@@ -1,4 +1,4 @@
-"""Test-only tool executor that never reaches external side effects."""
+"""Strict, deterministic tool simulations for text test runs."""
 
 from __future__ import annotations
 
@@ -9,25 +9,141 @@ from typing import Any
 from models import RuntimeTool, RuntimeToolBinding
 from services.runtime_variables import DynamicVariableStore
 from services.tool_executor import ToolExecutionError, ToolExecutor
-from test_schemas import ToolCallExpectedBehavior
+from test_schemas import ToolSimulation, ToolSimulationResponse, ToolSimulationStep
+
+
+TOOL_SIMULATION_ERROR_CODES = {
+    "UNMOCKED_TOOL_CALL",
+    "TOOL_SIMULATION_EXHAUSTED",
+    "TOOL_SIMULATION_SCOPE_MISMATCH",
+}
+
+
+class ToolSimulationError(ToolExecutionError):
+    """A non-retryable test configuration error raised at tool-call time."""
+
+    def __init__(self, code: str, message: str):
+        self.code = code
+        super().__init__(f"{code}: {message}")
+
+
+class ToolSimulationResolver:
+    """Resolve one case's fixed or ordered simulated tool responses."""
+
+    def __init__(self, simulations: list[ToolSimulation]):
+        self._by_tool_id = {item.tool_id: item for item in simulations}
+        self._by_function = {item.function_name: item for item in simulations}
+        self._positions = {item.tool_id: 0 for item in simulations}
+        self._phase = "startup"
+        self._turn_id: str | None = None
+        self._failure: ToolSimulationError | None = None
+
+    def enter_startup(self) -> None:
+        self._phase = "startup"
+        self._turn_id = None
+
+    def enter_turn(self, turn_id: str) -> None:
+        self._phase = "turn"
+        self._turn_id = turn_id
+
+    async def resolve(
+        self,
+        *,
+        tool_id: str | None,
+        function_name: str,
+    ) -> dict[str, Any]:
+        simulation = (
+            self._by_tool_id.get(tool_id or "")
+            or self._by_function.get(function_name)
+        )
+        if simulation is None:
+            raise self._record_failure(
+                "UNMOCKED_TOOL_CALL",
+                f"工具 {function_name} 未配置工具模拟",
+            )
+
+        response: ToolSimulationResponse | ToolSimulationStep
+        if simulation.mode == "fixed":
+            response = simulation.response
+        else:
+            position = self._positions[simulation.tool_id]
+            if position >= len(simulation.steps):
+                raise self._record_failure(
+                    "TOOL_SIMULATION_EXHAUSTED",
+                    f"工具 {function_name} 的模拟响应已耗尽（第 {position + 1} 次调用）",
+                )
+            step = simulation.steps[position]
+            if not self._scope_matches(step):
+                raise self._record_failure(
+                    "TOOL_SIMULATION_SCOPE_MISMATCH",
+                    (
+                        f"工具 {function_name} 的第 {position + 1} 条模拟步骤"
+                        f"期望发生在{self._scope_label(step)}，"
+                        f"实际发生在{self._current_scope_label()}"
+                    ),
+                )
+            response = step
+            self._positions[simulation.tool_id] = position + 1
+
+        if response.delay_ms:
+            await asyncio.sleep(response.delay_ms / 1000)
+        return self._normalized_result(response)
+
+    @property
+    def failure(self) -> ToolSimulationError | None:
+        return self._failure
+
+    def raise_if_failed(self) -> None:
+        """Re-raise errors swallowed by a higher-level client message stage."""
+
+        if self._failure is not None:
+            raise self._failure
+
+    def _record_failure(self, code: str, message: str) -> ToolSimulationError:
+        error = ToolSimulationError(code, message)
+        if self._failure is None:
+            self._failure = error
+        return error
+
+    def _scope_matches(self, step: ToolSimulationStep) -> bool:
+        if step.scope.phase == "any":
+            return True
+        if step.scope.phase == "startup":
+            return self._phase == "startup"
+        return self._phase == "turn" and step.scope.turn_id == self._turn_id
+
+    @staticmethod
+    def _scope_label(step: ToolSimulationStep) -> str:
+        if step.scope.phase == "any":
+            return "任意阶段"
+        if step.scope.phase == "startup":
+            return "启动阶段"
+        return f"轮次 {step.scope.turn_id}"
+
+    def _current_scope_label(self) -> str:
+        if self._phase == "startup":
+            return "启动阶段"
+        return f"轮次 {self._turn_id or 'unknown'}"
+
+    @staticmethod
+    def _normalized_result(response: ToolSimulationResponse) -> dict[str, Any]:
+        payload = json.loads(response.body)
+        status = "ok" if response.outcome == "success" else "error"
+        if isinstance(payload, dict):
+            return {**payload, "status": status}
+        return {"status": status, "data": payload}
 
 
 class MockToolExecutor(ToolExecutor):
-    """Resolve the active turn's configured mocks and reject everything else."""
+    """Execute external runtime tools exclusively through one test resolver."""
 
-    def __init__(self, store: DynamicVariableStore):
-        super().__init__(store)
-        self._mocks_by_tool_id: dict[str, ToolCallExpectedBehavior] = {}
-        self._mocks_by_function: dict[str, ToolCallExpectedBehavior] = {}
-        self.unmocked_calls: list[str] = []
-
-    def set_turn_behaviors(
+    def __init__(
         self,
-        behaviors: list[ToolCallExpectedBehavior],
-    ) -> None:
-        self._mocks_by_tool_id = {item.tool_id: item for item in behaviors}
-        self._mocks_by_function = {item.function_name: item for item in behaviors}
-        self.unmocked_calls = []
+        store: DynamicVariableStore,
+        resolver: ToolSimulationResolver,
+    ):
+        super().__init__(store)
+        self._resolver = resolver
 
     async def execute(
         self,
@@ -37,26 +153,10 @@ class MockToolExecutor(ToolExecutor):
         binding: RuntimeToolBinding | None = None,
         result_assignments: dict[str, str] | None = None,
     ) -> dict[str, Any]:
-        behavior = self._mocks_by_tool_id.get(tool.id) or self._mocks_by_function.get(
-            tool.function_name
+        result = await self._resolver.resolve(
+            tool_id=tool.id,
+            function_name=tool.function_name,
         )
-        if behavior is None:
-            self.unmocked_calls.append(tool.function_name)
-            raise ToolExecutionError(
-                f"UNMOCKED_TOOL_CALL: 工具 {tool.function_name} 未配置 Mock 返回值"
-            )
-
-        delay_seconds = behavior.mock_response.delay_ms / 1000
-        if delay_seconds:
-            await asyncio.sleep(delay_seconds)
-
-        payload = json.loads(behavior.mock_response.body)
-        status = "ok" if behavior.mock_response.outcome == "success" else "error"
-        if isinstance(payload, dict):
-            result: dict[str, Any] = dict(payload)
-            result["status"] = status
-        else:
-            result = {"status": status, "data": payload}
         if result.get("status") != "ok":
             return {**result, "updated_variables": []}
         return self._apply_result_assignments(
@@ -71,13 +171,16 @@ class MockToolExecutor(ToolExecutor):
 
 
 class TextClientToolPort:
-    """Automatically acknowledge built-in message stages in text tests."""
+    """Return configured client-tool results without a browser transport."""
 
-    def __init__(self) -> None:
+    def __init__(self, resolver: ToolSimulationResolver) -> None:
+        self._resolver = resolver
         self.auto_confirmed_messages: list[dict[str, Any]] = []
 
-    def supports(self, function_name: str) -> bool:
-        return function_name == "show_message"
+    def supports(self, _function_name: str) -> bool:
+        # Test runs advertise the virtual client, then fail strictly in call().
+        # Returning False would silently skip an unconfigured client tool.
+        return True
 
     async def call(
         self,
@@ -85,12 +188,31 @@ class TextClientToolPort:
         arguments: dict[str, Any],
         **_kwargs: Any,
     ) -> dict[str, Any]:
-        if function_name == "show_message":
-            self.auto_confirmed_messages.append(dict(arguments))
-            return {
-                "status": "ok",
-                "data": {"action": "confirmed", "arguments": arguments},
-            }
-        raise ToolExecutionError(
-            f"UNMOCKED_CLIENT_TOOL: 客户端工具 {function_name} 不能脱离 Mock 执行"
+        resolved = await self._resolver.resolve(
+            tool_id=None,
+            function_name=function_name,
         )
+        status = str(resolved.get("status") or "error")
+        payload = (
+            resolved.get("data")
+            if set(resolved).issubset({"status", "data"}) and "data" in resolved
+            else {key: value for key, value in resolved.items() if key != "status"}
+        )
+        # ClientToolPort mirrors the browser transport envelope. The JSON
+        # configured in the editor remains the client handler's root object.
+        result: dict[str, Any] = {"status": status, "data": payload}
+        if (
+            status != "ok"
+            and isinstance(payload, dict)
+            and payload.get("message") is not None
+        ):
+            result["message"] = str(payload["message"])
+        data = result.get("data")
+        action = data.get("action") if isinstance(data, dict) else None
+        if (
+            function_name == "show_message"
+            and result.get("status") == "ok"
+            and action == "confirmed"
+        ):
+            self.auto_confirmed_messages.append(dict(arguments))
+        return result
diff --git a/backend/services/test_runs/text_runner.py b/backend/services/test_runs/text_runner.py
index 45c232d..a7ef68e 100644
--- a/backend/services/test_runs/text_runner.py
+++ b/backend/services/test_runs/text_runner.py
@@ -21,7 +21,6 @@ from pipecat.frames.frames import (
     ErrorFrame,
     FunctionCallInProgressFrame,
     FunctionCallResultFrame,
-    LLMContextFrame,
     LLMFullResponseEndFrame,
     LLMFullResponseStartFrame,
     LLMMessagesAppendFrame,
@@ -54,7 +53,13 @@ from services.pipecat.processors import (
 from services.pipecat.service_factory import config_with_resource
 from services.pipecat.workflow_services import build_workflow_llm_switcher
 from services.test_runs.errors import TestExecutionError
-from services.test_runs.mock_tools import MockToolExecutor, TextClientToolPort
+from services.test_runs.mock_tools import (
+    TOOL_SIMULATION_ERROR_CODES,
+    MockToolExecutor,
+    TextClientToolPort,
+    ToolSimulationError,
+    ToolSimulationResolver,
+)
 from services.vision import build_image_user_message
 from test_schemas import ContextTurn, FixedInputTurn, TestCaseDefinition, TestImageRef
 
@@ -197,18 +202,21 @@ class _TextCaptureProcessor(FrameProcessor):
         """Release every wait even when Pipecat reports the error upstream."""
 
         message = frame.error or "Pipeline 执行失败"
-        unmocked_tool = "UNMOCKED_TOOL_CALL" in message
+        simulation_code = next(
+            (code for code in TOOL_SIMULATION_ERROR_CODES if code in message),
+            None,
+        )
         self.error = TestExecutionError(
-            code="UNMOCKED_TOOL_CALL" if unmocked_tool else "PIPELINE_ERROR",
+            code=simulation_code or "PIPELINE_ERROR",
             message=message,
             stage=(
                 "tool"
-                if unmocked_tool
+                if simulation_code
                 else "model"
                 if frame.processor
                 else "pipeline"
             ),
-            retryable=not unmocked_tool,
+            retryable=simulation_code is None,
         )
         self.window.pending_tools.clear()
         self.window.touch()
@@ -573,14 +581,12 @@ class TextPipelineRunner:
         async def on_pipeline_error(_worker, frame):
             capture.fail(frame)
 
-        executor_holder: dict[str, MockToolExecutor] = {}
-        text_client_tools = TextClientToolPort()
+        simulation_resolver = ToolSimulationResolver(definition.tool_simulations)
+        text_client_tools = TextClientToolPort(simulation_resolver)
         workflow_vision_scope: dict[str, Any] = {"enabled": False}
 
         def create_test_executor(store) -> MockToolExecutor:
-            executor = MockToolExecutor(store)
-            executor_holder["executor"] = executor
-            return executor
+            return MockToolExecutor(store, simulation_resolver)
 
         async def queue_call_end(reason: str) -> None:
             logger.debug(f"文本测试收到会话结束请求: {reason}")
@@ -738,8 +744,6 @@ class TextPipelineRunner:
                 tool_executor_factory=create_test_executor,
             ),
         )
-        executor = executor_holder["executor"]
-
         # A transportless text run cannot synthesize audio, but the configured
         # greeting has already completed before scripted inputs begin. Mirror
         # production's provider-safe context fact so the model does not repeat it.
@@ -772,25 +776,11 @@ class TextPipelineRunner:
                 # response is therefore the equivalent of playback ending.
                 await call_end.finish()
 
-        all_tool_behaviors = [
-            behavior
-            for turn in definition.turns
-            for behavior in turn.behaviors
-            if behavior.type == "tool_call"
-        ]
-        executor.set_turn_behaviors(all_tool_behaviors)
-
         runner_task: asyncio.Task[None] | None = None
         try:
             startup_runtime_record_marker = len(capture.runtime_records)
+            simulation_resolver.enter_startup()
             await brain.run_preflight()
-            if executor.unmocked_calls:
-                raise TestExecutionError(
-                    code="UNMOCKED_TOOL_CALL",
-                    message="启动阶段调用了未配置 Mock 的工具: "
-                    + "、".join(executor.unmocked_calls),
-                    stage="tool",
-                )
 
             await runner.add_workers(worker)
             runner_task = asyncio.create_task(
@@ -803,12 +793,14 @@ class TextPipelineRunner:
             marker = capture.window.activity_count
             await brain.on_connected(greeting_pending=False)
             await brain.on_client_ready()
+            simulation_resolver.raise_if_failed()
             await _wait_for_startup(
                 worker,
                 capture,
                 runner_task,
                 marker,
             )
+            simulation_resolver.raise_if_failed()
             startup_greeting = _resolve_startup_greeting(
                 greeting,
                 capture.opening_speech_outputs,
@@ -844,7 +836,7 @@ class TextPipelineRunner:
                     worker,
                     runner_task,
                     capture,
-                    executor,
+                    simulation_resolver,
                     turn,
                     index,
                     image_assets=image_assets,
@@ -873,9 +865,24 @@ class TextPipelineRunner:
             )
         except TestExecutionError:
             raise
+        except ToolSimulationError as exc:
+            raise TestExecutionError(
+                code=exc.code,
+                message=str(exc),
+                stage="tool",
+                retryable=False,
+            ) from exc
         except asyncio.CancelledError:
             raise
         except Exception as exc:
+            if simulation_resolver.failure is not None:
+                error = simulation_resolver.failure
+                raise TestExecutionError(
+                    code=error.code,
+                    message=str(error),
+                    stage="tool",
+                    retryable=False,
+                ) from exc
             raise TestExecutionError(
                 code="PIPELINE_EXECUTION_FAILED",
                 message=str(exc) or type(exc).__name__,
@@ -894,17 +901,14 @@ class TextPipelineRunner:
         worker: PipelineWorker,
         runner_task: asyncio.Task[None],
         capture: _TextCaptureProcessor,
-        executor: MockToolExecutor,
+        simulation_resolver: ToolSimulationResolver,
         turn: FixedInputTurn,
         index: int,
         *,
         image_assets: dict[str, bytes],
         workflow_vision_scope: dict[str, Any],
     ) -> RawTurnResult:
-        tool_behaviors = [
-            behavior for behavior in turn.behaviors if behavior.type == "tool_call"
-        ]
-        executor.set_turn_behaviors(tool_behaviors)
+        simulation_resolver.enter_turn(turn.id)
         activity_marker = capture.window.activity_count
         output_marker = len(capture.outputs)
         tool_marker = len(capture.tool_calls)
@@ -942,13 +946,7 @@ class TextPipelineRunner:
             LLMMessagesAppendFrame(messages=[user_message], run_llm=True)
         )
         await _wait_for_quiescence(capture, runner_task, activity_marker)
-        if executor.unmocked_calls:
-            raise TestExecutionError(
-                code="UNMOCKED_TOOL_CALL",
-                message="本轮调用了未配置 Mock 的工具: "
-                + "、".join(executor.unmocked_calls),
-                stage="tool",
-            )
+        simulation_resolver.raise_if_failed()
         return RawTurnResult(
             id=turn.id,
             index=index,
diff --git a/backend/test_schemas.py b/backend/test_schemas.py
index 1fb1c77..0c5e8b3 100644
--- a/backend/test_schemas.py
+++ b/backend/test_schemas.py
@@ -77,7 +77,7 @@ class ToolParamAssertion(CamelModel):
         return self
 
 
-class ToolMockResponse(CamelModel):
+class ToolSimulationResponse(CamelModel):
     outcome: Literal["success", "error"] = "success"
     body: str = Field(default='{"status":"ok"}', max_length=100_000)
     delay_ms: int = Field(default=0, ge=0, le=60_000)
@@ -93,6 +93,63 @@ class ToolMockResponse(CamelModel):
         return self
 
 
+class AnyToolSimulationScope(CamelModel):
+    phase: Literal["any"] = "any"
+
+
+class StartupToolSimulationScope(CamelModel):
+    phase: Literal["startup"] = "startup"
+
+
+class TurnToolSimulationScope(CamelModel):
+    phase: Literal["turn"] = "turn"
+    turn_id: str = Field(min_length=1, max_length=128)
+
+
+ToolSimulationScope = Annotated[
+    Union[
+        AnyToolSimulationScope,
+        StartupToolSimulationScope,
+        TurnToolSimulationScope,
+    ],
+    Field(discriminator="phase"),
+]
+
+
+class ToolSimulationStep(ToolSimulationResponse):
+    id: str = Field(min_length=1, max_length=128)
+    scope: ToolSimulationScope = Field(default_factory=AnyToolSimulationScope)
+
+
+class ToolSimulationBase(CamelModel):
+    tool_id: str = Field(min_length=1, max_length=40)
+    function_name: str = Field(min_length=1, max_length=128)
+
+    @field_validator("tool_id", "function_name")
+    @classmethod
+    def strip_tool_identity(cls, value: str) -> str:
+        value = value.strip()
+        if not value:
+            raise ValueError("必须选择有效的工具")
+        return value
+
+
+class FixedToolSimulation(ToolSimulationBase):
+    mode: Literal["fixed"] = "fixed"
+    response: ToolSimulationResponse = Field(default_factory=ToolSimulationResponse)
+
+
+class SequenceToolSimulation(ToolSimulationBase):
+    mode: Literal["sequence"] = "sequence"
+    steps: list[ToolSimulationStep] = Field(min_length=1, max_length=10_000)
+
+
+ToolSimulation = Annotated[
+    Union[FixedToolSimulation, SequenceToolSimulation],
+    Field(discriminator="mode"),
+]
+
+
 class ReplyExpectedBehavior(CamelModel):
     id: str = Field(min_length=1, max_length=128)
     type: Literal["reply"] = "reply"
@@ -122,7 +179,6 @@ class ToolCallExpectedBehavior(CamelModel):
     min_calls: int = Field(default=1, ge=0, le=100)
     max_calls: int | None = Field(default=None, ge=0, le=100)
     param_assertions: list[ToolParamAssertion] = Field(default_factory=list, max_length=50)
-    mock_response: ToolMockResponse = Field(default_factory=ToolMockResponse)
 
     @model_validator(mode="after")
     def validate_call_expectation(self):
@@ -191,14 +247,176 @@ class OverallCriterion(CamelModel):
 class TestCaseDefinition(CamelModel):
     context_turns: list[ContextTurn] = Field(default_factory=list, max_length=100)
     turns: list[FixedInputTurn] = Field(min_length=1, max_length=100)
+    tool_simulations: list[ToolSimulation] = Field(default_factory=list, max_length=100)
     overall_criteria: list[OverallCriterion] = Field(default_factory=list, max_length=50)
 
+    @model_validator(mode="before")
+    @classmethod
+    def migrate_legacy_tool_mocks(cls, raw):
+        """Normalize the former per-assertion mockResponse contract on read."""
+
+        if not isinstance(raw, dict):
+            return raw
+        if "tool_simulations" in raw or "toolSimulations" in raw:
+            return raw
+
+        groups: dict[tuple[str, str], list[dict[str, Any]]] = {}
+        for turn_index, turn in enumerate(raw.get("turns") or []):
+            if not isinstance(turn, dict):
+                continue
+            turn_id = str(turn.get("id") or "").strip()
+            for behavior_index, behavior in enumerate(turn.get("behaviors") or []):
+                if not isinstance(behavior, dict):
+                    continue
+                if behavior.get("type") != "tool_call":
+                    continue
+                if behavior.get("expectation") != "called":
+                    continue
+                mock = behavior.get("mock_response") or behavior.get("mockResponse")
+                if not isinstance(mock, dict):
+                    continue
+                tool_id = str(
+                    behavior.get("tool_id") or behavior.get("toolId") or ""
+                ).strip()
+                function_name = str(
+                    behavior.get("function_name")
+                    or behavior.get("functionName")
+                    or ""
+                ).strip()
+                if not tool_id or not function_name:
+                    continue
+                response = {
+                    "outcome": str(mock.get("outcome") or "success"),
+                    "body": str(mock.get("body") or '{"status":"ok"}'),
+                    "delayMs": int(
+                        mock.get("delay_ms")
+                        if mock.get("delay_ms") is not None
+                        else mock.get("delayMs") or 0
+                    ),
+                }
+                groups.setdefault((tool_id, function_name), []).append(
+                    {
+                        "response": response,
+                        "turnId": turn_id,
+                        "turnIndex": turn_index,
+                        "behaviorIndex": behavior_index,
+                        "minCalls": max(
+                            1,
+                            int(
+                                behavior.get("min_calls")
+                                if behavior.get("min_calls") is not None
+                                else behavior.get("minCalls") or 1
+                            ),
+                        ),
+                    }
+                )
+
+        migrated: list[dict[str, Any]] = []
+        for (tool_id, function_name), entries in groups.items():
+            response_signatures = {
+                json.dumps(entry["response"], ensure_ascii=False, sort_keys=True)
+                for entry in entries
+            }
+            if len(response_signatures) == 1:
+                migrated.append(
+                    {
+                        "toolId": tool_id,
+                        "functionName": function_name,
+                        "mode": "fixed",
+                        "response": entries[0]["response"],
+                    }
+                )
+                continue
+
+            steps: list[dict[str, Any]] = []
+            for entry in entries:
+                for repeat_index in range(entry["minCalls"]):
+                    steps.append(
+                        {
+                            "id": (
+                                f"legacy_step_{entry['turnIndex'] + 1}_"
+                                f"{entry['behaviorIndex'] + 1}_{repeat_index + 1}"
+                            ),
+                            "scope": {
+                                "phase": "turn",
+                                "turnId": entry["turnId"],
+                            },
+                            **entry["response"],
+                        }
+                    )
+            migrated.append(
+                {
+                    "toolId": tool_id,
+                    "functionName": function_name,
+                    "mode": "sequence",
+                    "steps": steps,
+                }
+            )
+
+        normalized = dict(raw)
+        normalized["toolSimulations"] = migrated
+        return normalized
+
     @model_validator(mode="after")
     def validate_has_expectation(self):
         if not any(turn.behaviors for turn in self.turns) and not self.overall_criteria:
             raise ValueError("至少配置一项有效的预期行为或整体评估标准")
         return self
 
+    @model_validator(mode="after")
+    def validate_tool_simulations(self):
+        tool_ids: set[str] = set()
+        step_ids: set[str] = set()
+        turn_ids = {turn.id for turn in self.turns}
+        simulations_by_tool: dict[str, ToolSimulation] = {}
+
+        for simulation in self.tool_simulations:
+            if simulation.tool_id in tool_ids:
+                raise ValueError(
+                    f"工具 {simulation.function_name} 只能配置一条工具模拟"
+                )
+            tool_ids.add(simulation.tool_id)
+            simulations_by_tool[simulation.tool_id] = simulation
+            if simulation.mode != "sequence":
+                continue
+            for step in simulation.steps:
+                if step.id in step_ids:
+                    raise ValueError(f"工具模拟步骤 ID 重复: {step.id}")
+                step_ids.add(step.id)
+                if step.scope.phase == "turn" and step.scope.turn_id not in turn_ids:
+                    raise ValueError(
+                        f"工具模拟步骤引用了不存在的轮次: {step.scope.turn_id}"
+                    )
+
+        for turn in self.turns:
+            for behavior in turn.behaviors:
+                if behavior.type != "tool_call" or behavior.expectation != "called":
+                    continue
+                simulation = simulations_by_tool.get(behavior.tool_id)
+                if simulation is None:
+                    raise ValueError(
+                        f"工具 {behavior.function_name} 配置了应调用断言，但缺少工具模拟"
+                    )
+                if simulation.function_name != behavior.function_name:
+                    raise ValueError(
+                        f"工具 {behavior.function_name} 的断言与模拟不匹配"
+                    )
+                if simulation.mode == "fixed":
+                    continue
+                covers_turn = any(
+                    step.scope.phase == "any"
+                    or (
+                        step.scope.phase == "turn"
+                        and step.scope.turn_id == turn.id
+                    )
+                    for step in simulation.steps
+                )
+                if not covers_turn:
+                    raise ValueError(
+                        f"工具 {behavior.function_name} 缺少覆盖本轮的模拟步骤"
+                    )
+        return self
+
     @model_validator(mode="after")
     def validate_tool_context_pairs(self):
         pending: list[tuple[str | None, str]] = []
@@ -275,6 +493,7 @@ class TestCaseWrite(CamelModel):
     input_mode: TestCaseInputMode = "fixed_script_text"
     context_turns: list[ContextTurn] = Field(default_factory=list, max_length=100)
     turns: list[FixedInputTurn] = Field(min_length=1, max_length=100)
+    tool_simulations: list[ToolSimulation] = Field(default_factory=list, max_length=100)
     overall_criteria: list[OverallCriterion] = Field(default_factory=list, max_length=50)
 
     @field_validator("name")
@@ -292,6 +511,7 @@ class TestCaseWrite(CamelModel):
         TestCaseDefinition(
             context_turns=self.context_turns,
             turns=self.turns,
+            tool_simulations=self.tool_simulations,
             overall_criteria=self.overall_criteria,
         )
         return self
@@ -300,6 +520,7 @@ class TestCaseWrite(CamelModel):
         return TestCaseDefinition(
             context_turns=self.context_turns,
             turns=self.turns,
+            tool_simulations=self.tool_simulations,
             overall_criteria=self.overall_criteria,
         )
 
diff --git a/backend/tests/test_debug_auto_test.py b/backend/tests/test_debug_auto_test.py
index 45a3fb1..4121990 100644
--- a/backend/tests/test_debug_auto_test.py
+++ b/backend/tests/test_debug_auto_test.py
@@ -7,7 +7,11 @@ from pipecat.processors.aggregators.llm_context import LLMContext
 from services.brains import BrainRuntime, build_brain
 from services.brains.base import GREETING_CONTEXT_MARKER
 from services.runtime_variables import DynamicVariableStore, prepare_dynamic_config
-from services.test_runs.mock_tools import MockToolExecutor, TextClientToolPort
+from services.test_runs.mock_tools import (
+    MockToolExecutor,
+    TextClientToolPort,
+    ToolSimulationResolver,
+)
 from services.test_runs.text_runner import (
     _TextCaptureProcessor,
     _opening_speech_from_message,
@@ -16,7 +20,7 @@ from services.test_runs.text_runner import (
     _wait_for_startup,
     _wait_for_quiescence,
 )
-from test_schemas import BatchRunCreate, BatchStartupResult
+from test_schemas import BatchRunCreate, BatchStartupResult, FixedToolSimulation
 
 
 class FakeLLM:
@@ -154,8 +158,24 @@ class DebugAutoTestContractTests(unittest.TestCase):
             "message_started",
         )
 
-    def test_text_client_port_records_automatic_message_confirmation(self):
-        port = TextClientToolPort()
+    def test_text_client_port_records_configured_message_confirmation(self):
+        resolver = ToolSimulationResolver(
+            [
+                FixedToolSimulation.model_validate(
+                    {
+                        "toolId": "tool_message",
+                        "functionName": "show_message",
+                        "mode": "fixed",
+                        "response": {
+                            "outcome": "success",
+                            "body": '{"action":"confirmed"}',
+                            "delayMs": 0,
+                        },
+                    }
+                )
+            ]
+        )
+        port = TextClientToolPort(resolver)
 
         result = asyncio.run(
             port.call(
@@ -230,12 +250,9 @@ class DebugAutoTestRuntimeTests(unittest.IsolatedAsyncioTestCase):
         brain = build_brain(cfg)
         llm = FakeLLM()
         call_end = FakeCallEnd()
-        executor_holder = {}
 
         def create_executor(store):
-            executor = MockToolExecutor(store)
-            executor_holder["executor"] = executor
-            return executor
+            return MockToolExecutor(store, ToolSimulationResolver([]))
 
         await brain.setup(
             cfg,
@@ -256,7 +273,6 @@ class DebugAutoTestRuntimeTests(unittest.IsolatedAsyncioTestCase):
         self.assertTrue(call_end.ending)
         self.assertEqual(call_end.reason, "测试完成")
         self.assertEqual(params.result["action"], "ending_call")
-        self.assertEqual(executor_holder["executor"].unmocked_calls, [])
 
     async def test_text_capture_keeps_runtime_records_for_auto_test_ui(self):
         event = {
diff --git a/backend/tests/test_startup_actions.py b/backend/tests/test_startup_actions.py
index 3ccfcd4..95fb80f 100644
--- a/backend/tests/test_startup_actions.py
+++ b/backend/tests/test_startup_actions.py
@@ -7,7 +7,11 @@ from fastapi import HTTPException
 from routes.assistants import _validate_startup_actions
 from models import RuntimeToolBinding
 from schemas import AssistantUpsert
-from services.config_resolver import _runtime_tool_ids
+from services.config_resolver import (
+    _implicit_startup_tool_names,
+    _runtime_tool_ids,
+    _tools_for,
+)
 
 
 def startup_body(
@@ -57,6 +61,25 @@ class FakeSession:
         return self.tool if self.tool is not None and tool_id == self.tool.id else None
 
 
+class FakeToolQueryResult:
+    def __init__(self, tools):
+        self.tools = tools
+
+    def scalars(self):
+        return self
+
+    def all(self):
+        return self.tools
+
+
+class FakeToolCatalogSession:
+    def __init__(self, tools):
+        self.tools = tools
+
+    async def execute(self, _statement):
+        return FakeToolQueryResult(self.tools)
+
+
 class StartupActionValidationTests(unittest.IsolatedAsyncioTestCase):
     def test_runtime_pool_includes_bound_and_startup_only_tools(self):
         self.assertEqual(
@@ -75,6 +98,51 @@ class StartupActionValidationTests(unittest.IsolatedAsyncioTestCase):
             ["conversation_tool", "shared_tool", "opening_tool"],
         )
 
+    def test_confirmation_opening_declares_implicit_show_message_tool(self):
+        self.assertEqual(
+            _implicit_startup_tool_names(
+                {
+                    "opening_mode": "confirmation",
+                    "opening_message": {
+                        "title": "重要提示",
+                        "message": "请确认已阅读。",
+                    },
+                }
+            ),
+            ["show_message"],
+        )
+        self.assertEqual(
+            _implicit_startup_tool_names(
+                {
+                    "openingMode": "playback",
+                    "openingMessage": {"message": "无需确认"},
+                }
+            ),
+            [],
+        )
+
+    async def test_implicit_show_message_enters_runtime_pool_without_binding(self):
+        tool = SimpleNamespace(
+            id="tool_show_message_default",
+            name="显示确认消息",
+            function_name="show_message",
+            type="client",
+            description="显示确认弹窗",
+            definition={"config": {}},
+            secrets={},
+            mcp_server_id=None,
+        )
+
+        runtime_tools = await _tools_for(
+            FakeToolCatalogSession([tool]),
+            SimpleNamespace(type="prompt"),
+            [],
+            ["show_message"],
+        )
+
+        self.assertEqual([item.id for item in runtime_tools], [tool.id])
+        self.assertEqual(runtime_tools[0].function_name, "show_message")
+
     def test_realtime_rejects_startup_actions(self):
         with self.assertRaisesRegex(ValueError, "Realtime"):
             AssistantUpsert(
diff --git a/backend/tests/test_tool_simulations.py b/backend/tests/test_tool_simulations.py
new file mode 100644
index 0000000..2a2f6c7
--- /dev/null
+++ b/backend/tests/test_tool_simulations.py
@@ -0,0 +1,312 @@
+import unittest
+
+from pydantic import ValidationError
+
+from services.message_policy import MESSAGE_CONFIRMATION
+from services.message_stage import (
+    MessageDisplaySpec,
+    MessageStageRunner,
+    MessageStageSpec,
+)
+from services.test_runs.mock_tools import (
+    TextClientToolPort,
+    ToolSimulationError,
+    ToolSimulationResolver,
+)
+from test_schemas import TestCaseDefinition
+
+
+def definition_with_simulation(simulation: dict) -> TestCaseDefinition:
+    return TestCaseDefinition.model_validate(
+        {
+            "turns": [
+                {
+                    "id": "turn_1",
+                    "userInput": "查询订单",
+                    "behaviors": [
+                        {
+                            "id": "behavior_1",
+                            "type": "tool_call",
+                            "toolId": "tool_order",
+                            "functionName": "query_order",
+                            "expectation": "called",
+                            "minCalls": 1,
+                            "maxCalls": 1,
+                            "paramAssertions": [],
+                        }
+                    ],
+                }
+            ],
+            "toolSimulations": [simulation],
+        }
+    )
+
+
+class ToolSimulationSchemaTests(unittest.TestCase):
+    def test_called_assertion_requires_simulation(self):
+        with self.assertRaisesRegex(ValidationError, "缺少工具模拟"):
+            TestCaseDefinition.model_validate(
+                {
+                    "turns": [
+                        {
+                            "id": "turn_1",
+                            "userInput": "查询订单",
+                            "behaviors": [
+                                {
+                                    "id": "behavior_1",
+                                    "type": "tool_call",
+                                    "toolId": "tool_order",
+                                    "functionName": "query_order",
+                                    "expectation": "called",
+                                    "minCalls": 1,
+                                    "paramAssertions": [],
+                                }
+                            ],
+                        }
+                    ]
+                }
+            )
+
+    def test_sequence_turn_scope_must_reference_existing_turn(self):
+        with self.assertRaisesRegex(ValidationError, "不存在的轮次"):
+            definition_with_simulation(
+                {
+                    "toolId": "tool_order",
+                    "functionName": "query_order",
+                    "mode": "sequence",
+                    "steps": [
+                        {
+                            "id": "step_1",
+                            "scope": {"phase": "turn", "turnId": "missing"},
+                            "outcome": "success",
+                            "body": "{}",
+                            "delayMs": 0,
+                        }
+                    ],
+                }
+            )
+
+    def test_legacy_equal_responses_become_fixed_simulation(self):
+        definition = TestCaseDefinition.model_validate(
+            {
+                "turns": [
+                    {
+                        "id": "turn_1",
+                        "userInput": "第一次查询",
+                        "behaviors": [
+                            {
+                                "id": "behavior_1",
+                                "type": "tool_call",
+                                "toolId": "tool_order",
+                                "functionName": "query_order",
+                                "expectation": "called",
+                                "minCalls": 1,
+                                "paramAssertions": [],
+                                "mockResponse": {
+                                    "outcome": "success",
+                                    "body": '{"orderId":"A1"}',
+                                    "delayMs": 0,
+                                },
+                            }
+                        ],
+                    },
+                    {
+                        "id": "turn_2",
+                        "userInput": "再次查询",
+                        "behaviors": [
+                            {
+                                "id": "behavior_2",
+                                "type": "tool_call",
+                                "toolId": "tool_order",
+                                "functionName": "query_order",
+                                "expectation": "called",
+                                "minCalls": 1,
+                                "paramAssertions": [],
+                                "mockResponse": {
+                                    "outcome": "success",
+                                    "body": '{"orderId":"A1"}',
+                                    "delayMs": 0,
+                                },
+                            }
+                        ],
+                    },
+                ]
+            }
+        )
+
+        self.assertEqual(definition.tool_simulations[0].mode, "fixed")
+        dumped = definition.model_dump(mode="json", by_alias=True)
+        self.assertNotIn("mockResponse", dumped["turns"][0]["behaviors"][0])
+
+    def test_legacy_different_responses_become_scoped_sequence(self):
+        definition = TestCaseDefinition.model_validate(
+            {
+                "turns": [
+                    {
+                        "id": "turn_1",
+                        "userInput": "第一次查询",
+                        "behaviors": [
+                            {
+                                "id": "behavior_1",
+                                "type": "tool_call",
+                                "toolId": "tool_order",
+                                "functionName": "query_order",
+                                "expectation": "called",
+                                "minCalls": 2,
+                                "paramAssertions": [],
+                                "mockResponse": {
+                                    "outcome": "success",
+                                    "body": '{"orderId":"A1"}',
+                                    "delayMs": 0,
+                                },
+                            }
+                        ],
+                    },
+                    {
+                        "id": "turn_2",
+                        "userInput": "再次查询",
+                        "behaviors": [
+                            {
+                                "id": "behavior_2",
+                                "type": "tool_call",
+                                "toolId": "tool_order",
+                                "functionName": "query_order",
+                                "expectation": "called",
+                                "minCalls": 1,
+                                "paramAssertions": [],
+                                "mockResponse": {
+                                    "outcome": "error",
+                                    "body": '{"message":"not found"}',
+                                    "delayMs": 0,
+                                },
+                            }
+                        ],
+                    },
+                ]
+            }
+        )
+
+        simulation = definition.tool_simulations[0]
+        self.assertEqual(simulation.mode, "sequence")
+        self.assertEqual(len(simulation.steps), 3)
+        self.assertEqual(simulation.steps[0].scope.turn_id, "turn_1")
+        self.assertEqual(simulation.steps[2].scope.turn_id, "turn_2")
+
+
+class ToolSimulationResolverTests(unittest.IsolatedAsyncioTestCase):
+    async def test_client_message_stage_keeps_unmocked_failure_for_runner(self):
+        resolver = ToolSimulationResolver([])
+        stage = MessageStageRunner(TextClientToolPort(resolver))
+
+        result = await stage.run(
+            MessageStageSpec(
+                display=MessageDisplaySpec(title="确认", message="请确认"),
+                completion_policy=MESSAGE_CONFIRMATION,
+            )
+        )
+
+        self.assertFalse(result.succeeded)
+        self.assertIn("UNMOCKED_TOOL_CALL", result.error or "")
+        with self.assertRaisesRegex(ToolSimulationError, "UNMOCKED_TOOL_CALL"):
+            resolver.raise_if_failed()
+
+    async def test_fixed_response_can_be_reused(self):
+        definition = definition_with_simulation(
+            {
+                "toolId": "tool_order",
+                "functionName": "query_order",
+                "mode": "fixed",
+                "response": {
+                    "outcome": "success",
+                    "body": '{"orderId":"A1"}',
+                    "delayMs": 0,
+                },
+            }
+        )
+        resolver = ToolSimulationResolver(definition.tool_simulations)
+        resolver.enter_turn("turn_1")
+
+        first = await resolver.resolve(
+            tool_id="tool_order", function_name="query_order"
+        )
+        second = await resolver.resolve(
+            tool_id="tool_order", function_name="query_order"
+        )
+
+        self.assertEqual(first, {"orderId": "A1", "status": "ok"})
+        self.assertEqual(second, first)
+
+    async def test_sequence_consumes_client_choices_then_exhausts(self):
+        definition = definition_with_simulation(
+            {
+                "toolId": "tool_order",
+                "functionName": "query_order",
+                "mode": "sequence",
+                "steps": [
+                    {
+                        "id": "step_1",
+                        "scope": {"phase": "turn", "turnId": "turn_1"},
+                        "outcome": "success",
+                        "body": '{"action":"selected","value":"basic"}',
+                        "delayMs": 0,
+                    },
+                    {
+                        "id": "step_2",
+                        "scope": {"phase": "any"},
+                        "outcome": "success",
+                        "body": '{"action":"dismissed"}',
+                        "delayMs": 0,
+                    },
+                ],
+            }
+        )
+        resolver = ToolSimulationResolver(definition.tool_simulations)
+        resolver.enter_turn("turn_1")
+        port = TextClientToolPort(resolver)
+
+        selected = await port.call("query_order", {})
+        dismissed = await port.call("query_order", {})
+
+        self.assertEqual(selected["data"]["value"], "basic")
+        self.assertEqual(dismissed["data"]["action"], "dismissed")
+        with self.assertRaisesRegex(
+            ToolSimulationError, "TOOL_SIMULATION_EXHAUSTED"
+        ):
+            await port.call("query_order", {})
+
+    async def test_missing_and_scope_mismatch_are_distinct(self):
+        definition = definition_with_simulation(
+            {
+                "toolId": "tool_order",
+                "functionName": "query_order",
+                "mode": "sequence",
+                "steps": [
+                    {
+                        "id": "step_1",
+                        "scope": {"phase": "turn", "turnId": "turn_1"},
+                        "outcome": "success",
+                        "body": "{}",
+                        "delayMs": 0,
+                    }
+                ],
+            }
+        )
+        resolver = ToolSimulationResolver(definition.tool_simulations)
+        resolver.enter_startup()
+
+        with self.assertRaisesRegex(
+            ToolSimulationError, "TOOL_SIMULATION_SCOPE_MISMATCH"
+        ):
+            await resolver.resolve(
+                tool_id="tool_order", function_name="query_order"
+            )
+        with self.assertRaisesRegex(
+            ToolSimulationError, "TOOL_SIMULATION_SCOPE_MISMATCH"
+        ):
+            resolver.raise_if_failed()
+        with self.assertRaisesRegex(ToolSimulationError, "UNMOCKED_TOOL_CALL"):
+            await resolver.resolve(tool_id="missing", function_name="missing")
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/frontend/src/components/assistant-editor/debug-auto-test.tsx b/frontend/src/components/assistant-editor/debug-auto-test.tsx
index c769528..6ddeb69 100644
--- a/frontend/src/components/assistant-editor/debug-auto-test.tsx
+++ b/frontend/src/components/assistant-editor/debug-auto-test.tsx
@@ -36,11 +36,12 @@ import {
 } from "@/components/ui/popover";
 import { SearchInput } from "@/components/ui/search-input";
 import { Tabs, TabsList, TabsTrigger } from "@/components/ui/tabs";
-import type {
-  BatchEvaluationResult,
-  BatchRunCase,
-  BatchRunSnapshot,
-  BatchStartupResult,
+import {
+  testExecutionErrorLabel,
+  type BatchEvaluationResult,
+  type BatchRunCase,
+  type BatchRunSnapshot,
+  type BatchStartupResult,
 } from "@/data/batch-run";
 import {
   getTestCaseValidationMessage,
@@ -422,7 +423,7 @@ export function AutoTestLivePanel({
   ) {
     return (
       <ErrorPanel
-        title={state.runCase.executionError.code}
+        title={testExecutionErrorLabel(state.runCase.executionError.code)}
         description={state.runCase.executionError.message}
       />
     );
@@ -504,7 +505,7 @@ export function AutoTestLivePanel({
         {state.phase === "error" && state.runCase?.executionError && (
           <div className="rounded-xl border border-amber-500/25 bg-amber-500/[0.06] px-3 py-2.5 text-xs leading-5 text-amber-600 dark:text-amber-400">
             <span className="font-medium">
-              {state.runCase.executionError.code}
+              {testExecutionErrorLabel(state.runCase.executionError.code)}
             </span>
             <span> · {state.runCase.executionError.message}</span>
           </div>
diff --git a/frontend/src/components/batch-test/batch-case-detail-drawer.tsx b/frontend/src/components/batch-test/batch-case-detail-drawer.tsx
index ddd3ed5..3db8304 100644
--- a/frontend/src/components/batch-test/batch-case-detail-drawer.tsx
+++ b/frontend/src/components/batch-test/batch-case-detail-drawer.tsx
@@ -23,6 +23,7 @@ import { BatchCaseStatusBadge } from "@/components/batch-test/batch-case-status"
 import { Button } from "@/components/ui/button";
 import {
   firstEvaluationFailureReason,
+  testExecutionErrorLabel,
   type BatchEvaluationResult,
   type BatchRunCase,
   type BatchToolCallRecord,
@@ -115,7 +116,7 @@ export function BatchCaseDetailDrawer({
               <div className="mt-2.5 rounded-2xl border border-amber-500/25 bg-amber-500/[0.06] px-4 py-3.5">
                 <div className="flex flex-wrap items-center gap-2">
                   <code className="text-xs font-medium text-foreground">
-                    {item.executionError.code}
+                    {testExecutionErrorLabel(item.executionError.code)}
                   </code>
                   <span className="rounded-full border border-amber-500/25 px-2 py-0.5 text-[10px] text-amber-600 dark:text-amber-400">
                     {item.executionError.stage}
diff --git a/frontend/src/components/pages/TestCasesPage.tsx b/frontend/src/components/pages/TestCasesPage.tsx
index 5fe9b36..cf57b67 100644
--- a/frontend/src/components/pages/TestCasesPage.tsx
+++ b/frontend/src/components/pages/TestCasesPage.tsx
@@ -56,6 +56,7 @@ import {
   DEFAULT_INPUT_MODE,
   getTestCaseValidationMessage,
   normalizeOverallCriteria,
+  prepareFixedToolSimulations,
   TEST_CASE_INPUT_MODE_LABEL,
   TEST_CASE_INPUT_MODE_SHORT_LABEL,
   type TestCase,
@@ -420,6 +421,7 @@ function caseToDraft(item: TestCase): CaseEditorDraft {
     inputMode: item.inputMode,
     contextTurns: item.contextTurns.map((turn) => ({ ...turn })),
     turns: cloneTurns(item.turns),
+    toolSimulations: prepareFixedToolSimulations(item.toolSimulations),
     overallCriteria: cloneOverallCriteria(item.overallCriteria),
   };
 }
@@ -434,6 +436,7 @@ function createEmptyCaseDraft(): CaseEditorDraft {
     inputMode: DEFAULT_INPUT_MODE,
     contextTurns: [],
     turns: [createEmptyFixedInputTurn()],
+    toolSimulations: [],
     overallCriteria: [],
   };
 }
@@ -625,6 +628,7 @@ function SuiteDetailView({
       inputMode: draft.inputMode,
       contextTurns: draft.contextTurns,
       turns: draft.turns,
+      toolSimulations: draft.toolSimulations,
       overallCriteria: normalizeOverallCriteria(draft.overallCriteria),
     };
 
diff --git a/frontend/src/components/test-cases/next-reply-editor.tsx b/frontend/src/components/test-cases/next-reply-editor.tsx
index 2585f60..9188f10 100644
--- a/frontend/src/components/test-cases/next-reply-editor.tsx
+++ b/frontend/src/components/test-cases/next-reply-editor.tsx
@@ -55,6 +55,10 @@ import {
   SelectValue,
 } from "@/components/ui/select";
 import { Textarea } from "@/components/ui/textarea";
+import {
+  ToolSimulationEditor,
+  type EditorToolOption,
+} from "@/components/test-cases/tool-simulation-editor";
 import {
   ASSERTION_TYPE_LABEL,
   KEYWORD_MATCH_MODE_LABEL,
@@ -76,6 +80,7 @@ import {
   type TestImageRef,
   type TestCaseInputMode,
   type ToolCallExpectedBehavior,
+  type ToolSimulation,
   type ToolParamAssertion,
   type ToolParamMatchMode,
 } from "@/data/test-suites";
@@ -87,6 +92,7 @@ export type CaseEditorDraft = {
   inputMode: TestCaseInputMode;
   contextTurns: ContextTurn[];
   turns: FixedInputTurn[];
+  toolSimulations: ToolSimulation[];
   overallCriteria: OverallCriterion[];
 };
 
@@ -104,15 +110,7 @@ function contextPlaceholder(role: ContextRole): string {
   return '{\n  "status": "accepted"\n}';
 }
 
-/** 编辑器内使用的工具选项（来自 Agent 工具库） */
-export type EditorToolOption = {
-  id: string;
-  functionName: string;
-  label: string;
-  parameters: { name: string; description: string }[];
-};
-
-type EditorSectionId = "context" | "main" | "overall";
+type EditorSectionId = "context" | "main" | "tools" | "overall";
 
 function editorSections(): {
   id: EditorSectionId;
@@ -121,6 +119,7 @@ function editorSections(): {
   return [
     { id: "context", label: "对话上下文" },
     { id: "main", label: "固定输入" },
+    { id: "tools", label: "工具模拟" },
     { id: "overall", label: "整体评估" },
   ];
 }
@@ -156,6 +155,7 @@ function toEditorToolOption(tool: Tool): EditorToolOption {
     id: tool.id,
     functionName: tool.functionName || tool.name,
     label: tool.name || tool.functionName,
+    type: tool.type,
     parameters: toolParametersFromDefinition(tool),
   };
 }
@@ -175,6 +175,7 @@ export function NextReplyEditorBody({
   const sectionRefs = useRef<Record<EditorSectionId, HTMLElement | null>>({
     context: null,
     main: null,
+    tools: null,
     overall: null,
   });
   const selectedAnchorRef = useRef<EditorSectionId | null>(null);
@@ -230,7 +231,12 @@ export function NextReplyEditorBody({
 
       const containerTop = scrollContainer.getBoundingClientRect().top;
       const activationLine = containerTop + 24;
-      const sectionIds: EditorSectionId[] = ["context", "main", "overall"];
+      const sectionIds: EditorSectionId[] = [
+        "context",
+        "main",
+        "tools",
+        "overall",
+      ];
       let nextSection: EditorSectionId = sectionIds[0];
 
       for (const sectionId of sectionIds) {
@@ -394,7 +400,10 @@ export function NextReplyEditorBody({
 
   function removeTurn(turnId: string) {
     if (draft.turns.length <= 1) return;
-    updateTurns(draft.turns.filter((turn) => turn.id !== turnId));
+    onChange({
+      ...draft,
+      turns: draft.turns.filter((turn) => turn.id !== turnId),
+    });
   }
 
   function handleTurnDragEnd(event: DragEndEvent) {
@@ -681,6 +690,28 @@ export function NextReplyEditorBody({
             </SectionCard>
           </section>
 
+          <section
+            ref={(element) => {
+              sectionRefs.current.tools = element;
+            }}
+            className="scroll-mt-3 space-y-3"
+          >
+            <SectionCard
+              icon={<Wrench size={15} />}
+              title="工具模拟"
+              description="为外部工具配置固定返回；每次调用复用同一结果，缺少模拟时测试立即停止。"
+            >
+              <ToolSimulationEditor
+                value={draft.toolSimulations}
+                tools={tools}
+                turns={draft.turns}
+                onChange={(toolSimulations) =>
+                  onChange({ ...draft, toolSimulations })
+                }
+              />
+            </SectionCard>
+          </section>
+
           <section
             ref={(element) => {
               sectionRefs.current.overall = element;
@@ -1094,7 +1125,7 @@ function SortableFixedInputTurn({
               >
                 <Wrench
                   size={15}
-                  className="mt-0.5 shrink-0 text-violet-400/90"
+                  className="mt-0.5 shrink-0 text-muted-foreground"
                 />
                 <span className="min-w-0">
                   <span className="block text-sm font-medium text-foreground">
@@ -1140,7 +1171,7 @@ function ExpectedBehaviorCard({
         className={cn(
           "group/card w-full rounded-xl border px-3 py-2.5 text-left transition-colors hover:bg-canvas-soft/60",
           behavior.type === "tool_call"
-            ? "border-violet-500/25 bg-violet-500/[0.04]"
+            ? "border-hairline bg-surface-strong/25"
             : "border-hairline bg-canvas-soft/30",
           validationMessage && "border-destructive/40",
         )}
@@ -1168,7 +1199,7 @@ function ExpectedBehaviorCard({
       className={cn(
         "rounded-xl border px-3 py-2.5",
         behavior.type === "tool_call"
-          ? "border-violet-500/30 bg-violet-500/[0.04]"
+          ? "border-hairline bg-surface-strong/25"
           : "border-hairline bg-canvas-soft/40",
         validationMessage && "border-destructive/40",
       )}
@@ -1184,7 +1215,7 @@ function ExpectedBehaviorCard({
           {behavior.type === "reply" ? (
             <MessageSquareText size={14} className="shrink-0 text-muted-foreground" />
           ) : (
-            <Wrench size={14} className="shrink-0 text-violet-400/90" />
+            <Wrench size={14} className="shrink-0 text-muted-foreground" />
           )}
           <span className="text-xs font-medium text-foreground">
             {behavior.type === "reply" ? "回复要求" : "工具调用"}
@@ -1257,7 +1288,7 @@ function BehaviorSummary({ behavior }: { behavior: ExpectedBehavior }) {
   return (
     <div className="space-y-1">
       <div className="flex items-center gap-1.5 text-xs font-medium text-foreground">
-        <Wrench size={13} className="text-violet-400/90" />
+        <Wrench size={13} className="text-muted-foreground" />
         {behavior.expectation === "not_called" ? "禁止调用 · " : ""}
         {name}
       </div>
@@ -1712,87 +1743,9 @@ function ToolCallBehaviorEditor({
         </div>
       )}
 
-      {selectedTool && behavior.expectation === "called" && (
-        <div className="space-y-2 rounded-lg border border-violet-500/20 bg-background/60 p-2.5">
-          <div>
-            <div className="text-xs font-medium text-foreground">
-              Mock 工具响应
-            </div>
-            <p className="mt-0.5 text-[11px] leading-4 text-muted-soft">
-              工具被调用后将此结果注入 pipeline，使下一轮可稳定复现。
-            </p>
-          </div>
-          <div className="grid gap-2 sm:grid-cols-[minmax(0,1fr)_112px]">
-            <Select
-              value={behavior.mockResponse.outcome}
-              onValueChange={(value) =>
-                onChange({
-                  ...behavior,
-                  mockResponse: {
-                    ...behavior.mockResponse,
-                    outcome: value as "success" | "error",
-                  },
-                })
-              }
-            >
-              <SelectTrigger
-                aria-label="Mock 工具响应类型"
-                className="h-8 border-hairline-strong bg-background text-xs"
-              >
-                <SelectValue />
-              </SelectTrigger>
-              <SelectContent>
-                <SelectItem value="success">成功响应</SelectItem>
-                <SelectItem value="error">错误响应</SelectItem>
-              </SelectContent>
-            </Select>
-            <div className="flex items-center gap-1.5">
-              <Input
-                type="number"
-                min={0}
-                max={60000}
-                step={100}
-                value={behavior.mockResponse.delayMs}
-                onChange={(event) => {
-                  const next = Number(event.target.value);
-                  onChange({
-                    ...behavior,
-                    mockResponse: {
-                      ...behavior.mockResponse,
-                      delayMs: Number.isFinite(next)
-                        ? Math.min(60000, Math.max(0, Math.round(next)))
-                        : 0,
-                    },
-                  });
-                }}
-                aria-label="Mock 工具响应延迟"
-                className="h-8 border-hairline-strong bg-background text-xs"
-              />
-              <span className="text-[11px] text-muted-soft">ms</span>
-            </div>
-          </div>
-          <Textarea
-            value={behavior.mockResponse.body}
-            onChange={(event) =>
-              onChange({
-                ...behavior,
-                mockResponse: {
-                  ...behavior.mockResponse,
-                  body: event.target.value,
-                },
-              })
-            }
-            rows={5}
-            aria-label="Mock 工具响应 JSON"
-            placeholder={'{\n  "status": "ok"\n}'}
-            className="field-sizing-fixed min-h-[112px] resize-y border-hairline-strong bg-background font-mono text-xs"
-          />
-        </div>
-      )}
-
       {selectedTool && behavior.expectation === "not_called" && (
         <p className="rounded-lg border border-hairline bg-background/60 px-2.5 py-2 text-[11px] leading-4 text-muted-soft">
-          本轮该工具的允许调用次数为 0；参数与 Mock 响应无需配置。
+          本轮该工具的允许调用次数为 0；参数无需配置。
         </p>
       )}
     </div>
diff --git a/frontend/src/components/test-cases/tool-simulation-editor.tsx b/frontend/src/components/test-cases/tool-simulation-editor.tsx
new file mode 100644
index 0000000..07a7595
--- /dev/null
+++ b/frontend/src/components/test-cases/tool-simulation-editor.tsx
@@ -0,0 +1,381 @@
+"use client";
+
+import { MonitorSmartphone, Plus, Server, Trash2, Wrench } from "lucide-react";
+import { useMemo, useState } from "react";
+
+import { Button } from "@/components/ui/button";
+import { Input } from "@/components/ui/input";
+import {
+  Popover,
+  PopoverContent,
+  PopoverTrigger,
+} from "@/components/ui/popover";
+import {
+  Select,
+  SelectContent,
+  SelectItem,
+  SelectTrigger,
+  SelectValue,
+} from "@/components/ui/select";
+import { Textarea } from "@/components/ui/textarea";
+import {
+  createToolSimulation,
+  getToolSimulationValidationMessage,
+  type FixedInputTurn,
+  type ToolSimulation,
+  type ToolSimulationResponse,
+} from "@/data/test-suites";
+import { cn } from "@/lib/utils";
+
+export type EditorToolOption = {
+  id: string;
+  functionName: string;
+  label: string;
+  type: "http" | "mcp" | "client";
+  parameters: { name: string; description: string }[];
+};
+
+const TOOL_TYPE_LABEL: Record<EditorToolOption["type"], string> = {
+  http: "服务端 · HTTP",
+  mcp: "服务端 · MCP",
+  client: "客户端交互",
+};
+
+export function ToolSimulationEditor({
+  value,
+  tools,
+  turns,
+  onChange,
+}: {
+  value: ToolSimulation[];
+  tools: EditorToolOption[];
+  turns: FixedInputTurn[];
+  onChange: (value: ToolSimulation[]) => void;
+}) {
+  const [addOpen, setAddOpen] = useState(false);
+  const configuredIds = useMemo(
+    () => new Set(value.map((simulation) => simulation.toolId)),
+    [value],
+  );
+  const availableTools = tools.filter((tool) => !configuredIds.has(tool.id));
+  const turnIds = useMemo(() => new Set(turns.map((turn) => turn.id)), [turns]);
+
+  function addTool(tool: EditorToolOption) {
+    onChange([
+      ...value,
+      createToolSimulation({
+        toolId: tool.id,
+        functionName: tool.functionName,
+        toolType: tool.type,
+      }),
+    ]);
+    setAddOpen(false);
+  }
+
+  function updateSimulation(index: number, simulation: ToolSimulation) {
+    onChange(
+      value.map((item, itemIndex) =>
+        itemIndex === index ? simulation : item,
+      ),
+    );
+  }
+
+  return (
+    <div className="space-y-3">
+      {value.length === 0 ? (
+        <div className="rounded-xl border border-dashed border-hairline-strong bg-canvas-soft/50 px-4 py-7 text-center">
+          <Wrench size={18} className="mx-auto text-muted-soft" />
+          <p className="mt-2 text-xs font-medium text-foreground">
+            暂未配置工具模拟
+          </p>
+          <p className="mx-auto mt-1 max-w-md text-[11px] leading-4 text-muted-soft">
+            运行中调用未配置的外部工具将立即停止当前用例。
+          </p>
+        </div>
+      ) : (
+        <div className="space-y-2.5">
+          {value.map((simulation, index) => {
+            const tool =
+              tools.find((candidate) => candidate.id === simulation.toolId) ??
+              tools.find(
+                (candidate) =>
+                  candidate.functionName === simulation.functionName,
+              ) ??
+              null;
+            return (
+              <ToolSimulationCard
+                key={simulation.toolId}
+                simulation={simulation}
+                tool={tool}
+                validationMessage={getToolSimulationValidationMessage(
+                  simulation,
+                  turnIds,
+                )}
+                onChange={(next) => updateSimulation(index, next)}
+                onRemove={() =>
+                  onChange(value.filter((_, itemIndex) => itemIndex !== index))
+                }
+              />
+            );
+          })}
+        </div>
+      )}
+
+      <Popover open={addOpen} onOpenChange={setAddOpen}>
+        <PopoverTrigger asChild>
+          <Button
+            type="button"
+            variant="outline"
+            size="sm"
+            className="h-8 gap-1.5 rounded-full border-hairline-strong text-xs text-muted-foreground"
+            disabled={availableTools.length === 0}
+          >
+            <Plus size={14} />
+            添加工具模拟
+          </Button>
+        </PopoverTrigger>
+        <PopoverContent align="start" className="w-80 p-1.5">
+          <div className="px-2.5 py-1.5 text-[11px] font-medium tracking-wide text-muted-soft uppercase">
+            选择工具
+          </div>
+          {availableTools.length === 0 ? (
+            <p className="px-2.5 py-3 text-xs text-muted-foreground">
+              所有可用工具均已配置
+            </p>
+          ) : (
+            availableTools.map((tool) => (
+              <button
+                key={tool.id}
+                type="button"
+                className="flex w-full items-center gap-2.5 rounded-lg px-2.5 py-2 text-left transition-colors hover:bg-surface-strong"
+                onClick={() => addTool(tool)}
+              >
+                <ToolTypeIcon type={tool.type} />
+                <span className="min-w-0 flex-1">
+                  <span className="block truncate text-sm font-medium text-foreground">
+                    {tool.label || tool.functionName}
+                  </span>
+                  <span className="block truncate font-mono text-[11px] text-muted-soft">
+                    {tool.functionName}
+                  </span>
+                </span>
+                <span className="shrink-0 rounded-full border border-hairline px-2 py-0.5 text-[10px] text-muted-soft">
+                  {TOOL_TYPE_LABEL[tool.type]}
+                </span>
+              </button>
+            ))
+          )}
+        </PopoverContent>
+      </Popover>
+    </div>
+  );
+}
+
+function ToolSimulationCard({
+  simulation,
+  tool,
+  validationMessage,
+  onChange,
+  onRemove,
+}: {
+  simulation: ToolSimulation;
+  tool: EditorToolOption | null;
+  validationMessage: string | null;
+  onChange: (simulation: ToolSimulation) => void;
+  onRemove: () => void;
+}) {
+  const toolType = tool?.type;
+  const response =
+    simulation.mode === "fixed"
+      ? simulation.response
+      : simulation.steps[0]
+        ? responseFromLegacyStep(simulation.steps[0])
+        : defaultResponse(toolType, simulation.functionName);
+
+  function updateFixedResponse(nextResponse: ToolSimulationResponse) {
+    onChange({
+      toolId: simulation.toolId,
+      functionName: simulation.functionName,
+      mode: "fixed",
+      response: nextResponse,
+    });
+  }
+
+  return (
+    <div
+      className={cn(
+        "rounded-xl border border-hairline bg-canvas-soft/30 p-3",
+        validationMessage && "border-destructive/40",
+      )}
+    >
+      <div className="flex items-start gap-2.5">
+        <div className="mt-0.5 flex h-8 w-8 shrink-0 items-center justify-center rounded-full bg-surface-strong text-muted-foreground">
+          {toolType ? <ToolTypeIcon type={toolType} /> : <Wrench size={14} />}
+        </div>
+        <div className="min-w-0 flex-1">
+          <div className="flex flex-wrap items-center gap-2">
+            <span className="truncate text-sm font-medium text-foreground">
+              {tool?.label || simulation.functionName}
+            </span>
+            <span className="rounded-full border border-hairline px-2 py-0.5 text-[10px] text-muted-soft">
+              {toolType ? TOOL_TYPE_LABEL[toolType] : "工具已不可用"}
+            </span>
+          </div>
+          <div className="mt-0.5 truncate font-mono text-[11px] text-muted-soft">
+            {simulation.functionName}
+          </div>
+        </div>
+        <Button
+          type="button"
+          variant="ghost"
+          size="icon-sm"
+          className="h-7 w-7 shrink-0 text-muted-soft hover:text-destructive"
+          onClick={onRemove}
+          aria-label={`删除 ${simulation.functionName} 工具模拟`}
+        >
+          <Trash2 size={13} />
+        </Button>
+      </div>
+
+      {simulation.mode === "sequence" ? (
+        <div className="mt-3 flex flex-wrap items-center justify-between gap-2 rounded-lg border border-hairline bg-surface-strong/45 px-3 py-2">
+          <p className="text-[11px] leading-4 text-muted-foreground">
+            这是旧版顺序配置。转换后将复用第一条响应作为固定返回。
+          </p>
+          <button
+            type="button"
+            className="h-7 rounded-full border border-hairline-strong px-2.5 text-[11px] text-foreground transition-colors hover:bg-background"
+            onClick={() => updateFixedResponse(response)}
+          >
+            转为固定返回
+          </button>
+        </div>
+      ) : null}
+
+      <div className="mt-3 rounded-lg border border-hairline bg-background/60 p-2.5">
+        <SimulationResponseFields
+          value={response}
+          toolType={toolType}
+          ariaPrefix={`${simulation.functionName} 固定返回`}
+          onChange={updateFixedResponse}
+        />
+        <p className="mt-2 text-[11px] leading-4 text-muted-soft">
+          每次调用都返回此结果；调用次数仍由预期行为单独判断。
+        </p>
+      </div>
+
+      {validationMessage ? (
+        <p className="mt-2 text-[11px] text-destructive" role="alert">
+          {validationMessage}
+        </p>
+      ) : null}
+    </div>
+  );
+}
+
+function SimulationResponseFields({
+  value,
+  toolType,
+  ariaPrefix,
+  onChange,
+}: {
+  value: ToolSimulationResponse;
+  toolType: EditorToolOption["type"] | undefined;
+  ariaPrefix: string;
+  onChange: (value: ToolSimulationResponse) => void;
+}) {
+  return (
+    <div className="space-y-2">
+      <div className="grid gap-2 sm:grid-cols-[minmax(0,1fr)_112px]">
+        <Select
+          value={value.outcome}
+          onValueChange={(outcome) =>
+            onChange({ ...value, outcome: outcome as "success" | "error" })
+          }
+        >
+          <SelectTrigger
+            aria-label={`${ariaPrefix}状态`}
+            className="h-8 border-hairline-strong bg-background text-xs"
+          >
+            <SelectValue />
+          </SelectTrigger>
+          <SelectContent>
+            <SelectItem value="success">成功响应</SelectItem>
+            <SelectItem value="error">错误响应</SelectItem>
+          </SelectContent>
+        </Select>
+        <div className="flex items-center gap-1.5">
+          <Input
+            type="number"
+            min={0}
+            max={60000}
+            step={100}
+            value={value.delayMs}
+            onChange={(event) => {
+              const next = Number(event.target.value);
+              onChange({
+                ...value,
+                delayMs: Number.isFinite(next)
+                  ? Math.min(60000, Math.max(0, Math.round(next)))
+                  : 0,
+              });
+            }}
+            aria-label={`${ariaPrefix}延迟`}
+            className="h-8 border-hairline-strong bg-background text-xs"
+          />
+          <span className="text-[11px] text-muted-soft">ms</span>
+        </div>
+      </div>
+      <Textarea
+        value={value.body}
+        onChange={(event) => onChange({ ...value, body: event.target.value })}
+        rows={5}
+        aria-label={`${ariaPrefix} JSON`}
+        placeholder={
+          toolType === "client"
+            ? '{\n  "action": "selected",\n  "value": "basic"\n}'
+            : '{\n  "status": "ok"\n}'
+        }
+        className="field-sizing-fixed min-h-[112px] resize-y border-hairline-strong bg-background font-mono text-xs"
+      />
+      {toolType === "client" ? (
+        <p className="text-[11px] leading-4 text-muted-soft">
+          JSON 表示客户端回传给 Agent 的用户操作，例如选择、关闭或取消。
+        </p>
+      ) : null}
+    </div>
+  );
+}
+
+function defaultResponse(
+  toolType: EditorToolOption["type"] | undefined,
+  functionName: string,
+): ToolSimulationResponse {
+  return {
+    outcome: "success",
+    body:
+      toolType === "client"
+        ? functionName === "show_message"
+          ? '{\n  "action": "confirmed"\n}'
+          : '{\n  "action": "selected",\n  "value": ""\n}'
+        : '{\n  "status": "ok"\n}',
+    delayMs: 0,
+  };
+}
+
+function responseFromLegacyStep(
+  step: Extract<ToolSimulation, { mode: "sequence" }>["steps"][number],
+): ToolSimulationResponse {
+  return {
+    outcome: step.outcome,
+    body: step.body,
+    delayMs: step.delayMs,
+  };
+}
+
+function ToolTypeIcon({ type }: { type: EditorToolOption["type"] }) {
+  if (type === "client") {
+    return <MonitorSmartphone size={14} className="shrink-0" />;
+  }
+  return <Server size={14} className="shrink-0" />;
+}
diff --git a/frontend/src/data/batch-run.ts b/frontend/src/data/batch-run.ts
index 5520483..ca1ea7c 100644
--- a/frontend/src/data/batch-run.ts
+++ b/frontend/src/data/batch-run.ts
@@ -31,6 +31,16 @@ export type BatchExecutionError = {
   retryable: boolean;
 };
 
+const TEST_EXECUTION_ERROR_LABEL: Record<string, string> = {
+  UNMOCKED_TOOL_CALL: "工具未配置模拟",
+  TOOL_SIMULATION_EXHAUSTED: "工具模拟响应已耗尽",
+  TOOL_SIMULATION_SCOPE_MISMATCH: "工具模拟阶段不匹配",
+};
+
+export function testExecutionErrorLabel(code: string): string {
+  return TEST_EXECUTION_ERROR_LABEL[code] ?? code;
+}
+
 export type BatchEvaluationResult = {
   id: string;
   label: string;
diff --git a/frontend/src/data/test-suites.ts b/frontend/src/data/test-suites.ts
index 90c6a6f..b105d3d 100644
--- a/frontend/src/data/test-suites.ts
+++ b/frontend/src/data/test-suites.ts
@@ -56,13 +56,37 @@ export type ReplyExpectedBehavior = {
   llmCriteria: string;
 };
 
-type ToolMockResponse = {
+export type ToolSimulationResponse = {
   outcome: "success" | "error";
   /** JSON 文本，运行时会作为工具返回值注入 pipeline */
   body: string;
   delayMs: number;
 };
 
+export type ToolSimulationScope =
+  | { phase: "any" }
+  | { phase: "startup" }
+  | { phase: "turn"; turnId: string };
+
+export type ToolSimulationStep = ToolSimulationResponse & {
+  id: string;
+  scope: ToolSimulationScope;
+};
+
+export type ToolSimulation = {
+  toolId: string;
+  functionName: string;
+} & (
+  | {
+      mode: "fixed";
+      response: ToolSimulationResponse;
+    }
+  | {
+      mode: "sequence";
+      steps: ToolSimulationStep[];
+    }
+);
+
 /** 预期行为：工具调用或禁止调用指定工具 */
 export type ToolCallExpectedBehavior = {
   id: string;
@@ -74,8 +98,6 @@ export type ToolCallExpectedBehavior = {
   maxCalls: number | null;
   /** 仅包含用户主动配置的参数；未列出的参数不校验 */
   paramAssertions: ToolParamAssertion[];
-  /** 调用工具后注入 pipeline 的稳定返回值 */
-  mockResponse: ToolMockResponse;
 };
 
 export type ExpectedBehavior = ReplyExpectedBehavior | ToolCallExpectedBehavior;
@@ -132,6 +154,8 @@ export type TestCase = {
   contextTurns: ContextTurn[];
   /** 固定脚本轮次 */
   turns: FixedInputTurn[];
+  /** 测试用例级工具模拟；与工具调用断言相互独立 */
+  toolSimulations: ToolSimulation[];
   /** 整段对话业务目标预期；所有标准均通过才算用例通过 */
   overallCriteria: OverallCriterion[];
   /** 套件内排序，越小越靠前 */
@@ -148,7 +172,11 @@ export const DEFAULT_INPUT_MODE: TestCaseInputMode = "fixed_script_text";
 export function getTestCaseValidationMessage(
   item: Pick<
     TestCase,
-    "inputMode" | "contextTurns" | "turns" | "overallCriteria"
+    | "inputMode"
+    | "contextTurns"
+    | "turns"
+    | "toolSimulations"
+    | "overallCriteria"
   >,
 ): string | null {
   if (item.inputMode !== "fixed_script_text") {
@@ -248,6 +276,49 @@ export function getTestCaseValidationMessage(
       }
     }
   }
+  const simulationToolIds = new Set<string>();
+  const simulationStepIds = new Set<string>();
+  const turnIds = new Set(item.turns.map((turn) => turn.id));
+  for (let index = 0; index < item.toolSimulations.length; index += 1) {
+    const simulation = item.toolSimulations[index];
+    if (simulationToolIds.has(simulation.toolId)) {
+      return `工具模拟 ${index + 1}：同一工具只能配置一条模拟`;
+    }
+    simulationToolIds.add(simulation.toolId);
+    const message = getToolSimulationValidationMessage(
+      simulation,
+      turnIds,
+      simulationStepIds,
+    );
+    if (message) return `工具模拟 ${index + 1}：${message}`;
+  }
+  for (let turnIndex = 0; turnIndex < item.turns.length; turnIndex += 1) {
+    const turn = item.turns[turnIndex];
+    for (const behavior of turn.behaviors) {
+      if (behavior.type !== "tool_call" || behavior.expectation !== "called") {
+        continue;
+      }
+      const simulation = item.toolSimulations.find(
+        (candidate) => candidate.toolId === behavior.toolId,
+      );
+      if (!simulation) {
+        return `第 ${turnIndex + 1} 轮工具 ${behavior.functionName} 缺少工具模拟`;
+      }
+      if (simulation.functionName !== behavior.functionName) {
+        return `第 ${turnIndex + 1} 轮工具 ${behavior.functionName} 的断言与模拟不匹配`;
+      }
+      if (
+        simulation.mode === "sequence" &&
+        !simulation.steps.some(
+          (step) =>
+            step.scope.phase === "any" ||
+            (step.scope.phase === "turn" && step.scope.turnId === turn.id),
+        )
+      ) {
+        return `第 ${turnIndex + 1} 轮工具 ${behavior.functionName} 缺少覆盖本轮的模拟步骤`;
+      }
+    }
+  }
   for (let index = 0; index < item.overallCriteria.length; index += 1) {
     const criterion = item.overallCriteria[index];
     if (!criterion.name.trim()) {
@@ -269,7 +340,11 @@ export function getTestCaseValidationMessage(
 export function isTestCaseRunnable(
   item: Pick<
     TestCase,
-    "inputMode" | "contextTurns" | "turns" | "overallCriteria"
+    | "inputMode"
+    | "contextTurns"
+    | "turns"
+    | "toolSimulations"
+    | "overallCriteria"
   >,
 ): boolean {
   return getTestCaseValidationMessage(item) === null;
@@ -312,7 +387,55 @@ export const TOOL_PARAM_MATCH_MODE_LABEL: Record<ToolParamMatchMode, string> = {
 };
 
 export const OVERALL_CRITERION_TEXT_MAX = 2000;
-const TOOL_MOCK_DELAY_MAX_MS = 60_000;
+const TOOL_SIMULATION_DELAY_MAX_MS = 60_000;
+
+function getSimulationResponseValidationMessage(
+  response: ToolSimulationResponse,
+): string | null {
+  if (!response.body.trim()) return "请填写工具返回值";
+  try {
+    JSON.parse(response.body);
+  } catch {
+    return "工具返回值必须是有效 JSON";
+  }
+  if (
+    !Number.isInteger(response.delayMs) ||
+    response.delayMs < 0 ||
+    response.delayMs > TOOL_SIMULATION_DELAY_MAX_MS
+  ) {
+    return "模拟延迟需为 0–60000 毫秒的整数";
+  }
+  return null;
+}
+
+export function getToolSimulationValidationMessage(
+  simulation: ToolSimulation,
+  turnIds: Set<string>,
+  seenStepIds: Set<string> = new Set(),
+): string | null {
+  if (!simulation.toolId.trim() || !simulation.functionName.trim()) {
+    return "请选择有效工具";
+  }
+  if (simulation.mode === "fixed") {
+    return getSimulationResponseValidationMessage(simulation.response);
+  }
+  if (simulation.steps.length === 0) return "顺序返回至少需要一个步骤";
+  for (let index = 0; index < simulation.steps.length; index += 1) {
+    const step = simulation.steps[index];
+    if (!step.id.trim()) return `步骤 ${index + 1} 缺少 ID`;
+    if (seenStepIds.has(step.id)) return `步骤 ID 重复：${step.id}`;
+    seenStepIds.add(step.id);
+    if (
+      step.scope.phase === "turn" &&
+      !turnIds.has(step.scope.turnId)
+    ) {
+      return `步骤 ${index + 1} 引用了不存在的轮次`;
+    }
+    const message = getSimulationResponseValidationMessage(step);
+    if (message) return `步骤 ${index + 1}：${message}`;
+  }
+  return null;
+}
 
 export function getExpectedBehaviorValidationMessage(
   behavior: ExpectedBehavior,
@@ -353,21 +476,6 @@ export function getExpectedBehaviorValidationMessage(
       (item) => !item.name.trim() || !item.value.trim(),
     );
     if (emptyParam) return "工具参数断言的名称和值不能为空";
-    if (!behavior.mockResponse.body.trim()) {
-      return "请填写 Mock 工具返回值";
-    }
-    try {
-      JSON.parse(behavior.mockResponse.body);
-    } catch {
-      return "Mock 工具返回值必须是有效 JSON";
-    }
-    if (
-      !Number.isInteger(behavior.mockResponse.delayMs) ||
-      behavior.mockResponse.delayMs < 0 ||
-      behavior.mockResponse.delayMs > TOOL_MOCK_DELAY_MAX_MS
-    ) {
-      return "Mock 延迟需为 0–60000 毫秒的整数";
-    }
   }
   return null;
 }
@@ -387,6 +495,29 @@ function nextCriterionId() {
   return newEditorId("criterion");
 }
 
+export function createToolSimulation(input: {
+  toolId: string;
+  functionName: string;
+  toolType: "http" | "mcp" | "client";
+}): ToolSimulation {
+  const body =
+    input.toolType === "client"
+      ? input.functionName === "show_message"
+        ? '{\n  "action": "confirmed"\n}'
+        : '{\n  "action": "selected",\n  "value": ""\n}'
+      : '{\n  "status": "ok"\n}';
+  return {
+    toolId: input.toolId,
+    functionName: input.functionName,
+    mode: "fixed",
+    response: {
+      outcome: "success",
+      body,
+      delayMs: 0,
+    },
+  };
+}
+
 export function createOverallCriterion(
   name = "新的评估标准",
 ): OverallCriterion {
@@ -453,11 +584,6 @@ export function createToolCallBehavior(input?: {
     minCalls: 1,
     maxCalls: null,
     paramAssertions: [],
-    mockResponse: {
-      outcome: "success",
-      body: '{\n  "status": "ok"\n}',
-      delayMs: 0,
-    },
   };
 }
 
@@ -482,11 +608,6 @@ function cloneBehavior(behavior: ExpectedBehavior): ExpectedBehavior {
     minCalls: behavior.minCalls,
     maxCalls: behavior.maxCalls,
     paramAssertions: behavior.paramAssertions.map((item) => ({ ...item })),
-    mockResponse: {
-      outcome: behavior.mockResponse.outcome,
-      body: behavior.mockResponse.body,
-      delayMs: behavior.mockResponse.delayMs,
-    },
   };
 }
 
@@ -499,6 +620,39 @@ export function cloneTurns(turns: FixedInputTurn[]): FixedInputTurn[] {
   }));
 }
 
+export function prepareFixedToolSimulations(
+  values: ToolSimulation[] | null | undefined,
+): ToolSimulation[] {
+  return (values ?? []).map((simulation) =>
+    simulation.mode === "fixed"
+      ? {
+          toolId: simulation.toolId,
+          functionName: simulation.functionName,
+          mode: "fixed",
+          response: { ...simulation.response },
+        }
+      : {
+          toolId: simulation.toolId,
+          functionName: simulation.functionName,
+          mode: "fixed",
+          response: simulation.steps[0]
+            ? {
+                outcome: simulation.steps[0].outcome,
+                body: simulation.steps[0].body,
+                delayMs: simulation.steps[0].delayMs,
+              }
+            : {
+                outcome: "success",
+                body:
+                  simulation.functionName === "show_message"
+                    ? '{\n  "action": "confirmed"\n}'
+                    : '{\n  "status": "ok"\n}',
+                delayMs: 0,
+              },
+        },
+  );
+}
+
 export function formatUpdatedAt(value?: string | null) {
   if (!value) return "—";
   const date = new Date(value);
diff --git a/frontend/src/lib/api.ts b/frontend/src/lib/api.ts
index 6015d75..5765f5c 100644
--- a/frontend/src/lib/api.ts
+++ b/frontend/src/lib/api.ts
@@ -14,6 +14,7 @@ import type {
   TestCase,
   TestCaseInputMode,
   TestSuite,
+  ToolSimulation,
 } from "@/data/test-suites";
 
 const LOOPBACK_HOSTS = new Set(["localhost", "127.0.0.1", "[::1]"]);
@@ -851,6 +852,7 @@ export type TestCaseWrite = {
   inputMode: TestCaseInputMode;
   contextTurns: ContextTurn[];
   turns: FixedInputTurn[];
+  toolSimulations: ToolSimulation[];
   overallCriteria: OverallCriterion[];
 };