Losita/mai-bot
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat：新增按顺序选择 fix：修复timing gate意外tool问题

Browse Source
This commit is contained in:
SengokuCola
2026-04-27 10:53:13 +08:00
parent 859bbb9cbf
commit d32be4741a
7 changed files with 247 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

148
pytests/test_maisaka_timing_gate.py Normal file
View File

@@ -0,0 +1,148 @@
from datetime import datetime
from types import SimpleNamespace
import pytest
from src.core.tooling import ToolExecutionResult
from src.llm_models.payload_content.tool_option import ToolCall
from src.maisaka.chat_loop_service import ChatResponse, MaisakaChatLoopService
from src.maisaka.context_messages import AssistantMessage, TIMING_GATE_INVALID_TOOL_HINT_SOURCE
from src.maisaka.reasoning_engine import MaisakaReasoningEngine
def _build_chat_response(tool_calls: list[ToolCall]) -> ChatResponse:
return ChatResponse(
content="The model returned an invalid timing tool.",
tool_calls=tool_calls,
request_messages=[],
raw_message=AssistantMessage(
content="",
timestamp=datetime.now(),
source_kind="perception",
),
selected_history_count=1,
tool_count=len(tool_calls),
prompt_tokens=10,
built_message_count=1,
completion_tokens=3,
total_tokens=13,
prompt_section=None,
)
@pytest.mark.asyncio
async def test_timing_gate_invalid_tool_defaults_to_no_reply(monkeypatch: pytest.MonkeyPatch) -> None:
runtime = SimpleNamespace(
_force_next_timing_continue=False,
_chat_history=[],
log_prefix="[test]",
stopped=False,
)
def _enter_stop_state() -> None:
runtime.stopped = True
runtime._enter_stop_state = _enter_stop_state
engine = MaisakaReasoningEngine(runtime) # type: ignore[arg-type]
async def _fake_timing_gate_sub_agent(**kwargs: object) -> ChatResponse:
del kwargs
return _build_chat_response([
ToolCall(call_id="invalid-timing-tool", func_name="finish", args={}),
])
async def _fail_invoke_tool_call(*args: object, **kwargs: object) -> None:
del args, kwargs
raise AssertionError("invalid timing tools must not be executed")
monkeypatch.setattr(engine, "_run_timing_gate_sub_agent", _fake_timing_gate_sub_agent)
monkeypatch.setattr(engine, "_invoke_tool_call", _fail_invoke_tool_call)
action, response, tool_results, tool_monitor_results = await engine._run_timing_gate(object()) # type: ignore[arg-type]
assert action == "no_reply"
assert response.tool_calls[0].func_name == "finish"
assert runtime.stopped is True
assert tool_monitor_results == []
assert len(runtime._chat_history) == 1
assert runtime._chat_history[0].source == TIMING_GATE_INVALID_TOOL_HINT_SOURCE
assert "finish" in runtime._chat_history[0].processed_plain_text
assert tool_results == [
"- no_reply [非法 Timing 工具]: 返回了 finish，已停止本轮并等待新消息",
]
def test_timing_gate_invalid_tool_hint_keeps_only_latest() -> None:
old_hint = SimpleNamespace(source=TIMING_GATE_INVALID_TOOL_HINT_SOURCE)
runtime = SimpleNamespace(_chat_history=[old_hint])
engine = MaisakaReasoningEngine(runtime) # type: ignore[arg-type]
engine._append_timing_gate_invalid_tool_hint("finish")
engine._append_timing_gate_invalid_tool_hint("reply")
assert len(runtime._chat_history) == 1
hint_message = runtime._chat_history[0]
assert hint_message.source == TIMING_GATE_INVALID_TOOL_HINT_SOURCE
assert "reply" in hint_message.processed_plain_text
assert "finish" not in hint_message.processed_plain_text
def test_timing_gate_invalid_tool_hint_only_visible_to_timing_gate() -> None:
runtime = SimpleNamespace(_chat_history=[])
engine = MaisakaReasoningEngine(runtime) # type: ignore[arg-type]
engine._append_timing_gate_invalid_tool_hint("finish")
hint_message = runtime._chat_history[0]
timing_history = MaisakaChatLoopService._filter_history_for_request_kind(
[hint_message],
request_kind="timing_gate",
)
planner_history = MaisakaChatLoopService._filter_history_for_request_kind(
[hint_message],
request_kind="planner",
)
assert timing_history == [hint_message]
assert planner_history == []
def test_finish_tool_is_not_written_back_to_history() -> None:
finish_call = ToolCall(call_id="finish-call", func_name="finish", args={})
reply_call = ToolCall(call_id="reply-call", func_name="reply", args={})
assistant_message = AssistantMessage(
content="当前不需要继续回复。",
timestamp=datetime.now(),
tool_calls=[finish_call, reply_call],
)
runtime = SimpleNamespace(_chat_history=[assistant_message])
engine = MaisakaReasoningEngine(runtime) # type: ignore[arg-type]
engine._append_tool_execution_result(
finish_call,
ToolExecutionResult(
tool_name="finish",
success=True,
content="当前对话循环已结束本轮思考，等待新的消息到来。",
),
)
assert runtime._chat_history == [assistant_message]
assert [tool_call.func_name for tool_call in assistant_message.tool_calls] == ["reply"]
def test_finish_tool_removes_empty_assistant_history_message() -> None:
finish_call = ToolCall(call_id="finish-call", func_name="finish", args={})
assistant_message = AssistantMessage(
content="",
timestamp=datetime.now(),
tool_calls=[finish_call],
)
runtime = SimpleNamespace(_chat_history=[assistant_message])
engine = MaisakaReasoningEngine(runtime) # type: ignore[arg-type]
engine._append_tool_execution_result(
finish_call,
ToolExecutionResult(tool_name="finish", success=True),
)
assert runtime._chat_history == []

2
src/config/config.py
View File

@@ -57,7 +57,7 @@ MODEL_CONFIG_PATH: Path = (CONFIG_DIR / "model_config.toml").resolve().absolute(
LEGACY_ENV_PATH: Path = (PROJECT_ROOT / ".env").resolve().absolute()
MMC_VERSION: str = "1.0.0"
CONFIG_VERSION: str = "8.9.17"
MODEL_CONFIG_VERSION: str = "1.14.2"
MODEL_CONFIG_VERSION: str = "1.14.3"
logger = get_logger("config")

3
src/config/model_configs.py
View File

@@ -406,9 +406,10 @@ class TaskConfig(ConfigBase):
json_schema_extra={
"x-widget": "select",
"x-icon": "shuffle",
"options": ["balance", "random", "sequential"],
},
)
"""模型选择策略：balance（负载均衡）或 random（随机选择）"""
"""模型选择策略：balance（负载均衡）、random（随机选择）或 sequential（按配置顺序优先选择）"""
class ModelTaskConfig(ConfigBase):

7
src/llm_models/utils_model.py
View File

@@ -683,11 +683,16 @@ class LLMOrchestrator:
ensure_configured_clients_loaded()
strategy = self.model_for_task.selection_strategy.lower()
strategy = self.model_for_task.selection_strategy.strip().lower()
if strategy == "random":
# 随机选择策略
selected_model_name = random.choice(list(available_models.keys()))
elif strategy == "sequential":
# 顺序优先策略：按照配置顺序选择第一个尚未失败的模型。
selected_model_name = next(
model_name for model_name in self.model_for_task.model_list if model_name in available_models
)
elif strategy == "balance":
# 负载均衡策略：根据总tokens和惩罚值选择
selected_model_name = min(

10
src/maisaka/chat_loop_service.py
View File

@@ -32,6 +32,7 @@ from .builtin_tool import get_builtin_tools
from .context_messages import (
AssistantMessage,
LLMContextMessage,
TIMING_GATE_INVALID_TOOL_HINT_SOURCE,
ToolResultMessage,
build_llm_message_from_context,
)
@@ -704,6 +705,15 @@ class MaisakaChatLoopService:
) -> List[LLMContextMessage]:
"""按请求类型过滤不应暴露的历史工具链。"""
if request_kind == "timing_gate":
return selected_history
selected_history = [
message
for message in selected_history
if message.source != TIMING_GATE_INVALID_TOOL_HINT_SOURCE
]
if request_kind != "planner":
return selected_history

1
src/maisaka/context_messages.py
View File

@@ -29,6 +29,7 @@ from src.llm_models.payload_content.tool_option import ToolCall
from .message_adapter import parse_speaker_content
FORWARD_PREVIEW_LIMIT = 4
TIMING_GATE_INVALID_TOOL_HINT_SOURCE = "timing_gate_invalid_tool_hint"
def _guess_image_format(image_bytes: bytes) -> Optional[str]:

84
src/maisaka/reasoning_engine.py
View File

@@ -11,7 +11,7 @@ import traceback
from src.chat.heart_flow.heartFC_utils import CycleDetail
from src.chat.message_receive.message import SessionMessage
from src.common.data_models.message_component_data_model import EmojiComponent, ImageComponent, MessageSequence
from src.common.data_models.message_component_data_model import EmojiComponent, ImageComponent, MessageSequence, TextComponent
from src.common.logger import get_logger
from src.common.prompt_i18n import load_prompt
from src.core.tooling import ToolAvailabilityContext, ToolExecutionContext, ToolExecutionResult, ToolInvocation, ToolSpec
@@ -31,6 +31,7 @@ from .context_messages import (
ComplexSessionMessage,
LLMContextMessage,
SessionBackedMessage,
TIMING_GATE_INVALID_TOOL_HINT_SOURCE,
ToolResultMessage,
contains_complex_message,
)
@@ -54,6 +55,7 @@ logger = get_logger("maisaka_reasoning_engine")
TIMING_GATE_CONTEXT_LIMIT = 24
TIMING_GATE_MAX_TOKENS = 384
TIMING_GATE_TOOL_NAMES = {"continue", "no_reply", "wait"}
HISTORY_SILENT_TOOL_NAMES = {"finish"}
class MaisakaReasoningEngine:
@@ -259,8 +261,21 @@ class MaisakaReasoningEngine:
break
if selected_tool_call is None:
logger.warning(f"{self._runtime.log_prefix} Timing Gate 未返回有效控制工具，默认继续执行 Action Loop")
return "continue", response, tool_result_summaries, tool_monitor_results
invalid_tool_names = [
str(tool_call.func_name).strip()
for tool_call in response.tool_calls
if str(tool_call.func_name).strip()
]
invalid_tool_text = "、".join(invalid_tool_names) if invalid_tool_names else "无工具"
logger.warning(
f"{self._runtime.log_prefix} Timing Gate 未返回有效控制工具：{invalid_tool_text}，将按 no_reply 处理"
)
self._append_timing_gate_invalid_tool_hint(invalid_tool_text)
self._runtime._enter_stop_state()
tool_result_summaries.append(
f"- no_reply [非法 Timing 工具]: 返回了 {invalid_tool_text}，已停止本轮并等待新消息"
)
return "no_reply", response, tool_result_summaries, tool_monitor_results
append_history = False
store_record = selected_tool_call.func_name != "continue"
@@ -286,9 +301,13 @@ class MaisakaReasoningEngine:
timing_action = str(result.metadata.get("timing_action") or selected_tool_call.func_name).strip()
if timing_action not in TIMING_GATE_TOOL_NAMES:
logger.warning(
f"{self._runtime.log_prefix} Timing Gate 返回未知动作 {timing_action!r}，将按 continue 处理"
f"{self._runtime.log_prefix} Timing Gate 返回未知动作 {timing_action!r}，将按 no_reply 处理"
)
return "continue", response, tool_result_summaries, tool_monitor_results
self._runtime._enter_stop_state()
tool_result_summaries.append(
f"- no_reply [未知 Timing 动作]: 返回了 {timing_action!r}，已停止本轮并等待新消息"
)
return "no_reply", response, tool_result_summaries, tool_monitor_results
return timing_action, response, tool_result_summaries, tool_monitor_results
def _build_forced_continue_timing_result(
@@ -324,6 +343,29 @@ class MaisakaReasoningEngine:
[],
)
def _append_timing_gate_invalid_tool_hint(self, invalid_tool_text: str) -> None:
"""写入一条仅 Timing Gate 可见的非法工具提示，并保证最多保留最新一条。"""
self._runtime._chat_history = [
message
for message in self._runtime._chat_history
if message.source != TIMING_GATE_INVALID_TOOL_HINT_SOURCE
]
normalized_tool_text = invalid_tool_text.strip() or "无工具"
hint_content = (
"Timing Gate 上一轮选择了非法工具："
f"{normalized_tool_text}。\n"
"Timing Gate 只能调用 continue、wait 或 no_reply 中的一个工具。"
)
self._runtime._chat_history.append(
SessionBackedMessage(
raw_message=MessageSequence([TextComponent(hint_content)]),
visible_text=hint_content,
timestamp=datetime.now(),
source_kind=TIMING_GATE_INVALID_TOOL_HINT_SOURCE,
)
)
@staticmethod
def _mark_timing_gate_completed(timing_action: str) -> bool:
"""根据门控动作决定下一轮是否还需要重新执行 timing。"""
@@ -1210,6 +1252,10 @@ class MaisakaReasoningEngine:
result: 统一工具执行结果。
"""
if tool_call.func_name in HISTORY_SILENT_TOOL_NAMES:
self._remove_tool_call_from_history(tool_call)
return
history_content = result.get_history_content()
if not history_content:
history_content = "工具执行成功。" if result.success else f"工具 {tool_call.func_name} 执行失败。"
@@ -1224,6 +1270,34 @@ class MaisakaReasoningEngine:
)
)
def _remove_tool_call_from_history(self, tool_call: ToolCall) -> None:
"""从历史里的 assistant 消息中移除控制类工具调用。"""
tool_call_id = str(tool_call.call_id or "").strip()
if not tool_call_id:
return
for index in range(len(self._runtime._chat_history) - 1, -1, -1):
message = self._runtime._chat_history[index]
if not isinstance(message, AssistantMessage) or not message.tool_calls:
continue
remaining_tool_calls = [
existing_tool_call
for existing_tool_call in message.tool_calls
if str(existing_tool_call.call_id or "").strip() != tool_call_id
]
if len(remaining_tool_calls) == len(message.tool_calls):
continue
if remaining_tool_calls:
message.tool_calls = remaining_tool_calls
elif message.content.strip():
message.tool_calls = []
else:
del self._runtime._chat_history[index]
return
def _append_timing_gate_execution_result(
self,
response: ChatResponse,