refactor: tech debt Wave 1+2 (except Exception 收尾 + core/experts Any 治理) (#10)

src/agentkit/bus/memory_bus.py

@@ -70,7 +70,9 @@ class InMemoryMessageBus:
             for handler in self._subscribers[recipient]:
                 try:
                     await handler(message)
-                except Exception as e:
+                except asyncio.CancelledError:
+                    raise
+                except Exception as e:  # noqa: BLE001 — user-defined async handlers can throw arbitrary exceptions
                     logger.warning(f"Handler error for {recipient}: {e}")
 
         # Check if this is a response to a pending request

src/agentkit/bus/redis_bus.py

@@ -62,7 +62,7 @@ class RedisMessageBus:
 
         try:
             await redis.xadd(stream_key, {"data": json.dumps(data)})
-        except Exception as e:
+        except (ConnectionError, OSError, asyncio.TimeoutError, RuntimeError) as e:
             logger.error(f"Failed to publish message to {stream_key}: {e}")
             raise
 
@@ -103,7 +103,9 @@ class RedisMessageBus:
             await redis.xgroup_create(
                 stream_key, self._consumer_group, id="0", mkstream=True,
             )
-        except Exception:
+        except asyncio.CancelledError:
+            raise
+        except Exception:  # noqa: BLE001 — xgroup_create raises ResponseError on BUSYGROUP; redis is optional dep
             pass  # Group already exists
 
         while True:
@@ -126,12 +128,16 @@ class RedisMessageBus:
                                 for handler in self._subscribers.get(agent_name, []):
                                     try:
                                         await handler(message)
-                                    except Exception as e:
+                                    except asyncio.CancelledError:
+                                        raise
+                                    except Exception as e:  # noqa: BLE001 — user-defined async handlers can throw arbitrary exceptions
                                         logger.warning(f"Handler error for {agent_name}: {e}")
 
                                 # Acknowledge message
                                 await redis.xack(stream_key, self._consumer_group, msg_id)
-                            except Exception as e:
+                            except asyncio.CancelledError:
+                                raise
+                            except Exception as e:  # noqa: BLE001 — wraps json parse + handler + xack; multi-op block with diverse failure modes
                                 logger.warning(f"Failed to process message {msg_id}: {e}")
                                 # Move to dead letter after max retries
                                 await self._handle_failed_message(
@@ -139,7 +145,7 @@ class RedisMessageBus:
                                 )
             except asyncio.CancelledError:
                 break
-            except Exception as e:
+            except Exception as e:  # noqa: BLE001 — consumer loop top-level fallback; must keep running on transient errors
                 logger.error(f"Consumer error for {agent_name}: {e}")
                 await asyncio.sleep(1)
 
@@ -157,7 +163,7 @@ class RedisMessageBus:
             await redis.xadd(dead_key, fields)
             await redis.xack(stream_key, self._consumer_group, msg_id)
             logger.warning(f"Message {msg_id} moved to dead letter queue")
-        except Exception as e:
+        except (ConnectionError, OSError, asyncio.TimeoutError, RuntimeError) as e:
             logger.error(f"Failed to move message to dead letter: {e}")
 
     async def unsubscribe(self, agent_name: str) -> None:
@@ -205,7 +211,7 @@ class RedisMessageBus:
             stream_key = self._stream_key(agent_name)
             try:
                 await redis.xadd(stream_key, {"data": json.dumps(data)})
-            except Exception as e:
+            except (ConnectionError, OSError, asyncio.TimeoutError, RuntimeError) as e:
                 logger.error(f"Failed to broadcast to {agent_name}: {e}")
 
         # Check pending requests (only for replies)
@@ -222,7 +228,7 @@ class RedisMessageBus:
         try:
             redis = await self._get_redis()
             return await redis.ping()
-        except Exception:
+        except (ConnectionError, OSError, asyncio.TimeoutError, RuntimeError):
             return False
 
     @property
@@ -257,7 +263,9 @@ def create_message_bus(
             )
             logger.info(f"MessageBus backend: redis_streams ({redis_url})")
             return bus
-        except Exception as exc:
+        except asyncio.CancelledError:
+            raise
+        except Exception as exc:  # noqa: BLE001 — factory fallback to InMemoryMessageBus; must catch import/init errors broadly
             logger.warning(
                 f"Failed to initialise RedisMessageBus ({exc}), "
                 f"falling back to InMemoryMessageBus"

src/agentkit/chat/request_preprocessor.py

@@ -194,7 +194,7 @@ class RequestPreprocessor:
         """Resolve an explicitly specified skill via @skill:xxx prefix."""
         try:
             skill = registry.get(skill_name)
-        except Exception:
+        except (ValueError, KeyError, TypeError, RuntimeError):
             logger.warning(f"Skill '{skill_name}' not found, falling back to REACT")
             return SkillRoutingResult(
                 clean_content=clean_content,

src/agentkit/chat/skill_routing.py

@@ -200,7 +200,7 @@ async def resolve_skill_routing(
             result.match_method = "explicit"
             result.match_confidence = 1.0
             logger.info(f"Session {session_id}: using explicit skill '{explicit_skill}'")
-        except Exception as e:
+        except (ValueError, KeyError, TypeError, RuntimeError) as e:
             logger.warning(
                 f"Session {session_id}: explicit skill '{explicit_skill}' not found: {e}"
             )

src/agentkit/chat/sqlite_conversation_store.py

@@ -291,7 +291,7 @@ class SqliteConversationStore:
                     title_row = await title_cursor.fetchone()
                     if title_row:
                         conv.messages.append(ChatMessage(role="user", content=title_row["content"]))
-                except Exception:
+                except (aiosqlite.Error, ValueError, KeyError, TypeError):
                     pass
                 result.append(conv)
         return result
@@ -321,7 +321,7 @@ class SqliteConversationStore:
             )
             await db.execute("DELETE FROM conversations WHERE id = ?", (conversation_id,))
             await db.commit()
-        except Exception:
+        except (aiosqlite.Error, ValueError, KeyError, RuntimeError):
             await db.rollback()
             logger.exception("Failed to delete conversation %s; rolled back", conversation_id)
             raise
@@ -365,6 +365,6 @@ class SqliteConversationStore:
                 timestamp=self._str_to_dt(row["timestamp"]),
                 metadata=meta,
             )
-        except Exception as e:
+        except (aiosqlite.Error, ValueError, KeyError, TypeError, RuntimeError) as e:
             logger.warning(f"get_first_user_message failed for {conversation_id}: {e}")
             return None

src/agentkit/cli/chat.py

@@ -134,7 +134,7 @@ async def _chat_async(
                         data=data or {},
                     )
                 )
-            except Exception:
+            except (asyncio.QueueFull, RuntimeError, ConnectionError):
                 pass  # EQ is best-effort; never break CLI flow
 
         # Emit session.started event
@@ -177,7 +177,7 @@ async def _chat_async(
             elif p.is_file() and p.suffix in (".yaml", ".yml"):
                 try:
                     loader.load_from_file(str(p))
-                except Exception:
+                except (ValueError, TypeError, KeyError, OSError, RuntimeError):
                     pass
 
     # Build system prompt — inject memory into system prompt
@@ -428,7 +428,9 @@ async def _chat_async(
                         data={"output": full_content},
                     )
 
-        except Exception as e:
+        except asyncio.CancelledError:
+            raise
+        except Exception as e:  # noqa: BLE001 — CLI main loop top-level; must not crash on user-facing errors
             rprint(f"\n[red]Error: {e}[/red]")
             # Emit task.failed to EQ (if enabled)
             if _emit is not None:
@@ -461,7 +463,7 @@ async def _chat_async(
 
                 # Archive old daily logs
                 memory_store.archive_old_dailies(keep_days=2)
-        except Exception:
+        except (ConnectionError, OSError, asyncio.TimeoutError, ValueError, KeyError, RuntimeError):
             pass  # Daily log generation is best-effort
 
     # Close EventQueue if it was enabled (emit session.ended and close)
@@ -473,7 +475,7 @@ async def _chat_async(
                 session_id=session.session_id,
                 data={},
             )
-        except Exception:
+        except (asyncio.QueueFull, RuntimeError, ConnectionError):
             pass
         eq.close()
 
@@ -502,7 +504,7 @@ def _build_gateway(server_config: "ServerConfig") -> "LLMGateway":
         try:
             provider = _create_provider(name, pconf)
             gateway.register_provider(name, provider)
-        except Exception as e:
+        except (ValueError, TypeError, KeyError, RuntimeError, ConnectionError, OSError) as e:
             import logging
 
             logging.getLogger(__name__).warning(f"Failed to register LLM provider '{name}': {e}")
@@ -603,7 +605,9 @@ def _render_pm_collaboration_event(message: dict) -> bool:
                 )
             )
             return True
-    except Exception as e:
+    except asyncio.CancelledError:
+        raise
+    except Exception as e:  # noqa: BLE001 — best-effort rendering; must not break orchestration
         # ponytail: best-effort 渲染不中断编排，但记录日志便于调试
         import logging
 
@@ -782,7 +786,9 @@ async def _execute_team_cli(
             elif etype == "user_intervention":
                 pass  # User typed it themselves
             # Other events (expert_step, expert_result, expert_joined, etc.) are not rendered
-        except Exception:
+        except asyncio.CancelledError:
+            raise
+        except Exception:  # noqa: BLE001 — best-effort rendering; must not break orchestration
             pass  # Rendering is best-effort; never break orchestration
 
     team.handoff_transport.register_handler(team.team_channel, _event_handler)
@@ -816,7 +822,7 @@ async def _execute_team_cli(
             if readable:
                 try:
                     line = sys.stdin.readline()
-                except Exception:
+                except (OSError, ValueError, RuntimeError):
                     line = ""
                 if not line:
                     break  # EOF
@@ -846,12 +852,14 @@ async def _execute_team_cli(
                     )
                 )
 
-    except Exception as e:
+    except asyncio.CancelledError:
+        raise
+    except Exception as e:  # noqa: BLE001 — team execution top-level; must report errors to user without crash
         rprint(f"[red]团队执行错误: {e}[/red]")
     finally:
         try:
             await team.dissolve()
-        except Exception:
+        except (RuntimeError, asyncio.TimeoutError, ConnectionError):
             pass
 
     return True

src/agentkit/core/base.py

@@ -13,7 +13,7 @@ import logging
 import time
 from abc import ABC, abstractmethod
 from datetime import datetime, timezone
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING
 
 import redis.asyncio as aioredis
 
@@ -71,7 +71,7 @@ class BaseAgent(ABC):
         # 可插拔能力（由子类或配置注入）
         self._tools: list["Tool"] = []
         self._memory: "Memory | None" = None
-        self._memory_retriever: Any | None = None
+        self._memory_retriever: object | None = None
 
         # 外部依赖注入（由 start() 时设置）
         self._registry = None
@@ -192,7 +192,7 @@ class BaseAgent(ABC):
             lines.append(f"  - {msg}")
         return "\n".join(lines)
 
-    def _build_skill_context(self) -> dict[str, Any] | None:
+    def _build_skill_context(self) -> dict[str, object] | None:
         """从当前技能配置构建 skill_context，用于 QualityGate skill_match 校验"""
         if not self._skill:
             return None
@@ -216,17 +216,17 @@ class BaseAgent(ABC):
         self._memory = memory
         return self
 
-    def use_memory_retriever(self, retriever: Any) -> "BaseAgent":
+    def use_memory_retriever(self, retriever: object) -> "BaseAgent":
         """设置记忆检索器，用于上下文注入"""
         self._memory_retriever = retriever
         return self
 
-    def set_registry(self, registry: Any) -> "BaseAgent":
+    def set_registry(self, registry: object) -> "BaseAgent":
         """注入注册中心"""
         self._registry = registry
         return self
 
-    def set_dispatcher(self, dispatcher: Any) -> "BaseAgent":
+    def set_dispatcher(self, dispatcher: object) -> "BaseAgent":
         """注入任务分发器"""
         self._dispatcher = dispatcher
         return self
@@ -489,7 +489,7 @@ class BaseAgent(ABC):
         target_agent: str,
         task: TaskMessage,
         reason: str,
-        context: dict[str, Any] | None = None,
+        context: dict[str, object] | None = None,
     ):
         """将当前任务转交给另一个 Agent"""
         if self._redis is None:

src/agentkit/core/config_driven.py

@@ -10,7 +10,7 @@
 import json
 import logging
 import os
-from typing import Any, Callable, Coroutine
+from typing import Callable, Coroutine
 
 import yaml
 
@@ -39,12 +39,12 @@ class AgentConfig:
         task_mode: str = "llm_generate",
         supported_tasks: list[str] | None = None,
         max_concurrency: int = 1,
-        input_schema: dict[str, Any] | None = None,
-        output_schema: dict[str, Any] | None = None,
+        input_schema: dict[str, object] | None = None,
+        output_schema: dict[str, object] | None = None,
         prompt: dict[str, str] | None = None,
-        llm: dict[str, Any] | None = None,
+        llm: dict[str, object] | None = None,
         tools: list[str] | None = None,
-        memory: dict[str, Any] | None = None,
+        memory: dict[str, object] | None = None,
         custom_handler: str | None = None,
     ):
         self.name = name
@@ -96,7 +96,7 @@ class AgentConfig:
             )
 
     @classmethod
-    def from_dict(cls, data: dict[str, Any]) -> "AgentConfig":
+    def from_dict(cls, data: dict[str, object]) -> "AgentConfig":
         """从字典创建配置"""
         return cls(
             name=data["name"],
@@ -128,7 +128,7 @@ class AgentConfig:
             )
         return cls.from_dict(data)
 
-    def to_dict(self) -> dict[str, Any]:
+    def to_dict(self) -> dict[str, object]:
         """序列化为字典"""
         d = {
             "name": self.name,
@@ -197,11 +197,11 @@ class ConfigDrivenAgent(BaseAgent, EvolutionMixin):
         self,
         config: AgentConfig,
         tool_registry: ToolRegistry | None = None,
-        llm_client: Any = None,
+        llm_client: object | None = None,
         custom_handlers: dict[str, Callable[..., Coroutine]] | None = None,
-        llm_gateway: Any = None,  # NEW v2 param: LLMGateway
+        llm_gateway: object | None = None,  # NEW v2 param: LLMGateway
         mcp_servers: dict[str, str] | None = None,  # NEW v2 param: MCP server URLs
-        compressor: Any = None,  # CompressionStrategy | None
+        compressor: object | None = None,  # CompressionStrategy | None
     ):
         # v2: If SkillConfig, extract skill info
         from agentkit.skills.base import SkillConfig, Skill
@@ -310,12 +310,12 @@ class ConfigDrivenAgent(BaseAgent, EvolutionMixin):
                     logger.info(f"Merged skill tool '{tool.name}' into agent '{self.name}'")
 
         # v2: Register MCP tools if mcp_servers provided
-        self._mcp_clients: list[Any] = []
+        self._mcp_clients: list[object] = []
         self._mcp_servers: dict[str, str] = mcp_servers or {}
         self._mcp_tools_registered = False
 
         # Memory integration: 从 config.memory 自动实例化 MemoryRetriever
-        self._memory_retriever: Any | None = None
+        self._memory_retriever: object | None = None
         if config.memory:
             try:
                 from agentkit.memory.retriever import MemoryRetriever
@@ -903,7 +903,7 @@ class ConfigDrivenAgent(BaseAgent, EvolutionMixin):
         )
         return await self.handle_task(enhanced_task)
 
-    def _wrap_llm_client(self, llm_client: Any):
+    def _wrap_llm_client(self, llm_client: object):
         """Wrap legacy llm_client into LLMGateway"""
         from agentkit.llm.gateway import LLMGateway
         from agentkit.llm.protocol import LLMProvider, LLMRequest, LLMResponse, TokenUsage
@@ -911,7 +911,7 @@ class ConfigDrivenAgent(BaseAgent, EvolutionMixin):
         class ClientProvider(LLMProvider):
             """Adapter: wraps legacy llm_client as an LLMProvider"""
 
-            def __init__(self, raw_client: Any):
+            def __init__(self, raw_client: object):
                 self._raw_client = raw_client
 
             async def chat(self, request: LLMRequest) -> LLMResponse:

src/agentkit/core/fallback.py

@@ -9,7 +9,6 @@ no LLM). See ``docs/plans/2026-06-29-003-feat-agent-wave2-medium-coupling-plan.m
 """
 
 from dataclasses import dataclass
-from typing import Any
 
 from agentkit.core.exceptions import (
     LLMProviderError,
@@ -54,7 +53,7 @@ class EmergencyError:
     retryable: bool  # whether a user retry might succeed
     original_error: str  # str(exc) for traceability
 
-    def to_dict(self) -> dict[str, Any]:
+    def to_dict(self) -> dict[str, object]:
         return {
             "error_code": self.error_code,
             "message": self.message,

src/agentkit/core/middleware.py

@@ -24,7 +24,7 @@ from __future__ import annotations
 import json
 import logging
 from dataclasses import dataclass, field
-from typing import Any, Awaitable, Callable, Protocol, runtime_checkable
+from typing import Awaitable, Callable, Protocol, runtime_checkable
 
 logger = logging.getLogger(__name__)
 
@@ -37,14 +37,14 @@ class RequestContext:
     """
 
     messages: list[dict[str, str]]
-    tools: list[Any] = field(default_factory=list)
+    tools: list[object] = field(default_factory=list)
     system_prompt: str | None = None
     model: str = "default"
     agent_name: str = ""
     task_type: str = ""
     task_id: str | None = None
     # 中间件间共享状态（压缩结果、token 用量、循环检测状态等）
-    metadata: dict[str, Any] = field(default_factory=dict)
+    metadata: dict[str, object] = field(default_factory=dict)
 
 
 @runtime_checkable
@@ -57,7 +57,7 @@ class Middleware(Protocol):
 
     async def before(self, ctx: RequestContext) -> RequestContext: ...
 
-    async def after(self, ctx: RequestContext, result: Any) -> Any: ...
+    async def after(self, ctx: RequestContext, result: object) -> object: ...
 
 
 class MiddlewareChain:
@@ -81,8 +81,8 @@ class MiddlewareChain:
     async def execute(
         self,
         ctx: RequestContext,
-        handler: Callable[[RequestContext], Awaitable[Any]],
-    ) -> Any:
+        handler: Callable[[RequestContext], Awaitable[object]],
+    ) -> object:
         """执行中间件链 + handler。
 
         洋葱模型：before 顺序执行 → handler → after 逆序执行。
@@ -125,7 +125,7 @@ class SummarizationMiddleware:
     after: 无操作（压缩在 before 完成）
     """
 
-    def __init__(self, compressor: Any = None) -> None:
+    def __init__(self, compressor: object | None = None) -> None:
         self._compressor = compressor
 
     async def before(self, ctx: RequestContext) -> RequestContext:
@@ -143,7 +143,7 @@ class SummarizationMiddleware:
                 logger.warning(f"SummarizationMiddleware: compression failed: {e}")
         return ctx
 
-    async def after(self, ctx: RequestContext, result: Any) -> Any:
+    async def after(self, ctx: RequestContext, result: object) -> object:
         return result
 
 
@@ -157,7 +157,7 @@ class TokenUsageMiddleware:
     async def before(self, ctx: RequestContext) -> RequestContext:
         return ctx
 
-    async def after(self, ctx: RequestContext, result: Any) -> Any:
+    async def after(self, ctx: RequestContext, result: object) -> object:
         # 从 ReActResult 或类似结构提取 token 用量
         # ReActResult 有 total_tokens 属性（非 token_usage）
         usage = getattr(result, "total_tokens", None)
@@ -184,7 +184,7 @@ class LoopDetectionMiddleware:
         ctx.metadata["loop_detection_window"] = []
         return ctx
 
-    async def after(self, ctx: RequestContext, result: Any) -> Any:
+    async def after(self, ctx: RequestContext, result: object) -> object:
         trajectory = getattr(result, "trajectory", None) or []
         if len(trajectory) < self._threshold:
             return result

src/agentkit/core/plan_executor.py

@@ -16,9 +16,9 @@ import asyncio
 import logging
 import random
 import time
-from dataclasses import dataclass, field
+from dataclasses import dataclass
 from enum import Enum
-from typing import Any, Callable, Awaitable
+from typing import Callable, Awaitable
 
 from agentkit.core.plan_schema import ExecutionPlan, PlanStep, PlanStepStatus
 from agentkit.core.protocol import TaskMessage, TaskResult, TaskStatus
@@ -42,7 +42,7 @@ class StepExecutionResult:
 
     step_id: str
     status: PlanStepStatus
-    result: dict[str, Any] | None = None
+    result: dict[str, object] | None = None
     error: str | None = None
     retry_count: int = 0
     duration_ms: float = 0.0
@@ -91,7 +91,7 @@ class PlanExecutor:
 
     def __init__(
         self,
-        agent_pool: Any,
+        agent_pool: object,
         max_retries: int = 2,
         step_timeout: float = 300.0,
         max_parallel: int = 5,
@@ -207,7 +207,7 @@ class PlanExecutor:
     async def _execute_step_with_retry(
         self,
         step: PlanStep,
-        input_data: dict[str, Any],
+        input_data: dict[str, object],
         original_task: TaskMessage,
     ) -> StepExecutionResult:
         """执行单个步骤，支持重试
@@ -281,9 +281,9 @@ class PlanExecutor:
     async def _execute_step_once(
         self,
         step: PlanStep,
-        input_data: dict[str, Any],
+        input_data: dict[str, object],
         original_task: TaskMessage,
-    ) -> dict[str, Any]:
+    ) -> dict[str, object]:
         """执行单个步骤一次
 
         通过 AgentPool 创建 Agent 执行步骤。
@@ -463,7 +463,7 @@ class PlanExecutor:
         self,
         step: PlanStep,
         step_results: dict[str, StepExecutionResult],
-    ) -> dict[str, Any]:
+    ) -> dict[str, object]:
         """将依赖步骤的结果注入到当前步骤的输入中
 
         兼容 Orchestrator 的 subtask_results 累积模式。
@@ -471,7 +471,7 @@ class PlanExecutor:
         enriched = dict(step.input_data)
 
         if step.dependencies:
-            dep_results: dict[str, dict[str, Any]] = {}
+            dep_results: dict[str, dict[str, object]] = {}
             for dep_id in step.dependencies:
                 if dep_id in step_results:
                     dep_result = step_results[dep_id]

src/agentkit/core/react.py

@@ -13,7 +13,7 @@ from collections import Counter, deque
 from collections.abc import AsyncGenerator
 from dataclasses import dataclass, field
 from datetime import datetime, timezone
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING, Awaitable, Callable
 
 from agentkit.core.exceptions import LLMProviderError, LoopDetectedError, TaskCancelledError, TaskTimeoutError
 from agentkit.core.protocol import CancellationToken
@@ -43,13 +43,13 @@ class ReActStep:
     step: int
     action: str  # "tool_call" or "final_answer"
     tool_name: str | None = None
-    arguments: dict[str, Any] | None = None
-    result: Any = None
+    arguments: dict[str, object] | None = None
+    result: object = None
     content: str | None = None
     tokens: int = 0
 
 
-async def _ensure_async_iterable(obj: Any, label: str = "<obj>"):
+async def _ensure_async_iterable(obj: object, label: str = "<obj>"):
     """Defensive helper: ensure the given object is an async iterable.
 
     Guards against the recurring ``'async for' requires an object with
@@ -133,7 +133,7 @@ class ReActEvent:
 
     event_type: str  # "thinking","token","tool_call","tool_result","confirmation_request","confirmation_result","phase_violation","step","final_answer","final_result","error"
     step: int
-    data: dict[str, Any] = field(default_factory=dict)
+    data: dict[str, object] = field(default_factory=dict)
     timestamp: str = field(default_factory=lambda: datetime.now(timezone.utc).isoformat())
 
 
@@ -230,7 +230,7 @@ class ReActEngine:
         # step and yields phase_violation ReActEvents. Non-streaming execute()
         # simply ignores the accumulator (the error dict returned to the LLM is
         # the only signal there).
-        self._phase_violations: list[dict[str, Any]] = []
+        self._phase_violations: list[dict[str, object]] = []
 
     def reset(self) -> None:
         """Reset internal state for reuse across conversations.
@@ -299,8 +299,8 @@ class ReActEngine:
         return False
 
     def _check_phase_permission(
-        self, tool_name: str, arguments: dict[str, Any]
-    ) -> dict[str, Any] | None:
+        self, tool_name: str, arguments: dict[str, object]
+    ) -> dict[str, object] | None:
         """Return None if tool is allowed; return a structured error dict if blocked.
 
         The error dict replaces what `_execute_tool` would have returned —
@@ -351,7 +351,7 @@ class ReActEngine:
                 return violation
         return None
 
-    def _check_tool_loop(self, tool_calls: list[Any]) -> str | None:
+    def _check_tool_loop(self, tool_calls: list[object]) -> str | None:
         """检测重复工具调用模式。
 
         将当前步的工具调用 hash 加入滑动窗口，若同一 hash 在窗口内出现
@@ -406,10 +406,10 @@ class ReActEngine:
         memory_retriever: "MemoryRetriever | None" = None,
         task_id: str | None = None,
         compressor: "CompressionStrategy | None" = None,
-        retrieval_config: dict[str, Any] | None = None,
+        retrieval_config: dict[str, object] | None = None,
         cancellation_token: CancellationToken | None = None,
         timeout_seconds: float | None = None,
-        confirmation_handler: Any | None = None,
+        confirmation_handler: Callable[..., Awaitable[object]] | None = None,
     ) -> ReActResult:
         """执行 ReAct 循环
 
@@ -536,7 +536,7 @@ class ReActEngine:
 
     async def _run_loop_and_extract(
         self,
-        **kwargs: Any,
+        **kwargs: object,
     ) -> ReActResult:
         """Collect all events from _execute_loop and extract the final ReActResult.
 
@@ -564,9 +564,9 @@ class ReActEngine:
         memory_retriever: "MemoryRetriever | None" = None,
         task_id: str | None = None,
         compressor: "CompressionStrategy | None" = None,
-        retrieval_config: dict[str, Any] | None = None,
+        retrieval_config: dict[str, object] | None = None,
         cancellation_token: CancellationToken | None = None,
-        confirmation_handler: Any | None = None,
+        confirmation_handler: Callable[..., Awaitable[object]] | None = None,
         stream: bool = False,
         effective_timeout: float = 0.0,
     ) -> AsyncGenerator[ReActEvent, None]:
@@ -666,7 +666,7 @@ class ReActEngine:
                     )
 
             # 构建初始消息
-            conversation: list[dict[str, Any]] = []
+            conversation: list[dict[str, object]] = []
             system_content = self._build_system_message(
                 stable=system_prompt or "",
                 volatile=memory_context,
@@ -726,7 +726,7 @@ class ReActEngine:
                     # 流式模式：用 chat_stream，yield token events
                     stream_content_chunks: list[str] = []
                     stream_usage = None
-                    stream_tool_calls: list[Any] = []
+                    stream_tool_calls: list[object] = []
                     stream_model = model
                     # U3/G8: delta_flush 节流 buffer
                     _flush_buffer: list[str] = []
@@ -840,7 +840,7 @@ class ReActEngine:
                         )
 
                     # Act: 记录 assistant 消息（含 tool_calls）到对话历史
-                    assistant_msg: dict[str, Any] = {
+                    assistant_msg: dict[str, object] = {
                         "role": "assistant",
                         "content": response.content or "",
                         "tool_calls": [
@@ -874,7 +874,7 @@ class ReActEngine:
                             if i in parallelizable_set
                         ]
 
-                        all_results: list[Any] = [None] * len(response.tool_calls)
+                        all_results: list[object] = [None] * len(response.tool_calls)
 
                         # Execute serial tools first (handles confirmation flow)
                         for i, tc in serial_calls:
@@ -1452,10 +1452,10 @@ class ReActEngine:
         memory_retriever: "MemoryRetriever | None" = None,
         task_id: str | None = None,
         compressor: "CompressionStrategy | None" = None,
-        retrieval_config: dict[str, Any] | None = None,
+        retrieval_config: dict[str, object] | None = None,
         cancellation_token: CancellationToken | None = None,
         timeout_seconds: float | None = None,
-        confirmation_handler: Any | None = None,
+        confirmation_handler: Callable[..., Awaitable[object]] | None = None,
     ) -> AsyncGenerator[ReActEvent, None]:
         """Execute ReAct loop, yielding ReActEvent objects.
 
@@ -1514,7 +1514,7 @@ class ReActEngine:
         volatile: str,
         *,
         model: str,
-    ) -> str | list[dict[str, Any]] | None:
+    ) -> str | list[dict[str, object]] | None:
         """构建双块结构 system message(stable + volatile)。
 
         - prompt_cache_enable=False 或无 stable+volatile → 返回 str(或 None)
@@ -1535,7 +1535,7 @@ class ReActEngine:
 
         provider_name = self._get_provider_name(model)
         if provider_name == "anthropic":
-            blocks: list[dict[str, Any]] = []
+            blocks: list[dict[str, object]] = []
             if stable:
                 blocks.append(
                     {
@@ -1675,8 +1675,8 @@ class ReActEngine:
     @staticmethod
     def _build_response_from_stream(
         content: str,
-        tool_calls: list[Any],
-        usage: Any,
+        tool_calls: list[object],
+        usage: object,
         model: str,
     ) -> LLMResponse:
         """Build an LLMResponse from accumulated stream chunks."""
@@ -1723,7 +1723,7 @@ class ReActEngine:
     async def _build_tool_result_message(
         self,
         tool_call_id: str,
-        result: Any,
+        result: object,
         compressor: "CompressionStrategy | None" = None,
         tool_name: str | None = None,
     ) -> dict:
@@ -1742,7 +1742,7 @@ class ReActEngine:
         }
 
     async def _execute_tool(
-        self, tool_name: str, arguments: dict[str, Any], tools: list[Tool]
+        self, tool_name: str, arguments: dict[str, object], tools: list[Tool]
     ) -> dict:
         """执行工具调用，处理成功和失败情况"""
         # U3/G6: phase enforcement — check before dispatch. If the tool is
@@ -1788,11 +1788,11 @@ class ReActEngine:
 
     async def _execute_tool_with_confirmation(
         self,
-        tc: Any,
+        tc: object,
         tools: list[Tool],
         step: int,
-        confirmation_handler: Any,
-    ) -> tuple[Any, list[ReActEvent]]:
+        confirmation_handler: Callable[..., Awaitable[object]] | None,
+    ) -> tuple[object, list[ReActEvent]]:
         """Execute a tool call with confirmation flow support.
 
         Used in the parallel execution path for serial (non-parallelizable) tools
@@ -1886,7 +1886,7 @@ class ReActEngine:
 
         return tool_result, events
 
-    def _should_execute_parallel(self, tool_calls: list[Any]) -> bool:
+    def _should_execute_parallel(self, tool_calls: list[object]) -> bool:
         """Determine if tool calls should be executed in parallel.
 
         - parallel_tools=True: always parallel (if >1 tool)
@@ -1905,7 +1905,7 @@ class ReActEngine:
             return len(parallelizable_indices) > 1
         return False
 
-    def _get_parallelizable_indices(self, tool_calls: list[Any]) -> list[int]:
+    def _get_parallelizable_indices(self, tool_calls: list[object]) -> list[int]:
         """Get indices of tool_calls that have _parallelizable=true in arguments.
 
         LLM marks parallelizable tools by including _parallelizable: true
@@ -1918,7 +1918,7 @@ class ReActEngine:
                 indices.append(i)
         return indices
 
-    def _parse_text_tool_calls(self, content: str) -> list[dict[str, Any]]:
+    def _parse_text_tool_calls(self, content: str) -> list[dict[str, object]]:
         """从文本中解析工具调用模式
 
         支持格式：
@@ -1926,7 +1926,7 @@ class ReActEngine:
         2. ```tool\n{"name": "...", "arguments": {...}}\n```
         3. <tool_use>\n{"name": "...", "arguments": {...}}\n</tool_use>
         """
-        calls: list[dict[str, Any]] = []
+        calls: list[dict[str, object]] = []
 
         # 格式 1: Action: tool_name(args)
         action_pattern = re.compile(r"Action:\s*(\w+)\((.+?)\)", re.DOTALL)
@@ -1999,7 +1999,7 @@ class ReActEngine:
         return calls
 
     @staticmethod
-    def _extract_tool_call_from_malformed(text: str) -> dict[str, Any] | None:
+    def _extract_tool_call_from_malformed(text: str) -> dict[str, object] | None:
         """从畸形文本中尝试提取工具调用。
 
         处理场景：
@@ -2066,7 +2066,7 @@ class ReActEngine:
             return None
         name = name_match.group(1)
 
-        arguments: dict[str, Any] = {}
+        arguments: dict[str, object] = {}
         # 提取 "key": "value" 模式
         for kv_match in re.finditer(r'"(\w+)"\s*:\s*"([^"]*)"', text):
             key = kv_match.group(1)

src/agentkit/core/rewoo.py

@@ -11,7 +11,7 @@ import logging
 import re
 import time
 from dataclasses import dataclass, field
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING, Awaitable, Callable
 
 from agentkit.core.exceptions import LLMProviderError, TaskCancelledError, TaskTimeoutError
 from agentkit.core.protocol import CancellationToken
@@ -52,7 +52,7 @@ class ReWOOPlanStep:
 
     step_id: int
     tool_name: str
-    arguments: dict[str, Any]
+    arguments: dict[str, object]
     reasoning: str = ""
 
 
@@ -164,10 +164,10 @@ class ReWOOEngine:
         memory_retriever: "MemoryRetriever | None" = None,
         task_id: str | None = None,
         compressor: "CompressionStrategy | None" = None,
-        retrieval_config: dict[str, Any] | None = None,
+        retrieval_config: dict[str, object] | None = None,
         cancellation_token: CancellationToken | None = None,
         timeout_seconds: float | None = None,
-        confirmation_handler: Any | None = None,
+        confirmation_handler: Callable[..., Awaitable[object]] | None = None,
     ) -> ReActResult:
         """执行 ReWOO 三阶段流程
 
@@ -241,9 +241,9 @@ class ReWOOEngine:
         memory_retriever: "MemoryRetriever | None" = None,
         task_id: str | None = None,
         compressor: "CompressionStrategy | None" = None,
-        retrieval_config: dict[str, Any] | None = None,
+        retrieval_config: dict[str, object] | None = None,
         cancellation_token: CancellationToken | None = None,
-        confirmation_handler: Any | None = None,
+        confirmation_handler: Callable[..., Awaitable[object]] | None = None,
     ) -> ReActResult:
         tools = tools or []
         tool_schemas = self._build_tool_schemas(tools) if tools else None
@@ -350,7 +350,7 @@ class ReWOOEngine:
 
             # 如果计划为空（无需工具），直接让 LLM 回答
             if not plan.steps:
-                llm_messages: list[dict[str, Any]] = []
+                llm_messages: list[dict[str, object]] = []
                 if effective_system_prompt:
                     llm_messages.append({"role": "system", "content": effective_system_prompt})
                 llm_messages.extend(messages)
@@ -399,7 +399,7 @@ class ReWOOEngine:
                 )
 
             # ── Phase 2: Execution ──
-            tool_results: list[dict[str, Any]] = []
+            tool_results: list[dict[str, object]] = []
             for plan_step in plan.steps:
                 # 协作式取消检查
                 if cancellation_token is not None:
@@ -524,10 +524,10 @@ class ReWOOEngine:
         memory_retriever: "MemoryRetriever | None" = None,
         task_id: str | None = None,
         compressor: "CompressionStrategy | None" = None,
-        retrieval_config: dict[str, Any] | None = None,
+        retrieval_config: dict[str, object] | None = None,
         cancellation_token: CancellationToken | None = None,
         timeout_seconds: float | None = None,
-        confirmation_handler: Any | None = None,
+        confirmation_handler: Callable[..., Awaitable[object]] | None = None,
     ):
         """Execute ReWOO flow, yielding ReActEvent objects.
 
@@ -637,7 +637,7 @@ class ReWOOEngine:
 
             # Empty plan: direct answer
             if not plan.steps:
-                llm_messages: list[dict[str, Any]] = []
+                llm_messages: list[dict[str, object]] = []
                 if effective_system_prompt:
                     llm_messages.append({"role": "system", "content": effective_system_prompt})
                 llm_messages.extend(messages)
@@ -676,7 +676,7 @@ class ReWOOEngine:
                 return
 
             # ── Phase 2: Execution ──
-            tool_results: list[dict[str, Any]] = []
+            tool_results: list[dict[str, object]] = []
             for plan_step in plan.steps:
                 if cancellation_token is not None:
                     cancellation_token.check()
@@ -806,10 +806,10 @@ class ReWOOEngine:
         memory_retriever: "MemoryRetriever | None" = None,
         task_id: str | None = None,
         compressor: "CompressionStrategy | None" = None,
-        retrieval_config: dict[str, Any] | None = None,
+        retrieval_config: dict[str, object] | None = None,
         cancellation_token: CancellationToken | None = None,
         total_tokens: int = 0,
-        confirmation_handler: Any | None = None,
+        confirmation_handler: Callable[..., Awaitable[object]] | None = None,
     ):
         """Stream version: try fallback strategies in configured order, yielding events from the first successful one.
 
@@ -902,7 +902,7 @@ class ReWOOEngine:
                     "reasoning": plan.reasoning,
                     "steps": [{"step_id": s.step_id, "tool_name": s.tool_name, "arguments": s.arguments, "reasoning": s.reasoning} for s in plan.steps],
                 })
-                tool_results: list[dict[str, Any]] = []
+                tool_results: list[dict[str, object]] = []
                 for plan_step in plan.steps:
                     if cancellation_token is not None:
                         cancellation_token.check()
@@ -934,9 +934,9 @@ class ReWOOEngine:
         memory_retriever: "MemoryRetriever | None" = None,
         task_id: str | None = None,
         compressor: "CompressionStrategy | None" = None,
-        retrieval_config: dict[str, Any] | None = None,
+        retrieval_config: dict[str, object] | None = None,
         cancellation_token: CancellationToken | None = None,
-        confirmation_handler: Any | None = None,
+        confirmation_handler: Callable[..., Awaitable[object]] | None = None,
     ):
         """Stream: ReAct fallback"""
         logger.warning("ReWOO planning failed in stream mode, falling back to ReActEngine")
@@ -969,7 +969,7 @@ class ReWOOEngine:
         """Stream: Direct LLM fallback"""
         logger.warning("Falling back to direct LLM call in stream mode")
         try:
-            direct_messages: list[dict[str, Any]] = []
+            direct_messages: list[dict[str, object]] = []
             if effective_system_prompt:
                 direct_messages.append({"role": "system", "content": effective_system_prompt})
             direct_messages.extend(messages)
@@ -1012,7 +1012,7 @@ class ReWOOEngine:
                     "reasoning": plan.reasoning,
                     "steps": [{"step_id": s.step_id, "tool_name": s.tool_name, "arguments": s.arguments, "reasoning": s.reasoning} for s in plan.steps],
                 })
-                tool_results: list[dict[str, Any]] = []
+                tool_results: list[dict[str, object]] = []
                 for plan_step in plan.steps:
                     if cancellation_token is not None:
                         cancellation_token.check()
@@ -1043,11 +1043,11 @@ class ReWOOEngine:
         memory_retriever: "MemoryRetriever | None" = None,
         task_id: str | None = None,
         compressor: "CompressionStrategy | None" = None,
-        retrieval_config: dict[str, Any] | None = None,
+        retrieval_config: dict[str, object] | None = None,
         cancellation_token: CancellationToken | None = None,
         trajectory: list[ReActStep] | None = None,
         total_tokens: int = 0,
-        confirmation_handler: Any | None = None,
+        confirmation_handler: Callable[..., Awaitable[object]] | None = None,
     ) -> ReActResult | None:
         """按配置的 fallback 策略顺序尝试回退，返回第一个成功的结果
 
@@ -1136,7 +1136,7 @@ class ReWOOEngine:
                 # Execute the simplified plan
                 trajectory: list[ReActStep] = []
                 total_tokens = simplified_tokens
-                tool_results: list[dict[str, Any]] = []
+                tool_results: list[dict[str, object]] = []
                 for plan_step in plan.steps:
                     if cancellation_token is not None:
                         cancellation_token.check()
@@ -1195,9 +1195,9 @@ class ReWOOEngine:
         memory_retriever: "MemoryRetriever | None" = None,
         task_id: str | None = None,
         compressor: "CompressionStrategy | None" = None,
-        retrieval_config: dict[str, Any] | None = None,
+        retrieval_config: dict[str, object] | None = None,
         cancellation_token: CancellationToken | None = None,
-        confirmation_handler: Any | None = None,
+        confirmation_handler: Callable[..., Awaitable[object]] | None = None,
     ) -> ReActResult | None:
         """ReAct fallback: delegate to ReActEngine"""
         logger.warning("ReWOO planning failed, falling back to ReActEngine")
@@ -1240,7 +1240,7 @@ class ReWOOEngine:
         """Direct fallback: simple LLM call without tools"""
         logger.warning("Falling back to direct LLM call")
         try:
-            direct_messages: list[dict[str, Any]] = []
+            direct_messages: list[dict[str, object]] = []
             if effective_system_prompt:
                 direct_messages.append({"role": "system", "content": effective_system_prompt})
             direct_messages.extend(messages)
@@ -1319,7 +1319,7 @@ class ReWOOEngine:
             if plan is not None and plan.steps:
                 trajectory: list[ReActStep] = []
                 total_tokens = plan_tokens
-                tool_results: list[dict[str, Any]] = []
+                tool_results: list[dict[str, object]] = []
                 for plan_step in plan.steps:
                     if cancellation_token is not None:
                         cancellation_token.check()
@@ -1396,7 +1396,7 @@ class ReWOOEngine:
         tool_descriptions = self._build_tool_descriptions(tools)
 
         # 构建规划消息
-        planning_messages: list[dict[str, Any]] = [
+        planning_messages: list[dict[str, object]] = [
             {"role": "system", "content": _PLANNING_SYSTEM_PROMPT},
         ]
 
@@ -1451,7 +1451,7 @@ class ReWOOEngine:
     async def _synthesis_phase(
         self,
         messages: list[dict[str, str]],
-        tool_results: list[dict[str, Any]],
+        tool_results: list[dict[str, object]],
         model: str,
         agent_name: str,
         task_type: str,
@@ -1468,7 +1468,7 @@ class ReWOOEngine:
             cancellation_token.check()
 
         # 构建综合消息
-        synthesis_messages: list[dict[str, Any]] = [
+        synthesis_messages: list[dict[str, object]] = [
             {"role": "system", "content": _SYNTHESIS_SYSTEM_PROMPT},
         ]
 
@@ -1600,7 +1600,7 @@ class ReWOOEngine:
         return None
 
     async def _execute_tool(
-        self, tool_name: str, arguments: dict[str, Any], tools: list[Tool]
+        self, tool_name: str, arguments: dict[str, object], tools: list[Tool]
     ) -> dict:
         """执行工具调用，处理成功和失败情况"""
         tool = self._find_tool(tool_name, tools)

src/agentkit/evolution/ab_tester.py

@@ -8,6 +8,8 @@ import math
 from dataclasses import dataclass
 from typing import TYPE_CHECKING
 
+from sqlalchemy.exc import DBAPIError
+
 if TYPE_CHECKING:
     from agentkit.evolution.evolution_store import InMemoryEvolutionStore
 
@@ -129,7 +131,7 @@ class ABTester:
                 sample_count=len(experiment_metrics),
             )
             logger.info(f"A/B test results persisted for test '{test_id}'")
-        except Exception as e:
+        except (DBAPIError, RuntimeError, ValueError, KeyError) as e:
             logger.error(f"Failed to persist A/B test results: {e}")
 
     async def evaluate(self, test_id: str) -> ABTestResult | None:

src/agentkit/evolution/experience_store.py

@@ -10,6 +10,7 @@
 
 from __future__ import annotations
 
+import asyncio
 import logging
 import math
 import re
@@ -18,6 +19,7 @@ from datetime import datetime, timedelta, timezone
 from typing import Any
 
 from sqlalchemy import text
+from sqlalchemy.exc import DBAPIError
 
 _SAFE_TABLE_NAME_PATTERN = re.compile(r'^[a-zA-Z_][a-zA-Z0-9_]*$')
 
@@ -96,7 +98,7 @@ class ExperienceStore:
             text = experience.text_for_embedding()
             try:
                 experience.embedding = await self._embedder.embed(text)
-            except Exception as e:
+            except (ConnectionError, RuntimeError, asyncio.TimeoutError, ValueError) as e:
                 logger.warning(f"Failed to generate embedding for experience {experience.experience_id}: {e}")
 
         async with self._session_factory() as db:
@@ -122,7 +124,7 @@ class ExperienceStore:
                     f"task_type={experience.task_type} outcome={experience.outcome}"
                 )
                 return experience.experience_id
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 await db.rollback()
                 logger.error(f"Failed to record experience: {e}")
                 raise
@@ -149,7 +151,7 @@ class ExperienceStore:
                 if self._pgvector_enabled and self._embedder:
                     return await self._search_pgvector(db, query, top_k, task_type, search_multiplier)
                 return await self._search_client_side(db, query, top_k, task_type, search_multiplier)
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 logger.error(f"Failed to search experiences: {e}")
                 return []
 
@@ -343,7 +345,7 @@ class ExperienceStore:
                         )
                     )
                 return metrics_list
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 logger.error(f"Failed to get metrics: {e}")
                 return []
 
@@ -375,7 +377,7 @@ class InMemoryExperienceStore:
             text = experience.text_for_embedding()
             try:
                 experience.embedding = await self._embedder.embed(text)
-            except Exception as e:
+            except (ConnectionError, RuntimeError, asyncio.TimeoutError, ValueError) as e:
                 logger.warning(f"Failed to generate embedding for experience {experience.experience_id}: {e}")
 
         # 存储副本，避免外部修改影响内部状态
@@ -411,7 +413,7 @@ class InMemoryExperienceStore:
         if self._embedder:
             try:
                 query_embedding = await self._embedder.embed(query)
-            except Exception as e:
+            except (ConnectionError, RuntimeError, asyncio.TimeoutError, ValueError) as e:
                 logger.warning(f"Failed to generate query embedding: {e}")
 
         # 筛选候选

src/agentkit/evolution/genetic.py

@@ -13,6 +13,7 @@
 
 from __future__ import annotations
 
+import asyncio
 import copy
 import logging
 import random
@@ -292,7 +293,7 @@ class MutationOperator:
                     model="default",
                 )
                 return response.content.strip() or instructions
-            except Exception as e:
+            except (ConnectionError, RuntimeError, asyncio.TimeoutError, ValueError) as e:
                 logger.warning(f"LLM instruction mutation failed: {e}")
 
         # Fallback: simple text mutation (shuffle paragraphs)

src/agentkit/evolution/lifecycle.py

@@ -8,6 +8,8 @@ from dataclasses import dataclass, field
 from datetime import datetime, timezone
 from typing import Any
 
+from sqlalchemy.exc import DBAPIError
+
 from agentkit.core.protocol import EvolutionEvent, TaskMessage, TaskResult
 from agentkit.evolution.ab_tester import ABTestConfig, ABTestResult, ABTester
 from agentkit.evolution.evolution_store import EvolutionStore
@@ -371,7 +373,7 @@ class EvolutionMixin:
                     entry.event_id = event_id
                     break
             return True
-        except Exception as e:
+        except (DBAPIError, RuntimeError, ValueError, KeyError) as e:
             logger.error(f"Failed to apply evolution change: {e}")
             return False
 
@@ -382,7 +384,7 @@ class EvolutionMixin:
 
         try:
             return await self._evolution_store.rollback(log_entry.event_id)
-        except Exception as e:
+        except (DBAPIError, RuntimeError, ValueError, KeyError) as e:
             logger.error(f"Failed to rollback evolution change: {e}")
             return False
 

src/agentkit/evolution/llm_reflector.py

@@ -3,6 +3,7 @@
 通过 LLM 分析执行轨迹生成结构化反思，比 RuleBasedReflector 提供更深入的洞察。
 """
 
+import asyncio
 import json
 import logging
 import re
@@ -62,7 +63,7 @@ class LLMReflector:
                 task_type="reflection",
             )
             return self._parse_reflection_response(response.content, task, result)
-        except Exception as e:
+        except (ConnectionError, RuntimeError, asyncio.TimeoutError, ValueError, KeyError) as e:
             logger.warning(f"LLM reflection failed, returning default: {e}")
             return Reflection(
                 task_id=getattr(task, "task_id", "unknown"),

src/agentkit/evolution/pg_store.py

@@ -13,6 +13,7 @@ from typing import Any
 from sqlalchemy import select
 from sqlalchemy.dialects.postgresql import JSONB
 from sqlalchemy import Column, DateTime, Float, Integer, String, Text, UniqueConstraint
+from sqlalchemy.exc import DBAPIError
 from sqlalchemy.orm import declarative_base
 
 from agentkit.core.protocol import EvolutionEvent
@@ -154,7 +155,7 @@ class PostgreSQLEvolutionStore:
                 event.event_id = event_id
                 logger.info(f"Evolution event recorded: {event_id} for agent '{event.agent_name}'")
                 return event_id
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 await db.rollback()
                 logger.error(f"Failed to record evolution event: {e}")
                 raise
@@ -178,7 +179,7 @@ class PostgreSQLEvolutionStore:
                 await db.commit()
                 logger.info(f"Evolution event {event_id} rolled back")
                 return True
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 await db.rollback()
                 logger.error(f"Failed to rollback evolution event {event_id}: {e}")
                 return False
@@ -218,7 +219,7 @@ class PostgreSQLEvolutionStore:
                     }
                     for e in entries
                 ]
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 logger.error(f"Failed to list evolution events: {e}")
                 return []
 
@@ -242,7 +243,7 @@ class PostgreSQLEvolutionStore:
                 db.add(entry)
                 await db.commit()
                 return vid
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 await db.rollback()
                 logger.error(f"Failed to record skill version: {e}")
                 raise
@@ -271,7 +272,7 @@ class PostgreSQLEvolutionStore:
                     }
                     for e in entries
                 ]
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 logger.error(f"Failed to list skill versions: {e}")
                 return []
 
@@ -295,7 +296,7 @@ class PostgreSQLEvolutionStore:
                 db.add(entry)
                 await db.commit()
                 return rid
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 await db.rollback()
                 logger.error(f"Failed to record A/B test result: {e}")
                 raise
@@ -324,6 +325,6 @@ class PostgreSQLEvolutionStore:
                     }
                     for e in entries
                 ]
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 logger.error(f"Failed to get A/B test results: {e}")
                 return []

src/agentkit/evolution/pitfall_detector.py

@@ -151,7 +151,7 @@ class PitfallDetector:
                 task_type=task_type,
             )
             return results
-        except Exception as e:
+        except (RuntimeError, ValueError, KeyError) as e:
             logger.error(f"Failed to search experiences for pitfall detection: {e}")
             return []
 

src/agentkit/evolution/prompt_optimizer.py

@@ -10,6 +10,7 @@
 - LLMPromptOptimizer: 基于 LLM 分析反思结果生成改进指令
 """
 
+import asyncio
 import logging
 from dataclasses import dataclass, field
 from typing import Any
@@ -202,7 +203,7 @@ class LLMPromptOptimizer:
         """
         try:
             optimized_instruction = await self._llm_optimize_instruction(module, trace, reflection)
-        except Exception as e:
+        except (ConnectionError, RuntimeError, asyncio.TimeoutError, ValueError) as e:
             logger.warning(f"LLM prompt optimization failed, falling back to bootstrap: {e}")
             return await self._bootstrap.optimize(module)
 

src/agentkit/evolution/risk_guard_learner.py

@@ -8,6 +8,7 @@
 
 from __future__ import annotations
 
+import asyncio
 import json
 import logging
 import re
@@ -79,7 +80,7 @@ class RiskGuardLearner:
                 top_k=top_k,
                 task_type=skill_name,
             )
-        except Exception as e:
+        except (ConnectionError, RuntimeError, asyncio.TimeoutError, ValueError, KeyError) as e:
             logger.warning(f"RiskGuardLearner: failed to search experiences: {e}")
             return []
 
@@ -95,7 +96,7 @@ class RiskGuardLearner:
         # 2. 构建 LLM prompt
         try:
             prompt = self._build_prompt(failures)
-        except Exception as e:
+        except (ValueError, KeyError, RuntimeError) as e:
             logger.warning(f"RiskGuardLearner: failed to build prompt: {e}")
             return []
 
@@ -117,14 +118,14 @@ class RiskGuardLearner:
                 agent_name="risk_guard_learner",
                 task_type="risk_guard_learning",
             )
-        except Exception as e:
+        except (ConnectionError, RuntimeError, asyncio.TimeoutError, ValueError) as e:
             logger.warning(f"RiskGuardLearner: LLM call failed: {e}")
             return []
 
         # 4. 解析响应
         try:
             return self._parse_response(response.content, source_ids)
-        except Exception as e:
+        except (ValueError, KeyError, TypeError) as e:
             logger.warning(f"RiskGuardLearner: failed to parse response: {e}")
             return []
 

src/agentkit/experts/_debate_runner.py

@@ -9,7 +9,7 @@ import asyncio
 import json
 import logging
 import re
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING
 
 from .expert import Expert
 from .plan import PhaseStatus, PlanPhase, TeamPlan
@@ -28,7 +28,7 @@ class DebateRunnerMixin:
     _phase_semaphore: asyncio.Semaphore
     MAX_DEBATE_ROUNDS: int
 
-    async def _execute_debate_phase(self, phase: PlanPhase, plan: TeamPlan) -> dict[str, Any]:
+    async def _execute_debate_phase(self, phase: PlanPhase, plan: TeamPlan) -> dict[str, object]:
         """Execute a DEBATE phase: Lead-facilitated structured debate (5 stages).
         Parse config → Lead opens → experts argue in parallel rounds → Lead
         summarizes → Lead adjudicates → write conclusion to workspace."""
@@ -86,7 +86,7 @@ class DebateRunnerMixin:
         )
 
         # Debate history for context (Lead opening + expert arguments + Lead summaries)
-        history: list[dict[str, Any]] = [
+        history: list[dict[str, object]] = [
             {"expert": lead.config.name, "content": opening, "round": 0, "role": "moderator"}
         ]
 
@@ -103,7 +103,7 @@ class DebateRunnerMixin:
                 break
 
             # Experts argue in parallel (with concurrency limit)
-            async def _bounded_debate(e: Any) -> str:
+            async def _bounded_debate(e: object) -> str:
                 async with self._phase_semaphore:
                     return await self._generate_debate_argument(e, topic, history, round_num)
 
@@ -234,7 +234,7 @@ class DebateRunnerMixin:
             return f"辩论主题：{topic}。请各位专家发表看法。"
 
     async def _generate_debate_argument(
-        self, expert: Expert, topic: str, history: list[dict[str, Any]], round_num: int
+        self, expert: Expert, topic: str, history: list[dict[str, object]], round_num: int
     ) -> str:
         """Generate an expert's debate argument for the current round."""
         gateway = self._get_llm_gateway(expert)
@@ -269,7 +269,7 @@ class DebateRunnerMixin:
         return response.content.strip()
 
     async def _generate_debate_summary(
-        self, lead: Expert, topic: str, history: list[dict[str, Any]], round_num: int
+        self, lead: Expert, topic: str, history: list[dict[str, object]], round_num: int
     ) -> str:
         """Generate Lead's summary of the current debate round."""
         gateway = self._get_llm_gateway(lead)
@@ -307,8 +307,8 @@ class DebateRunnerMixin:
             return f"[第 {round_num} 轮辩论完成，小结生成失败]"
 
     async def _generate_debate_verdict(
-        self, lead: Expert, topic: str, history: list[dict[str, Any]]
-    ) -> dict[str, Any]:
+        self, lead: Expert, topic: str, history: list[dict[str, object]]
+    ) -> dict[str, object]:
         """Generate Lead's final verdict for the debate."""
         gateway = self._get_llm_gateway(lead)
         if not gateway:
@@ -371,7 +371,7 @@ class DebateRunnerMixin:
                 "conclusion": f"辩论主题：{topic}。裁决生成失败，建议参考辩论历史自行判断。",
             }
 
-    def _format_debate_history(self, history: list[dict[str, Any]]) -> str:
+    def _format_debate_history(self, history: list[dict[str, object]]) -> str:
         """Format debate history as readable text for LLM prompts."""
         if not history:
             return ""

src/agentkit/experts/_divergence_detector.py

@@ -6,7 +6,7 @@
 from __future__ import annotations
 
 import logging
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING
 
 from .expert import Expert
 from .plan import PhaseStatus, PhaseType, PlanPhase, TeamPlan
@@ -23,7 +23,7 @@ class DivergenceDetectorMixin:
     # Shared state provided by TeamOrchestrator (annotations only)
     _team: ExpertTeam
     _debate_count: int
-    _checkpoint: Any
+    _checkpoint: object
     MAX_DEBATES: int
 
     async def _maybe_add_plan_review_debate(self, lead: Expert, plan: TeamPlan, task: str) -> None:

src/agentkit/experts/_phase_executor.py

@@ -9,7 +9,7 @@ import asyncio
 import copy
 import logging
 from datetime import datetime, timezone
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING
 
 from agentkit.core.config_driven import ConfigDrivenAgent
 from agentkit.core.protocol import TaskMessage, TaskResult, TaskStatus
@@ -37,7 +37,7 @@ class PhaseExecutorMixin:
     # U4: State offloading helpers — keep memory lean for long-horizon runs.
     _OFFLOAD_SUMMARY_LIMIT = 500
 
-    def _offload_result(self, content: str, ref_key: str) -> dict[str, Any]:
+    def _offload_result(self, content: str, ref_key: str) -> dict[str, object]:
         """Create an offloaded result: summary in memory, full content in workspace."""
         if not isinstance(content, str):
             content = str(content) if content is not None else ""
@@ -64,17 +64,17 @@ class PhaseExecutorMixin:
                     logger.warning(f"Failed to read offloaded output '{ref_key}': {e}")
         return content
 
-    async def _execute_phase(self, phase: PlanPhase, plan: TeamPlan) -> dict[str, Any]:
+    async def _execute_phase(self, phase: PlanPhase, plan: TeamPlan) -> dict[str, object]:
         """Execute a single phase, dispatching by phase_type."""
         if phase.phase_type == PhaseType.DEBATE:
             return await self._execute_debate_phase(phase, plan)
         return await self._execute_execution_phase(phase, plan)
 
-    async def _execute_execution_phase(self, phase: PlanPhase, plan: TeamPlan) -> dict[str, Any]:
+    async def _execute_execution_phase(self, phase: PlanPhase, plan: TeamPlan) -> dict[str, object]:
         """Execute a standard EXECUTION phase. Split into 3 sub-methods (U2, KTD3 isolation)."""
         expert, agent, lead = await self._prepare_phase_context(phase, plan)
         last_error: str | None = None
-        result: dict[str, Any] | None = None
+        result: dict[str, object] | None = None
 
         try:
             # U3: 返工循环 — 最多 MAX_REWORKS + 1 次（1 次初始 + MAX_REWORKS 次返工）
@@ -135,11 +135,11 @@ class PhaseExecutorMixin:
         self,
         expert: Expert,
         phase: PlanPhase,
-        dependency_outputs: dict[str, Any],
+        dependency_outputs: dict[str, object],
         collaboration_outputs: dict[str, str],
     ) -> TaskMessage:
         """Build TaskMessage for execution with context isolation."""
-        input_data: dict[str, Any] = {
+        input_data: dict[str, object] = {
             "task": phase.task_description,
             "team_id": self._team.team_id,
             "phase_id": phase.id,
@@ -180,12 +180,12 @@ class PhaseExecutorMixin:
         lead: Expert,
         phase: PlanPhase,
         plan: TeamPlan,
-    ) -> tuple[dict[str, Any], str | None, bool, str, bool]:
+    ) -> tuple[dict[str, object], str | None, bool, str, bool]:
         """Run one rework iteration: read deps, build input, execute, review. Returns
         (result, last_error, passed, feedback, degraded). Raises RuntimeError on retry
         exhaustion."""
         # 每次迭代重新读取依赖输出（前置阶段可能在返工期间完成）
-        dependency_outputs: dict[str, Any] = {}
+        dependency_outputs: dict[str, object] = {}
         for dep_id in phase.depends_on:
             dep_phase = plan.get_phase(dep_id)
             if dep_phase and dep_phase.status == PhaseStatus.COMPLETED and dep_phase.result:
@@ -216,7 +216,7 @@ class PhaseExecutorMixin:
 
         # 执行专家任务（带重试，MAX_RETRIES 处理瞬时失败）
         last_error: str | None = None
-        result: dict[str, Any] | None = None
+        result: dict[str, object] | None = None
         for attempt in range(self.MAX_RETRIES + 1):
             try:
                 task_result: TaskResult = await agent.execute(task_msg)
@@ -265,7 +265,7 @@ class PhaseExecutorMixin:
         lead: Expert,
         phase: PlanPhase,
         plan: TeamPlan,
-        result: dict[str, Any],
+        result: dict[str, object],
         passed: bool,
         feedback: str,
         degraded: bool = False,

src/agentkit/experts/_review_gate.py

@@ -10,7 +10,6 @@ import json
 import logging
 import re
 from dataclasses import dataclass
-from typing import Any
 
 from agentkit.core.exceptions import LLMProviderError
 
@@ -45,7 +44,7 @@ class ReviewGateMixin:
     """Mixin: Lead 验收阶段输出质量 + 解析风险标记。由 TeamOrchestrator 组合。"""
 
     async def _review_phase_output(
-        self, lead: Expert, phase: PlanPhase, result: dict[str, Any]
+        self, lead: Expert, phase: PlanPhase, result: dict[str, object]
     ) -> ReviewResult:
         """Lead 验收阶段输出质量。
 
@@ -93,7 +92,7 @@ class ReviewGateMixin:
             )
 
         # P2: 优先尝试直接解析整个响应为 JSON，避免贪婪正则匹配过多
-        review: dict[str, Any] | None = None
+        review: dict[str, object] | None = None
         try:
             review = json.loads(response.content)
         except (json.JSONDecodeError, TypeError):

src/agentkit/experts/_rollback_handler.py

@@ -6,7 +6,7 @@
 from __future__ import annotations
 
 import logging
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING
 
 from agentkit.orchestrator.rollback import RollbackExecutor
 
@@ -27,7 +27,7 @@ class RollbackHandlerMixin:
     _rollback_timeout: float
 
     async def _mark_dependents_failed(
-        self, failed_phase_id: str, plan: TeamPlan, phase_results: dict[str, dict[str, Any]]
+        self, failed_phase_id: str, plan: TeamPlan, phase_results: dict[str, dict[str, object]]
     ) -> None:
         """Mark all phases that depend on the failed phase as FAILED."""
         for ph in plan.phases:

src/agentkit/experts/_synthesizer.py

@@ -7,7 +7,7 @@ from __future__ import annotations
 
 import logging
 from datetime import datetime, timezone
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING
 
 from agentkit.core.protocol import TaskMessage, TaskResult
 
@@ -29,7 +29,7 @@ class SynthesizerMixin:
 
     async def _synthesize_results(
         self, lead: Expert, task: str, completed_phases: list[PlanPhase]
-    ) -> dict[str, Any]:
+    ) -> dict[str, object]:
         """Lead Expert synthesizes results using BEST strategy.
 
         The Lead Expert evaluates all completed phase results and produces
@@ -114,8 +114,8 @@ class SynthesizerMixin:
         self,
         task: str,
         plan: TeamPlan,
-        phase_results: dict[str, dict[str, Any]],
-    ) -> dict[str, Any]:
+        phase_results: dict[str, dict[str, object]],
+    ) -> dict[str, object]:
         """Fallback to single agent mode when pipeline execution fails.
 
         Uses the lead expert (or first active expert) to complete the original task.
@@ -128,7 +128,7 @@ class SynthesizerMixin:
             active = self._team.active_experts
             expert = active[0] if active else None
 
-        fallback_result: dict[str, Any] | None = None
+        fallback_result: dict[str, object] | None = None
         if expert:
             try:
                 task_msg = TaskMessage(

src/agentkit/experts/board.py

@@ -16,7 +16,6 @@ import enum
 import logging
 import time
 import uuid
-from typing import Any
 
 from .config import ExpertConfig
 from .expert import Expert
@@ -72,7 +71,7 @@ class BoardTeam:
 
         # Discussion state
         self._topic: str = ""
-        self._history: list[dict[str, Any]] = []
+        self._history: list[dict[str, object]] = []
         self._current_round: int = 0
         self._max_rounds: int = max_rounds
         self._user_interventions: list[str] = []  # Pending user messages
@@ -125,7 +124,7 @@ class BoardTeam:
         return self._max_rounds
 
     @property
-    def history(self) -> list[dict[str, Any]]:
+    def history(self) -> list[dict[str, object]]:
         return self._history.copy()
 
     def get_expert(self, name: str) -> Expert | None:
@@ -254,7 +253,7 @@ class BoardTeam:
 
         return "\n\n---\n\n".join(lines)
 
-    async def compress_history(self, moderator: Expert, llm_gateway: Any) -> None:
+    async def compress_history(self, moderator: Expert, llm_gateway: object) -> None:
         """Compress discussion history when it exceeds token threshold.
 
         The moderator summarizes each round's key points, replacing
@@ -291,7 +290,7 @@ class BoardTeam:
 
             # Parse compressed history back to entries
             # This is a best-effort compression; if parsing fails, keep original
-            new_history: list[dict[str, Any]] = []
+            new_history: list[dict[str, object]] = []
             current_round = 0
             for line in compressed.split("\n"):
                 line = line.strip()

src/agentkit/experts/board_orchestrator.py

@@ -19,7 +19,6 @@ from __future__ import annotations
 
 import asyncio
 import logging
-from typing import Any
 
 from .expert import Expert
 from .board import BoardTeam, BoardStatus
@@ -43,7 +42,7 @@ class BoardOrchestrator:
     def __init__(self, team: BoardTeam) -> None:
         self._team = team
 
-    async def execute(self, topic: str) -> dict[str, Any]:
+    async def execute(self, topic: str) -> dict[str, object]:
         """Execute a board meeting discussion.
 
         Flow:
@@ -351,7 +350,7 @@ class BoardOrchestrator:
             logger.warning(f"Moderator summary generation failed: {e}")
             return f"[第 {round} 轮讨论完成，主持人小结生成失败]"
 
-    async def _generate_final_conclusion(self, moderator: Expert, topic: str) -> dict[str, Any]:
+    async def _generate_final_conclusion(self, moderator: Expert, topic: str) -> dict[str, object]:
         """Generate moderator's final conclusion.
 
         The moderator gives:
@@ -428,7 +427,7 @@ class BoardOrchestrator:
                 "dissent_points": [],
             }
 
-    async def _generate_fallback_conclusion(self, moderator: Expert, topic: str) -> dict[str, Any]:
+    async def _generate_fallback_conclusion(self, moderator: Expert, topic: str) -> dict[str, object]:
         """Generate a fallback conclusion when execution fails.
 
         Uses existing discussion history to provide a basic summary.
@@ -491,7 +490,7 @@ class BoardOrchestrator:
                 return True
         return False
 
-    def _get_llm_gateway(self, expert: Expert | None = None) -> Any:
+    def _get_llm_gateway(self, expert: Expert | None = None) -> object:
         """Get LLM gateway from the given expert or the moderator's agent.
 
         Falls back to other active experts if the primary target has no gateway.
@@ -509,7 +508,7 @@ class BoardOrchestrator:
                     return gateway
         return None
 
-    async def _broadcast_event(self, event_type: str, data: dict[str, Any]) -> None:
+    async def _broadcast_event(self, event_type: str, data: dict[str, object]) -> None:
         """Broadcast a board event to the team channel.
 
         Events are emitted via handoff_transport for WebSocket relay.

src/agentkit/experts/config.py

@@ -3,7 +3,6 @@
 from __future__ import annotations
 
 from dataclasses import dataclass
-from typing import Any
 
 from agentkit.core.config_driven import AgentConfig
 
@@ -24,12 +23,12 @@ class ExpertConfig(AgentConfig):
         task_mode: str = "llm_generate",
         supported_tasks: list[str] | None = None,
         max_concurrency: int = 1,
-        input_schema: dict[str, Any] | None = None,
-        output_schema: dict[str, Any] | None = None,
+        input_schema: dict[str, object] | None = None,
+        output_schema: dict[str, object] | None = None,
         prompt: dict[str, str] | None = None,
-        llm: dict[str, Any] | None = None,
+        llm: dict[str, object] | None = None,
         tools: list[str] | None = None,
-        memory: dict[str, Any] | None = None,
+        memory: dict[str, object] | None = None,
         custom_handler: str | None = None,
         # Expert 专属字段
         persona: str = "",
@@ -70,7 +69,7 @@ class ExpertConfig(AgentConfig):
         self.decision_framework = decision_framework
 
     @classmethod
-    def from_dict(cls, data: dict[str, Any]) -> ExpertConfig:
+    def from_dict(cls, data: dict[str, object]) -> ExpertConfig:
         """从字典创建配置"""
         return cls(
             name=data["name"],
@@ -98,7 +97,7 @@ class ExpertConfig(AgentConfig):
             decision_framework=data.get("decision_framework", ""),
         )
 
-    def to_dict(self) -> dict[str, Any]:
+    def to_dict(self) -> dict[str, object]:
         """序列化为字典，包含 Expert 专属字段"""
         d = super().to_dict()
         d["persona"] = self.persona
@@ -125,7 +124,7 @@ class ExpertTemplate:
     is_builtin: bool = False
     description: str = ""
 
-    def to_dict(self) -> dict[str, Any]:
+    def to_dict(self) -> dict[str, object]:
         """序列化为字典"""
         return {
             "name": self.name,
@@ -135,7 +134,7 @@ class ExpertTemplate:
         }
 
     @classmethod
-    def from_dict(cls, data: dict[str, Any]) -> ExpertTemplate:
+    def from_dict(cls, data: dict[str, object]) -> ExpertTemplate:
         """从字典创建模板"""
         config_data = data["config"]
         config = ExpertConfig.from_dict(config_data)

src/agentkit/experts/orchestrator.py

@@ -14,7 +14,6 @@ import asyncio
 import json
 import logging
 import re
-from typing import Any
 
 from agentkit.core.exceptions import LLMProviderError
 from agentkit.llm.gateway import LLMGateway
@@ -72,7 +71,7 @@ class TeamOrchestrator(
         self,
         team: ExpertTeam,
         max_concurrent_phases: int | None = None,
-        checkpoint: Any = None,
+        checkpoint: object | None = None,
         workspace_root: str | None = None,
         rollback_timeout: float | None = None,
     ) -> None:
@@ -95,7 +94,7 @@ class TeamOrchestrator(
         self._workspace_root = workspace_root
         self._rollback_timeout = rollback_timeout or self.DEFAULT_ROLLBACK_TIMEOUT
 
-    async def execute(self, task: str) -> dict[str, Any]:
+    async def execute(self, task: str) -> dict[str, object]:
         """Execute a task in pipeline mode. Lead decomposes → topological sort →
         execute layers (parallel within layer) → synthesize. Returns dict with
         status/result/phase_results/plan."""
@@ -175,7 +174,7 @@ class TeamOrchestrator(
 
         # 4. Set EXECUTING status, execute phases
         self._team.set_status(TeamStatus.EXECUTING)
-        phase_results: dict[str, dict[str, Any]] = {}
+        phase_results: dict[str, dict[str, object]] = {}
 
         return await self._run_pipeline(lead, plan, phase_results, task)
 
@@ -183,9 +182,9 @@ class TeamOrchestrator(
         self,
         lead: Expert,
         plan: TeamPlan,
-        phase_results: dict[str, dict[str, Any]],
+        phase_results: dict[str, dict[str, object]],
         task: str,
-    ) -> dict[str, Any]:
+    ) -> dict[str, object]:
         """Execute the pipeline loop: run pending phases, synthesize, return result.
 
         Shared by execute() and resume(). phase_results may be pre-populated
@@ -220,7 +219,7 @@ class TeamOrchestrator(
                     break
 
                 # Execute all phases in this layer in parallel (with concurrency limit)
-                async def _bounded_phase(ph: PlanPhase) -> dict[str, Any]:
+                async def _bounded_phase(ph: PlanPhase) -> dict[str, object]:
                     async with self._phase_semaphore:
                         return await self._execute_phase(ph, plan)
 
@@ -335,7 +334,7 @@ class TeamOrchestrator(
             await self._broadcast_event("team_dissolved", {"team_id": self._team.team_id})
             return await self._fallback_to_single_agent(task, plan, phase_results)
 
-    async def resume(self, plan_id: str) -> dict[str, Any]:
+    async def resume(self, plan_id: str) -> dict[str, object]:
         """Resume from last checkpoint: load plan, restore completed/failed phases,
         continue via _run_pipeline. Returns same dict shape as execute()."""
         if self._checkpoint is None:
@@ -362,7 +361,7 @@ class TeamOrchestrator(
 
         # 3. Load checkpoints, mark completed phases
         checkpoints = await self._checkpoint.list_checkpoints(plan_id)
-        phase_results: dict[str, dict[str, Any]] = {}
+        phase_results: dict[str, dict[str, object]] = {}
         completed_phase_ids: set[str] = set()
         failed_phase_ids: set[str] = set()
 
@@ -492,7 +491,7 @@ class TeamOrchestrator(
 
         # First pass: create phases with IDs, build name->id mapping
         name_to_id: dict[str, str] = {}
-        raw_phases: list[dict[str, Any]] = []
+        raw_phases: list[dict[str, object]] = []
 
         for item in items:
             if not isinstance(item, dict):
@@ -584,7 +583,7 @@ class TeamOrchestrator(
                     return gateway
         return None
 
-    async def _broadcast_event(self, event_type: str, data: dict[str, Any]) -> None:
+    async def _broadcast_event(self, event_type: str, data: dict[str, object]) -> None:
         """Broadcast an orchestration event to the team channel via handoff_transport."""
         if self._team.handoff_transport:
             try:

src/agentkit/experts/plan.py

@@ -15,7 +15,6 @@ from __future__ import annotations
 import enum
 import uuid
 from dataclasses import dataclass, field
-from typing import Any
 
 
 class MergeStrategy(str, enum.Enum):
@@ -82,9 +81,9 @@ class SubTask:
     description: str = ""
     assigned_expert: str = ""
     status: SubTaskStatus = SubTaskStatus.PENDING
-    result: dict[str, Any] | None = None
+    result: dict[str, object] | None = None
 
-    def to_dict(self) -> dict[str, Any]:
+    def to_dict(self) -> dict[str, object]:
         """序列化为字典"""
         return {
             "id": self.id,
@@ -95,7 +94,7 @@ class SubTask:
         }
 
     @classmethod
-    def from_dict(cls, data: dict[str, Any]) -> SubTask:
+    def from_dict(cls, data: dict[str, object]) -> SubTask:
         """从字典创建 SubTask"""
         return cls(
             id=data.get("id", str(uuid.uuid4())),
@@ -124,7 +123,7 @@ class CollaborationContract:
     content_description: str = ""
     status: str = "pending"
 
-    def to_dict(self) -> dict[str, Any]:
+    def to_dict(self) -> dict[str, object]:
         """序列化为字典"""
         return {
             "from_expert": self.from_expert,
@@ -134,7 +133,7 @@ class CollaborationContract:
         }
 
     @classmethod
-    def from_dict(cls, data: dict[str, Any]) -> CollaborationContract:
+    def from_dict(cls, data: dict[str, object]) -> CollaborationContract:
         """从字典创建 CollaborationContract"""
         return cls(
             from_expert=data.get("from_expert", ""),
@@ -176,9 +175,9 @@ class PlanPhase:
     task_description: str = ""
     depends_on: list[str] = field(default_factory=list)
     status: PhaseStatus = PhaseStatus.PENDING
-    result: dict[str, Any] | None = None
+    result: dict[str, object] | None = None
     phase_type: PhaseType = PhaseType.EXECUTION
-    debate_config: dict[str, Any] | None = None
+    debate_config: dict[str, object] | None = None
     collaboration_contracts: list[CollaborationContract] = field(default_factory=list)
     rework_count: int = 0
     review_feedback: str | None = None
@@ -188,7 +187,7 @@ class PlanPhase:
     validation_command: str | None = None
     rollback_command: str | None = None
 
-    def to_dict(self) -> dict[str, Any]:
+    def to_dict(self) -> dict[str, object]:
         """序列化为字典"""
         # Serialize result to string to match frontend ITeamPlanPhase.result type
         result_str: str | None = None
@@ -197,7 +196,7 @@ class PlanPhase:
                 result_str = self.result.get("content", str(self.result))
             else:
                 result_str = str(self.result)
-        out: dict[str, Any] = {
+        out: dict[str, object] = {
             "id": self.id,
             "name": self.name,
             "assigned_expert": self.assigned_expert,
@@ -219,7 +218,7 @@ class PlanPhase:
         return out
 
     @classmethod
-    def from_dict(cls, data: dict[str, Any]) -> PlanPhase:
+    def from_dict(cls, data: dict[str, object]) -> PlanPhase:
         """从字典创建 PlanPhase"""
         contracts_data = data.get("collaboration_contracts", [])
         if not isinstance(contracts_data, list):
@@ -272,7 +271,7 @@ class TeamPlan:
     status: PlanStatus = PlanStatus.DRAFT
     lead_expert: str = ""
 
-    def to_dict(self) -> dict[str, Any]:
+    def to_dict(self) -> dict[str, object]:
         """序列化为字典"""
         return {
             "id": self.id,
@@ -284,7 +283,7 @@ class TeamPlan:
         }
 
     @classmethod
-    def from_dict(cls, data: dict[str, Any]) -> TeamPlan:
+    def from_dict(cls, data: dict[str, object]) -> TeamPlan:
         """从字典创建 TeamPlan"""
         subtasks = [SubTask.from_dict(st) for st in data.get("subtasks", [])]
         phases = [PlanPhase.from_dict(ph) for ph in data.get("phases", [])]
@@ -307,7 +306,7 @@ class TeamPlan:
         return None
 
     def update_subtask_status(
-        self, subtask_id: str, status: SubTaskStatus, result: dict[str, Any] | None = None
+        self, subtask_id: str, status: SubTaskStatus, result: dict[str, object] | None = None
     ) -> None:
         """更新子任务状态和可选的结果"""
         st = self.get_subtask(subtask_id)
@@ -343,7 +342,7 @@ class TeamPlan:
         return None
 
     def update_phase_status(
-        self, phase_id: str, status: PhaseStatus, result: dict[str, Any] | None = None
+        self, phase_id: str, status: PhaseStatus, result: dict[str, object] | None = None
     ) -> None:
         """更新阶段状态和可选的结果"""
         ph = self.get_phase(phase_id)

src/agentkit/experts/team.py

@@ -17,7 +17,6 @@ import enum
 import logging
 import time
 import uuid
-from typing import Any
 
 from .config import ExpertConfig
 from .expert import Expert
@@ -63,7 +62,7 @@ class ExpertTeam:
         workspace: SharedWorkspace | None = None,
         pool: AgentPool | None = None,
         template_registry: ExpertTemplateRegistry | None = None,
-        redis_client: Any = None,
+        redis_client: object | None = None,
     ):
         self.team_id = team_id or str(uuid.uuid4())
         # U4: Accept redis_client for SharedWorkspace state offloading.

src/agentkit/memory/adapters/confluence.py

@@ -6,6 +6,7 @@
 
 from __future__ import annotations
 
+import asyncio
 import logging
 
 import httpx
@@ -83,7 +84,7 @@ class ConfluenceAdapter(KBAdapter):
             resp = await client.get("/rest/api/user/current")
             self._authenticated = resp.status_code == 200
             return self._authenticated
-        except Exception as e:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError) as e:
             logger.error(f"Confluence auth error: {e}")
             self._authenticated = False
             return False
@@ -139,7 +140,7 @@ class ConfluenceAdapter(KBAdapter):
                 f"Confluence search HTTP error: {e.response.status_code} — {e.response.text[:200]}"
             )
             return []
-        except Exception as e:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError) as e:
             logger.error(f"Confluence search error: {e}")
             return []
 
@@ -170,7 +171,7 @@ class ConfluenceAdapter(KBAdapter):
                     "version": page.get("version", {}).get("number", 0),
                 },
             )
-        except Exception as e:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError) as e:
             logger.error(f"Confluence get_document error: {e}")
             return None
 
@@ -202,7 +203,7 @@ class ConfluenceAdapter(KBAdapter):
                     )
                 ]
             )
-        except Exception as e:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError) as e:
             logger.error(f"Confluence list_sources error: {e}")
             return [
                 SourceInfo(
@@ -218,5 +219,7 @@ class ConfluenceAdapter(KBAdapter):
         try:
             resp = await client.get("/rest/api/space", params={"limit": 1})
             return resp.status_code == 200
-        except Exception:
+        except asyncio.CancelledError:
+            raise
+        except Exception:  # noqa: BLE001 — health_check 设计意图：任何异常都返回 False
             return False

src/agentkit/memory/adapters/feishu.py

@@ -6,6 +6,7 @@
 
 from __future__ import annotations
 
+import asyncio
 import logging
 import time
 from typing import TypeAlias
@@ -101,7 +102,7 @@ class FeishuKBAdapter(KBAdapter):
             else:
                 logger.error(f"Feishu auth failed: code={data.get('code')}, msg={data.get('msg')}")
                 return None
-        except Exception as e:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError) as e:
             logger.error(f"Feishu auth error: {e}")
             return None
 
@@ -166,7 +167,7 @@ class FeishuKBAdapter(KBAdapter):
                 f"Feishu search HTTP error: {e.response.status_code} — {e.response.text[:200]}"
             )
             return []
-        except Exception as e:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError) as e:
             logger.error(f"Feishu search error: {e}")
             return []
 
@@ -199,7 +200,7 @@ class FeishuKBAdapter(KBAdapter):
                     "obj_type": node.get("obj_type", ""),
                 },
             )
-        except Exception as e:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError) as e:
             logger.error(f"Feishu get_document error: {e}")
             return None
 
@@ -241,7 +242,7 @@ class FeishuKBAdapter(KBAdapter):
                     )
                 ]
             )
-        except Exception as e:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError) as e:
             logger.error(f"Feishu list_sources error: {e}")
             return [
                 SourceInfo(
@@ -256,5 +257,7 @@ class FeishuKBAdapter(KBAdapter):
         try:
             token = await self._get_access_token()
             return token is not None
-        except Exception:
+        except asyncio.CancelledError:
+            raise
+        except Exception:  # noqa: BLE001 — health_check 设计意图：任何异常都返回 False
             return False

src/agentkit/memory/adapters/generic_http.py

@@ -6,6 +6,7 @@
 
 from __future__ import annotations
 
+import asyncio
 import logging
 
 import httpx
@@ -138,7 +139,7 @@ class GenericHTTPAdapter(KBAdapter):
                 f"GenericHTTP search HTTP error: {e.response.status_code} — {e.response.text[:200]}"
             )
             return []
-        except Exception as e:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError) as e:
             logger.error(f"GenericHTTP search error: {e}")
             return []
 
@@ -179,7 +180,7 @@ class GenericHTTPAdapter(KBAdapter):
                 f"GenericHTTP ingest HTTP error: {e.response.status_code} — {e.response.text[:200]}"
             )
             return []
-        except Exception as e:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError) as e:
             logger.error(f"GenericHTTP ingest error: {e}")
             return []
 
@@ -194,7 +195,7 @@ class GenericHTTPAdapter(KBAdapter):
             if resp.status_code in (200, 204):
                 return True
             return False
-        except Exception as e:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError) as e:
             logger.error(f"GenericHTTP delete_by_id error: {e}")
             return False
 
@@ -216,7 +217,7 @@ class GenericHTTPAdapter(KBAdapter):
                 source_id=data.get("source_id", self._source_id),
                 metadata=data.get("metadata", {}),
             )
-        except Exception as e:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError) as e:
             logger.error(f"GenericHTTP get_document error: {e}")
             return None
 
@@ -254,7 +255,9 @@ class GenericHTTPAdapter(KBAdapter):
                         )
                     ]
                 )
-        except Exception as e:
+        except asyncio.CancelledError:
+            raise
+        except Exception as e:  # noqa: BLE001 — list_sources 设计意图：任何异常回退到默认信息源
             logger.debug(f"GenericHTTP list_sources error (endpoint may not exist): {e}")
 
         return [
@@ -272,14 +275,18 @@ class GenericHTTPAdapter(KBAdapter):
             resp = await client.get("/health")
             if resp.status_code == 200:
                 return True
-        except Exception:
+        except asyncio.CancelledError:
+            raise
+        except Exception:  # noqa: BLE001 — health_check 设计意图：任何异常继续尝试 fallback
             pass
 
         # Fallback: try the base endpoint
         try:
             resp = await client.get("/")
             return resp.status_code in (200, 401, 403)
-        except Exception:
+        except asyncio.CancelledError:
+            raise
+        except Exception:  # noqa: BLE001 — health_check 设计意图：任何异常都返回 False
             return False
 
     async def authenticate(self) -> bool:
@@ -289,6 +296,8 @@ class GenericHTTPAdapter(KBAdapter):
         """
         try:
             self._authenticated = await self.health_check()
-        except Exception:
+        except asyncio.CancelledError:
+            raise
+        except Exception:  # noqa: BLE001 — authenticate 设计意图：任何异常都返回 False
             self._authenticated = False
         return self._authenticated

src/agentkit/memory/contextual_retrieval.py

@@ -6,6 +6,7 @@
 
 from __future__ import annotations
 
+import asyncio
 import hashlib
 import logging
 from dataclasses import dataclass
@@ -199,7 +200,7 @@ class ContextualChunker:
             )
             context = response.content.strip()
             return context
-        except Exception as e:
+        except (ConnectionError, RuntimeError, asyncio.TimeoutError, ValueError) as e:
             logger.warning(f"Context generation failed for chunk: {e}")
             return ""
 

src/agentkit/memory/document_loader.py

@@ -9,6 +9,8 @@ from __future__ import annotations
 import io
 import logging
 import uuid
+import zipfile
+import asyncio
 from dataclasses import dataclass, field
 from datetime import datetime, timezone
 from pathlib import Path
@@ -194,7 +196,9 @@ class DocumentLoader:
             return text, meta
         except ImportError:
             pass
-        except Exception as e:
+        except asyncio.CancelledError:
+            raise
+        except Exception as e:  # noqa: BLE001 — 解析器设计意图：任何失败回退到文本
             logger.warning(f"PyMuPDF parsing failed for {filename}: {e}")
 
         # 尝试 pdfplumber
@@ -217,7 +221,9 @@ class DocumentLoader:
             return text, meta
         except ImportError:
             pass
-        except Exception as e:
+        except asyncio.CancelledError:
+            raise
+        except Exception as e:  # noqa: BLE001 — 解析器设计意图：任何失败回退到文本
             logger.warning(f"pdfplumber parsing failed for {filename}: {e}")
 
         # 回退到纯文本
@@ -264,7 +270,9 @@ class DocumentLoader:
         except ImportError:
             logger.warning(f"python-docx not available for {filename}, falling back to text")
             return self._parse_text(content, filename)
-        except Exception as e:
+        except asyncio.CancelledError:
+            raise
+        except Exception as e:  # noqa: BLE001 — 解析器设计意图：任何失败回退到文本
             logger.warning(f"python-docx parsing failed for {filename}: {e}")
             return self._parse_text(content, filename)
 
@@ -333,7 +341,9 @@ class DocumentLoader:
         except ImportError:
             logger.warning(f"openpyxl not available for {filename}, falling back to text")
             return self._parse_text(content, filename)
-        except Exception as e:
+        except asyncio.CancelledError:
+            raise
+        except Exception as e:  # noqa: BLE001 — 解析器设计意图：任何失败回退到文本
             logger.warning(f"openpyxl parsing failed for {filename}: {e}")
             return self._parse_text(content, filename)
 
@@ -407,7 +417,9 @@ class DocumentLoader:
         except ImportError:
             logger.warning(f"BeautifulSoup not available for {filename}, falling back to text")
             return self._parse_text(content, filename)
-        except Exception as e:
+        except asyncio.CancelledError:
+            raise
+        except Exception as e:  # noqa: BLE001 — 解析器设计意图：任何失败回退到文本
             logger.warning(f"BeautifulSoup parsing failed for {filename}: {e}")
             return self._parse_text(content, filename)
 

src/agentkit/memory/embedder.py

@@ -8,9 +8,7 @@ import os
 import time
 from abc import ABC, abstractmethod
 from collections import OrderedDict
-from typing import TYPE_CHECKING
 
-if TYPE_CHECKING:
 import httpx
 
 logger = logging.getLogger(__name__)
@@ -108,8 +106,6 @@ class OpenAIEmbedder(Embedder):
     def _get_client(self) -> httpx.AsyncClient:
         """Lazily create and reuse a single httpx.AsyncClient."""
         if self._client is None:
-            import httpx
-
             self._client = httpx.AsyncClient(timeout=30.0)
         return self._client
 
@@ -153,7 +149,7 @@ class OpenAIEmbedder(Embedder):
                 self._cache.put(text, embedding)
 
             return embedding
-        except Exception as e:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError) as e:
             logger.error(f"OpenAI embedding failed: {e}")
             raise
 

src/agentkit/memory/episodic.py

@@ -9,6 +9,7 @@ from datetime import datetime, timezone
 from typing import TYPE_CHECKING
 
 from sqlalchemy import text
+from sqlalchemy.exc import DBAPIError
 
 from agentkit.memory.base import Memory, MemoryItem, MetadataDict
 from agentkit.memory.embedder import Embedder
@@ -94,7 +95,7 @@ class EpisodicMemory(Memory):
                 )
                 db.add(entry)
                 await db.commit()
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 await db.rollback()
                 logger.error(f"Failed to store episodic memory: {e}")
                 raise
@@ -111,7 +112,7 @@ class EpisodicMemory(Memory):
                 if self._pgvector_enabled:
                     return await self._retrieve_pgvector(db, query_embedding)
                 return await self._retrieve_client_side(db, query_embedding)
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 logger.error(f"Failed to retrieve episodic memory: {e}")
                 return None
 
@@ -223,7 +224,7 @@ class EpisodicMemory(Memory):
                 if self._pgvector_enabled and self._embedder:
                     return await self._search_pgvector(db, query, top_k, filters, search_multiplier)
                 return await self._search_client_side(db, query, top_k, filters, search_multiplier)
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 logger.error(f"Failed to search episodic memory: {e}")
                 return []
 
@@ -405,7 +406,7 @@ class EpisodicMemory(Memory):
                 await db.execute(stmt)
                 await db.commit()
                 return True
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 await db.rollback()
                 logger.error(f"Failed to delete episodic memory: {e}")
                 return False

src/agentkit/memory/http_rag.py

@@ -147,7 +147,7 @@ class HttpRAGService:
         except httpx.RequestError as e:
             logger.error(f"RAG search request error: {e}")
             return []
-        except Exception as e:
+        except (ValueError, KeyError, TypeError, RuntimeError) as e:
             logger.error(f"RAG search unexpected error: {e}")
             return []
 
@@ -237,7 +237,7 @@ class HttpRAGService:
             except httpx.RequestError as e:
                 logger.error(f"RAG enhanced_search request error for KB {kb_id}: {e}")
                 raise
-            except Exception as e:
+            except (ValueError, KeyError, TypeError, RuntimeError) as e:
                 logger.error(f"RAG enhanced_search unexpected error for KB {kb_id}: {e}")
                 raise
 
@@ -302,7 +302,7 @@ class HttpRAGService:
         except httpx.HTTPStatusError as e:
             logger.error(f"RAG ingest HTTP error: {e.response.status_code}")
             return None
-        except Exception as e:
+        except (ValueError, KeyError, TypeError, RuntimeError) as e:
             logger.error(f"RAG ingest error: {e}")
             return None
 
@@ -312,7 +312,7 @@ class HttpRAGService:
         try:
             resp = await client.get("/bases")
             return resp.status_code in (200, 401)  # 401 = 服务在但需认证
-        except Exception:
+        except (httpx.HTTPError, ValueError, KeyError, TypeError):
             return False
 
     async def close(self) -> None:

src/agentkit/memory/local_rag.py

@@ -14,6 +14,8 @@ import re
 from datetime import datetime, timezone
 from typing import TYPE_CHECKING, TypeAlias
 
+from sqlalchemy.exc import DBAPIError
+
 from agentkit.memory.chunking import Chunk, StructuralChunker, TextChunker
 from agentkit.memory.document_loader import Document as LoaderDocument
 from agentkit.memory.embedder import Embedder
@@ -108,7 +110,7 @@ class LocalRAGService:
                 await self._store_chunks(doc, chunks)
                 ingested_ids.append(doc.doc_id)
                 logger.info(f"Ingested document '{doc.title}' with {len(chunks)} chunks")
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 logger.error(f"Failed to ingest document '{doc.title}': {e}")
 
         return ingested_ids
@@ -130,7 +132,7 @@ class LocalRAGService:
                 if self._pgvector_enabled:
                     return await self._query_pgvector(db, query_embedding, top_k)
                 return await self._query_client_side(db, query_embedding, top_k)
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 logger.error(f"Failed to query knowledge base: {e}")
                 return []
 
@@ -151,7 +153,7 @@ class LocalRAGService:
                 await db.execute(sql, {"doc_id": id})
                 await db.commit()
                 return True
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 await db.rollback()
                 logger.error(f"Failed to delete document {id}: {e}")
                 return False
@@ -190,7 +192,7 @@ class LocalRAGService:
                         )
                     )
                 return sources
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 logger.error(f"Failed to list sources: {e}")
                 return []
 
@@ -202,7 +204,7 @@ class LocalRAGService:
 
                 await db.execute(sql_text(f"SELECT 1 FROM {self._table_name} LIMIT 1"))
                 return True
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 logger.error(f"Health check failed: {e}")
                 return False
 
@@ -268,7 +270,7 @@ class LocalRAGService:
 
                 await db.execute(sql, params_list)
                 await db.commit()
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 await db.rollback()
                 logger.error(f"Failed to store chunks for document '{doc.title}': {e}")
                 raise
@@ -455,7 +457,7 @@ class InMemoryLocalRAGService:
                 }
                 ingested_ids.append(doc.doc_id)
                 logger.info(f"Ingested document '{doc.title}' with {len(chunks)} chunks")
-            except Exception as e:
+            except (DBAPIError, ValueError, KeyError, RuntimeError) as e:
                 logger.error(f"Failed to ingest document '{doc.title}': {e}")
 
         return ingested_ids

src/agentkit/memory/profile.py

@@ -378,7 +378,7 @@ class MemoryStore:
         try:
             new_prompt = self.refresh_system_prompt()
             self._on_change(new_prompt)
-        except Exception:
+        except (OSError, ValueError, RuntimeError, TypeError):
             import logging
 
             logging.getLogger(__name__).warning("Memory notify_change failed", exc_info=True)

src/agentkit/memory/retriever.py

@@ -392,5 +392,5 @@ class RetrieveKnowledgeTool(Tool):
                     }
                 )
             return {"query": query, "results": results, "call_count": self._call_count}
-        except Exception as e:
+        except (ConnectionError, RuntimeError, ValueError, KeyError, TypeError) as e:
             return {"error": str(e), "results": []}

src/agentkit/quality/cascade_state_store.py

@@ -13,6 +13,12 @@ import logging
 import time
 from typing import Any, Protocol, runtime_checkable
 
+# redis 可选依赖；未安装时回退为 Exception 以保留原 catch-all 语义（降级到 fallback）
+try:
+    from redis.exceptions import RedisError as _RedisError
+except ImportError:
+    _RedisError = Exception
+
 logger = logging.getLogger(__name__)
 
 
@@ -157,7 +163,7 @@ class RedisCascadeStateStore:
             pipe.expire(key, self._session_ttl)
             results = pipe.execute()
             return results[0]
-        except Exception as e:
+        except (ImportError, OSError, _RedisError, ValueError, KeyError, RuntimeError, TypeError) as e:
             logger.warning(f"Redis cascade increment failed: {e}")
             self._degrade_to_fallback()
             if self._fallback is not None:
@@ -171,7 +177,7 @@ class RedisCascadeStateStore:
             r = self._get_sync_redis()
             val = r.get(f"{self.INTER_PREFIX}{session_id}")
             return int(val) if val is not None else 0
-        except Exception as e:
+        except (ImportError, OSError, _RedisError, ValueError, KeyError, RuntimeError, TypeError) as e:
             logger.warning(f"Redis cascade get failed: {e}")
             if self._fallback is not None:
                 return self._fallback.get_interaction(session_id)
@@ -188,7 +194,7 @@ class RedisCascadeStateStore:
             pipe.set(key, depth)
             pipe.expire(key, self._session_ttl)
             pipe.execute()
-        except Exception as e:
+        except (ImportError, OSError, _RedisError, ValueError, KeyError, RuntimeError, TypeError) as e:
             logger.warning(f"Redis cascade set_depth failed: {e}")
             self._degrade_to_fallback()
             if self._fallback is not None:
@@ -201,7 +207,7 @@ class RedisCascadeStateStore:
             r = self._get_sync_redis()
             val = r.get(f"{self.DEPTH_PREFIX}{session_id}")
             return int(val) if val is not None else 0
-        except Exception as e:
+        except (ImportError, OSError, _RedisError, ValueError, KeyError, RuntimeError, TypeError) as e:
             logger.warning(f"Redis cascade get_depth failed: {e}")
             if self._fallback is not None:
                 return self._fallback.get_depth(session_id)
@@ -217,7 +223,7 @@ class RedisCascadeStateStore:
             pipe.delete(f"{self.INTER_PREFIX}{session_id}")
             pipe.delete(f"{self.DEPTH_PREFIX}{session_id}")
             pipe.execute()
-        except Exception as e:
+        except (ImportError, OSError, _RedisError, ValueError, KeyError, RuntimeError, TypeError) as e:
             logger.warning(f"Redis cascade reset failed: {e}")
             self._degrade_to_fallback()
             if self._fallback is not None:

src/agentkit/server/app.py

@@ -81,7 +81,7 @@ def _build_llm_gateway(config: ServerConfig) -> LLMGateway:
                 backend=config.usage_store.get("backend", "memory"),
                 redis_url=config.usage_store.get("redis_url", "redis://localhost:6379"),
             )
-        except Exception as e:
+        except (ConnectionError, OSError, asyncio.TimeoutError, ValueError, KeyError, RuntimeError) as e:
             logger.warning(f"Failed to initialize usage store: {e}, using in-memory")
 
     gateway = LLMGateway(config=config.llm_config, usage_store=usage_store)
@@ -93,7 +93,7 @@ def _build_llm_gateway(config: ServerConfig) -> LLMGateway:
         try:
             provider = _create_provider(name, pconf)
             gateway.register_provider(name, provider)
-        except Exception as e:
+        except (ValueError, TypeError, KeyError, RuntimeError, ConnectionError, OSError) as e:
             logger.warning(f"Failed to register LLM provider '{name}': {e}")
 
     return gateway
@@ -222,7 +222,7 @@ async def lifespan(app: FastAPI):
                     if agent is not None:
                         agent._system_prompt = new_prompt
                         updated += 1
-                except Exception:
+                except (AttributeError, TypeError, ValueError):
                     logger.warning(
                         f"Failed to update system prompt for agent '{agent_name}'", exc_info=True
                     )
@@ -283,7 +283,9 @@ async def lifespan(app: FastAPI):
                 agent._tool_registry.register(DocumentTool(service=doc_service))
                 app.state.document_service = doc_service
                 logger.info("DocumentTool registered with word/excel/pdf renderers")
-            except Exception:
+            except asyncio.CancelledError:
+                raise
+            except Exception:  # noqa: BLE001 — DocumentTool init + registry; diverse failure modes
                 logger.exception("Failed to register DocumentTool")
 
             # Override system prompt with memory-injected version + available tools
@@ -296,7 +298,9 @@ async def lifespan(app: FastAPI):
             )
 
             logger.info("GUI mode: created default chat agent with memory + tools")
-        except Exception as e:
+        except asyncio.CancelledError:
+            raise
+        except Exception as e:  # noqa: BLE001 — GUI mode agent creation; multi-step init with diverse failures
             logger.warning(f"GUI mode: failed to create default agent: {e}")
 
         # Load skills from config and register into SkillRegistry
@@ -310,7 +314,7 @@ async def lifespan(app: FastAPI):
             for tool in agent._tool_registry.list_tools():
                 try:
                     tool_registry.register(tool)
-                except Exception:
+                except (ValueError, KeyError, RuntimeError):
                     pass  # Already registered
 
             # Load skills from configured paths
@@ -331,11 +335,13 @@ async def lifespan(app: FastAPI):
                         try:
                             loader.load_from_file(str(p))
                             logger.info(f"GUI mode: loaded skill from {p}")
-                        except Exception as se:
+                        except (ValueError, TypeError, KeyError, OSError, RuntimeError) as se:
                             logger.warning(f"GUI mode: failed to load skill from {p}: {se}")
 
                 logger.info(f"GUI mode: {len(skill_registry.list_skills())} skills registered")
-        except Exception as e:
+        except asyncio.CancelledError:
+            raise
+        except Exception as e:  # noqa: BLE001 — GUI mode skills loading top-level; multi-step init
             logger.warning(f"GUI mode: failed to load skills: {e}")
     elif os.environ.get("AGENTKIT_GUI_MODE"):
         # Agent already exists (e.g. from config), still ensure memory store is available
@@ -369,7 +375,7 @@ async def lifespan(app: FastAPI):
                         if agent is not None:
                             agent._system_prompt = new_prompt
                             updated += 1
-                    except Exception:
+                    except (AttributeError, TypeError, ValueError):
                         logger.warning(
                             f"Failed to update system prompt for agent '{agent_name}'",
                             exc_info=True,
@@ -416,7 +422,9 @@ async def lifespan(app: FastAPI):
         app.state.expert_template_registry = expert_registry
         if total_loaded:
             logger.info(f"Total {total_loaded} ExpertTemplates registered for @board mode")
-    except Exception as e:
+    except asyncio.CancelledError:
+        raise
+    except Exception as e:  # noqa: BLE001 — ExpertTemplate registry loading; multi-step init
         logger.warning(f"Failed to load ExpertTemplates: {e}")
         # Ensure app.state.expert_template_registry always exists (empty registry)
         from agentkit.experts.registry import ExpertTemplateRegistry
@@ -468,7 +476,7 @@ async def lifespan(app: FastAPI):
                     _row = await _cur.fetchone()
                     if _row is not None:
                         default_cal_user_id = str(_row["id"])
-            except Exception:
+            except (ConnectionError, OSError, asyncio.TimeoutError, ValueError, KeyError, RuntimeError):
                 logger.debug("Could not resolve default user_id for CalendarTool", exc_info=True)
 
             calendar_tool = CalendarTool(
@@ -492,7 +500,7 @@ async def lifespan(app: FastAPI):
                 if default_agent is not None:
                     try:
                         default_agent._tool_registry.register(calendar_tool)
-                    except Exception:
+                    except (ValueError, KeyError, RuntimeError, AttributeError):
                         # ponytail: log at debug — CalendarTool double-registration
                         # is expected on reload, but silent pass hides real errors.
                         logger.debug("CalendarTool already registered or registration failed", exc_info=True)
@@ -512,12 +520,18 @@ async def lifespan(app: FastAPI):
                         "CalendarTool attached to default agent (tools=%d)",
                         len(default_agent._tool_registry.list_tools()),
                     )
-            except Exception:
+            except asyncio.CancelledError:
+                raise
+            except Exception:  # noqa: BLE001 — CalendarTool attach; multi-step with attribute access
                 logger.exception("Failed to attach CalendarTool to default agent")
-        except Exception:
+        except asyncio.CancelledError:
+            raise
+        except Exception:  # noqa: BLE001 — CalendarTool registration; multi-step init
             logger.exception("Failed to register CalendarTool")
         logger.info("Calendar subsystem initialized (service + reminder scheduler)")
-    except Exception:
+    except asyncio.CancelledError:
+        raise
+    except Exception:  # noqa: BLE001 — calendar subsystem top-level init; must not block app startup
         logger.exception("Failed to initialize calendar subsystem — calendar API unavailable")
 
     # Bitable subsystem: init DB, service, internal token (KTD11).
@@ -529,7 +543,9 @@ async def lifespan(app: FastAPI):
         app.state.bitable_service = BitableService(db=bitable_db)
         app.state.bitable_internal_token = os.environ.get("AGENTKIT_BITABLE_INTERNAL_TOKEN")
         logger.info("Bitable subsystem initialized")
-    except Exception:
+    except asyncio.CancelledError:
+        raise
+    except Exception:  # noqa: BLE001 — bitable subsystem top-level init; includes import + DB setup
         logger.exception("Failed to initialize bitable subsystem")
 
     # RAG platform subsystem (P1): wire KBStore, RetrievalEngine, HitProcessor,
@@ -552,7 +568,7 @@ async def lifespan(app: FastAPI):
                 from agentkit.rag_platform.store import ensure_tables
 
                 await ensure_tables(rag_database_url)
-            except Exception:
+            except (ConnectionError, OSError, asyncio.TimeoutError, ValueError, KeyError, RuntimeError):
                 logger.exception("Failed to ensure rag_platform tables")
 
             # KBStore — KB/Document persistence
@@ -561,7 +577,9 @@ async def lifespan(app: FastAPI):
 
                 app.state.kb_store = KBStore(session_factory=kb_session_factory)
                 logger.info("KBStore initialized")
-            except Exception:
+            except asyncio.CancelledError:
+                raise
+            except Exception:  # noqa: BLE001 — KBStore init includes import + DB setup
                 logger.exception("Failed to initialize KBStore")
 
             # RetrievalEngine — requires llama_index PGVectorStore
@@ -576,9 +594,13 @@ async def lifespan(app: FastAPI):
                     session_factory=kb_session_factory,
                 )
                 logger.info("RetrievalEngine initialized (pgvector ready)")
-            except Exception:
+            except asyncio.CancelledError:
+                raise
+            except Exception:  # noqa: BLE001 — RetrievalEngine init includes import + pgvector setup
                 logger.exception("Failed to initialize RetrievalEngine — vector search unavailable")
-        except Exception:
+        except asyncio.CancelledError:
+            raise
+        except Exception:  # noqa: BLE001 — rag_platform top-level init
             logger.exception("Failed to initialize rag_platform DB components")
 
     # HitProcessor — LLM-based answer generation from retrieval results
@@ -587,7 +609,9 @@ async def lifespan(app: FastAPI):
 
         app.state.hit_processor = HitProcessor(llm_gateway=app.state.llm_gateway)
         logger.info("HitProcessor initialized")
-    except Exception:
+    except asyncio.CancelledError:
+        raise
+    except Exception:  # noqa: BLE001 — HitProcessor init includes import
         logger.exception("Failed to initialize HitProcessor")
 
     # TaskManager — async vectorize/batch-index tasks
@@ -598,7 +622,9 @@ async def lifespan(app: FastAPI):
 
         app.state.task_manager = TaskManager()
         logger.info("TaskManager initialized (degraded mode: InMemoryTaskStore)")
-    except Exception:
+    except asyncio.CancelledError:
+        raise
+    except Exception:  # noqa: BLE001 — TaskManager init includes import
         logger.exception("Failed to initialize TaskManager")
 
     # KBSettingsStore — per-KB retrieval/hit-processing defaults (process singleton)
@@ -607,7 +633,9 @@ async def lifespan(app: FastAPI):
 
         app.state.kb_settings_store = get_settings_store()
         logger.info("KBSettingsStore initialized")
-    except Exception:
+    except asyncio.CancelledError:
+        raise
+    except Exception:  # noqa: BLE001 — KBSettingsStore init includes import
         logger.exception("Failed to initialize KBSettingsStore")
 
     yield
@@ -618,7 +646,7 @@ async def lifespan(app: FastAPI):
         from agentkit.bitable.db import close_bitable_db
 
         await close_bitable_db()
-    except Exception:
+    except (RuntimeError, asyncio.TimeoutError, ConnectionError, OSError):
         pass
     # Stop MCP servers
     if mcp_manager is not None:
@@ -660,7 +688,7 @@ async def lifespan(app: FastAPI):
         from agentkit.server.routes.channels import close_all_adapters
 
         await close_all_adapters()
-    except Exception:
+    except (RuntimeError, asyncio.TimeoutError, ConnectionError, OSError):
         logger.debug("close_all_adapters 异常已忽略")
 
 
@@ -706,7 +734,7 @@ def _on_config_change(app: FastAPI, config: ServerConfig) -> None:
                     if hasattr(app.state, "agent_pool") and app.state.agent_pool is not None:
                         app.state.agent_pool._llm_gateway = new_gateway
                     logger.info(f"LLM Gateway reloaded (config v{current_version})")
-                except Exception as e:
+                except (ValueError, TypeError, KeyError, RuntimeError, ConnectionError, OSError) as e:
                     logger.error(f"Failed to reload LLM Gateway: {e}")
 
                 # Reload skills if skill paths changed
@@ -730,13 +758,15 @@ def _on_config_change(app: FastAPI, config: ServerConfig) -> None:
                             elif p.is_file() and p.suffix in (".yaml", ".yml"):
                                 try:
                                     loader.load_from_file(str(p))
-                                except Exception:
+                                except (ValueError, TypeError, KeyError, OSError, RuntimeError):
                                     pass
                     app.state.skill_registry = new_skill_registry
                     if hasattr(app.state, "agent_pool") and app.state.agent_pool is not None:
                         app.state.agent_pool._skill_registry = new_skill_registry
                     logger.info(f"Skills reloaded (config v{current_version})")
-                except Exception as e:
+                except asyncio.CancelledError:
+                    raise
+                except Exception as e:  # noqa: BLE001 — skills reload top-level; multi-step init
                     logger.error(f"Failed to reload skills: {e}")
 
                 # Update config version on all agents
@@ -856,7 +886,7 @@ def create_app(
         from agentkit.server.middleware import _load_client_keys
 
         client_keys = _load_client_keys()
-    except Exception as e:
+    except (ValueError, TypeError, KeyError, OSError, RuntimeError) as e:
         logger.warning(f"Failed to load client keys for AuthMiddleware: {e}")
 
     app.add_middleware(
@@ -983,7 +1013,7 @@ def create_app(
 
         try:
             ts_config = {**ts_config, **_json.loads(ts_env)}
-        except Exception:
+        except (ValueError, TypeError):
             pass
     task_store = create_task_store(
         backend=ts_config.get("backend", "memory"),
@@ -1039,7 +1069,9 @@ def create_app(
                 db_path=evo_conf.get("db_path", "~/.agentkit/evolution.db"),
                 database_url=evo_conf.get("database_url"),
             )
-        except Exception as e:
+        except asyncio.CancelledError:
+            raise
+        except Exception as e:  # noqa: BLE001 — evolution store init includes import + DB setup
             logger.warning(f"Failed to initialize evolution store: {e}")
             app.state.evolution_store = None
     else:
@@ -1056,7 +1088,9 @@ def create_app(
                 redis_url=cs_conf.get("redis_url", "redis://localhost:6379"),
                 session_ttl=cs_conf.get("session_ttl", 86400),
             )
-        except Exception as e:
+        except asyncio.CancelledError:
+            raise
+        except Exception as e:  # noqa: BLE001 — cascade state store init includes import + setup
             logger.warning(f"Failed to initialize cascade state store: {e}")
             app.state.cascade_state_store = None
     else:
@@ -1138,7 +1172,7 @@ def create_app(
                         try:
                             epi_session_factory = create_episodic_session_factory(database_url)
                             epi_model = EpisodeModel
-                        except Exception as db_err:
+                        except (ConnectionError, OSError, asyncio.TimeoutError, ValueError, KeyError, RuntimeError, ImportError) as db_err:
                             import logging as _log
 
                             _log.getLogger(__name__).warning(
@@ -1155,7 +1189,9 @@ def create_app(
                         pgvector_enabled=epi_conf.get("pgvector_enabled", True),
                         table_name=epi_conf.get("table_name", "episodic_memories"),
                     )
-                except Exception as e:
+                except asyncio.CancelledError:
+                    raise
+                except Exception as e:  # noqa: BLE001 — episodic memory init; multi-step with embedder + DB
                     import logging
 
                     logging.getLogger(__name__).warning(
@@ -1174,7 +1210,9 @@ def create_app(
                 retrieve_tool = memory_retriever.create_retrieve_tool()
                 if retrieve_tool:
                     app.state.retrieve_knowledge_tool = retrieve_tool
-        except Exception as e:
+        except asyncio.CancelledError:
+            raise
+        except Exception as e:  # noqa: BLE001 — memory components top-level init; multi-step with imports + DB
             import logging
 
             logging.getLogger(__name__).warning(f"Failed to initialize memory components: {e}")

src/agentkit/session/store.py

@@ -7,6 +7,12 @@ import logging
 import os
 from typing import Any, Protocol, runtime_checkable
 
+# redis 可选依赖；未安装时回退为 Exception 以保留原 catch-all 语义
+try:
+    from redis.exceptions import RedisError as _RedisError
+except ImportError:
+    _RedisError = Exception
+
 from agentkit.session.models import Message, Session, SessionStatus
 
 logger = logging.getLogger(__name__)
@@ -207,7 +213,7 @@ class RedisSessionStore:
         try:
             redis = await self._get_redis()
             return await redis.ping()
-        except Exception:
+        except (ImportError, OSError, _RedisError, ValueError, KeyError, RuntimeError):
             return False
 
 
@@ -283,7 +289,7 @@ class FileSessionStore:
                 session = Session.from_dict(data["session"])
                 if agent_name is None or session.agent_name == agent_name:
                     sessions.append(session)
-            except Exception:
+            except (ValueError, KeyError, TypeError, OSError):
                 continue
         sessions.sort(key=lambda s: s.updated_at, reverse=True)
         return sessions[:limit]
@@ -343,7 +349,7 @@ def create_session_store(
             store = RedisSessionStore(redis_url=redis_url, ttl_seconds=ttl_seconds)
             logger.info(f"SessionStore backend: redis")
             return store
-        except Exception as exc:
+        except (ImportError, OSError, _RedisError, ValueError, KeyError, RuntimeError) as exc:
             logger.warning(f"Failed to initialise RedisSessionStore ({exc}), falling back to InMemorySessionStore")
 
     store = InMemorySessionStore()

src/agentkit/skills/loader.py

@@ -56,7 +56,7 @@ class SkillLoader:
             try:
                 skill = self._load_skill_from_file(yaml_path)
                 skills.append(skill)
-            except Exception as e:
+            except (ValueError, KeyError, TypeError, OSError, RuntimeError) as e:
                 logger.warning(f"Skipping invalid YAML file '{yaml_path}': {e}")
 
         # 加载 SKILL.md 文件
@@ -66,7 +66,7 @@ class SkillLoader:
             try:
                 skill = self.load_from_skill_md(md_path)
                 skills.append(skill)
-            except Exception as e:
+            except (ValueError, KeyError, TypeError, OSError, RuntimeError) as e:
                 logger.warning(f"Skipping invalid SKILL.md file '{md_path}': {e}")
 
         return skills
@@ -144,7 +144,7 @@ class SkillLoader:
                 from importlib.metadata import entry_points as _entry_points
 
                 eps = _entry_points().get(group_name, [])
-        except Exception as e:
+        except (ValueError, KeyError, TypeError, OSError, RuntimeError) as e:
             logger.warning(f"Failed to discover entry_points for group '{group_name}': {e}")
             return skills
 
@@ -192,7 +192,7 @@ class SkillLoader:
                 logger.info(
                     f"Loaded skill '{skill.name}' v{skill.version} from entry_point '{ep.name}'"
                 )
-            except Exception as e:
+            except (ValueError, KeyError, TypeError, OSError, RuntimeError) as e:
                 logger.warning(f"Failed to load skill from entry_point '{ep.name}': {e}")
 
         return skills
@@ -208,6 +208,6 @@ class SkillLoader:
                 tool = self._tool_registry.get(tool_name)
                 tools.append(tool)
                 logger.info(f"Bound tool '{tool_name}' to skill '{config.name}'")
-            except Exception as e:
+            except (ValueError, KeyError, TypeError, OSError, RuntimeError) as e:
                 logger.warning(f"Failed to bind tool '{tool_name}' to skill '{config.name}': {e}")
         return tools