diff --git a/src/agentkit/bus/memory_bus.py b/src/agentkit/bus/memory_bus.py
index ddf7a3b..d720c52 100644
--- a/src/agentkit/bus/memory_bus.py
+++ b/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
diff --git a/src/agentkit/bus/redis_bus.py b/src/agentkit/bus/redis_bus.py
index 3f41376..c846e0d 100644
--- a/src/agentkit/bus/redis_bus.py
+++ b/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"
diff --git a/src/agentkit/chat/request_preprocessor.py b/src/agentkit/chat/request_preprocessor.py
index 24704fd..f4c74a2 100644
--- a/src/agentkit/chat/request_preprocessor.py
+++ b/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,
diff --git a/src/agentkit/chat/skill_routing.py b/src/agentkit/chat/skill_routing.py
index 8ae04cd..0f6cff5 100644
--- a/src/agentkit/chat/skill_routing.py
+++ b/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}"
             )
diff --git a/src/agentkit/chat/sqlite_conversation_store.py b/src/agentkit/chat/sqlite_conversation_store.py
index a0a3b39..749be29 100644
--- a/src/agentkit/chat/sqlite_conversation_store.py
+++ b/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
diff --git a/src/agentkit/cli/chat.py b/src/agentkit/cli/chat.py
index f1ced52..0c5b1be 100644
--- a/src/agentkit/cli/chat.py
+++ b/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
diff --git a/src/agentkit/evolution/ab_tester.py b/src/agentkit/evolution/ab_tester.py
index b3a3b2d..9a01572 100644
--- a/src/agentkit/evolution/ab_tester.py
+++ b/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:
diff --git a/src/agentkit/evolution/experience_store.py b/src/agentkit/evolution/experience_store.py
index 5e47d13..642e02b 100644
--- a/src/agentkit/evolution/experience_store.py
+++ b/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
diff --git a/src/agentkit/evolution/genetic.py b/src/agentkit/evolution/genetic.py
index 38d9e4e..57beee1 100644
--- a/src/agentkit/evolution/genetic.py
+++ b/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)
diff --git a/src/agentkit/evolution/lifecycle.py b/src/agentkit/evolution/lifecycle.py
index 1cce14f..d0ef216 100644
--- a/src/agentkit/evolution/lifecycle.py
+++ b/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
 
diff --git a/src/agentkit/evolution/llm_reflector.py b/src/agentkit/evolution/llm_reflector.py
index 91a334a..aaf7fc4 100644
--- a/src/agentkit/evolution/llm_reflector.py
+++ b/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"),
diff --git a/src/agentkit/evolution/pg_store.py b/src/agentkit/evolution/pg_store.py
index aa0571a..ca2b17b 100644
--- a/src/agentkit/evolution/pg_store.py
+++ b/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 []
diff --git a/src/agentkit/evolution/pitfall_detector.py b/src/agentkit/evolution/pitfall_detector.py
index 31a1abd..c50205e 100644
--- a/src/agentkit/evolution/pitfall_detector.py
+++ b/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 []
 
diff --git a/src/agentkit/evolution/prompt_optimizer.py b/src/agentkit/evolution/prompt_optimizer.py
index 2bf9c99..a5dc655 100644
--- a/src/agentkit/evolution/prompt_optimizer.py
+++ b/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)
 
diff --git a/src/agentkit/evolution/risk_guard_learner.py b/src/agentkit/evolution/risk_guard_learner.py
index 06a658f..5a62341 100644
--- a/src/agentkit/evolution/risk_guard_learner.py
+++ b/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 []
 
diff --git a/src/agentkit/memory/adapters/confluence.py b/src/agentkit/memory/adapters/confluence.py
index 1aeb763..3f69a23 100644
--- a/src/agentkit/memory/adapters/confluence.py
+++ b/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
diff --git a/src/agentkit/memory/adapters/feishu.py b/src/agentkit/memory/adapters/feishu.py
index 69ef235..b84d323 100644
--- a/src/agentkit/memory/adapters/feishu.py
+++ b/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
diff --git a/src/agentkit/memory/adapters/generic_http.py b/src/agentkit/memory/adapters/generic_http.py
index 41dd26b..8a9b5c3 100644
--- a/src/agentkit/memory/adapters/generic_http.py
+++ b/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
diff --git a/src/agentkit/memory/contextual_retrieval.py b/src/agentkit/memory/contextual_retrieval.py
index aad0388..c1ae556 100644
--- a/src/agentkit/memory/contextual_retrieval.py
+++ b/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 ""
 
diff --git a/src/agentkit/memory/document_loader.py b/src/agentkit/memory/document_loader.py
index d991221..26a5c9a 100644
--- a/src/agentkit/memory/document_loader.py
+++ b/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)
 
diff --git a/src/agentkit/memory/embedder.py b/src/agentkit/memory/embedder.py
index 6c47c39..c12e04f 100644
--- a/src/agentkit/memory/embedder.py
+++ b/src/agentkit/memory/embedder.py
@@ -8,10 +8,8 @@ import os
 import time
 from abc import ABC, abstractmethod
 from collections import OrderedDict
-from typing import TYPE_CHECKING
 
-if TYPE_CHECKING:
-    import httpx
+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
 
diff --git a/src/agentkit/memory/episodic.py b/src/agentkit/memory/episodic.py
index ba8a4c5..6110311 100644
--- a/src/agentkit/memory/episodic.py
+++ b/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
diff --git a/src/agentkit/memory/http_rag.py b/src/agentkit/memory/http_rag.py
index 9f170c8..af09bbf 100644
--- a/src/agentkit/memory/http_rag.py
+++ b/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:
diff --git a/src/agentkit/memory/local_rag.py b/src/agentkit/memory/local_rag.py
index 8820131..817ea8c 100644
--- a/src/agentkit/memory/local_rag.py
+++ b/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
diff --git a/src/agentkit/memory/profile.py b/src/agentkit/memory/profile.py
index b7050d5..963048a 100644
--- a/src/agentkit/memory/profile.py
+++ b/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)
diff --git a/src/agentkit/memory/retriever.py b/src/agentkit/memory/retriever.py
index 01d9110..0fc3f20 100644
--- a/src/agentkit/memory/retriever.py
+++ b/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": []}
diff --git a/src/agentkit/quality/cascade_state_store.py b/src/agentkit/quality/cascade_state_store.py
index 9adde99..b45fc43 100644
--- a/src/agentkit/quality/cascade_state_store.py
+++ b/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:
diff --git a/src/agentkit/server/app.py b/src/agentkit/server/app.py
index ce2d3b9..60373c8 100644
--- a/src/agentkit/server/app.py
+++ b/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}")
diff --git a/src/agentkit/session/store.py b/src/agentkit/session/store.py
index 7199370..8f6c306 100644
--- a/src/agentkit/session/store.py
+++ b/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()
diff --git a/src/agentkit/skills/loader.py b/src/agentkit/skills/loader.py
index 941fda2..67d5e12 100644
--- a/src/agentkit/skills/loader.py
+++ b/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