fix(skills): curator 自我进化周期调度 + 并发/配置/安全修复

调度:
- app.py 新增常驻 curator 调度器,每小时检查一次"需不需要进化",
  替代原来"仅启动跑一次";多 worker 下用跨平台文件锁做 leader 选举,
  只有一个进程跑调度;启动延迟 5 分钟避开启动高峰

curator 逻辑修复:
- C1: should_run_now / apply_automatic_transitions / get_curator_status
  调用 _get_curator_config 时传 user_id,使 per-user 配置在运行时生效
- C3: usage.py 的 .usage.json 读-改-写加进程内锁,bump_* 改回调式,
  避免 curator 多线程并发丢状态更新
- C4a: curator 受限版 skill_manage 在工具层屏蔽 delete,
  堵住绕过 skill_archive 安全网删技能的口子
- C4b: 进化记录同时保存 LLM 自述(suggestions)与真实工具调用(actual_actions)
- 无候选技能时跳过 LLM 评审,省掉无意义的模型调用

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
xuhdandClaude Opus 4.7 committed 2026-05-15 19:31:05 +08:00
1 parent 3b951efd13
commit d592e1e535
3 files changed
+278 -45

No files matched your search

@@ -1,6 +1,7 @@
import asyncio
import logging
import os
import sys
from collections.abc import AsyncGenerator
from contextlib import asynccontextmanager
@@ -268,6 +269,93 @@ async def _run_db_migrations() -> None:
logger.exception("Database migration failed — schema may be out of date")
# Skill curator 调度参数。
# - 检查间隔:每隔这么久检查一次"需不需要进化";真正是否执行由 should_run_now() 判定。
# - 启动延迟:启动后先等这么久再首次检查,避开启动高峰。
_CURATOR_CHECK_INTERVAL_SECONDS = 3600 # 1 小时
_CURATOR_STARTUP_DELAY_SECONDS = 300 # 5 分钟
def _acquire_singleton_lock(lock_path):
"""跨平台非阻塞独占文件锁,用于多 worker 下的 leader 选举。
成功 -> 返回持有锁的文件句柄(调用方须保持其存活,直到进程退出/主动释放);
抢不到锁(已有其它 worker 持有)或打开失败 -> 返回 None。
进程异常退出时 OS 会自动释放该锁。
"""
try:
lock_path.parent.mkdir(parents=True, exist_ok=True)
handle = open(lock_path, "a+")
except OSError:
logger.exception("Curator scheduler: 打开锁文件失败")
return None
try:
if sys.platform == "win32":
import msvcrt
handle.seek(0)
msvcrt.locking(handle.fileno(), msvcrt.LK_NBLCK, 1)
else:
import fcntl
fcntl.flock(handle.fileno(), fcntl.LOCK_EX | fcntl.LOCK_NB)
except OSError:
handle.close()
return None
return handle
def _release_singleton_lock(handle) -> None:
"""释放并关闭 _acquire_singleton_lock 拿到的锁文件句柄。"""
try:
if sys.platform == "win32":
import msvcrt
handle.seek(0)
msvcrt.locking(handle.fileno(), msvcrt.LK_UNLCK, 1)
else:
import fcntl
fcntl.flock(handle.fileno(), fcntl.LOCK_UN)
except OSError:
pass
finally:
try:
handle.close()
except OSError:
pass
async def _curator_scheduler_loop() -> None:
"""常驻后台任务:每隔 _CURATOR_CHECK_INTERVAL_SECONDS 检查一次需不需要进化。
多 worker 部署下用文件锁做 leader 选举 —— 只有抢到锁的那个 worker 进程真正
运行调度,其余 worker 直接退出,避免 curator 被重复触发、重复调用 LLM。
被 cancel 时 CancelledError 会穿过 ``except Exception``(它是 BaseException
子类,不会被捕获),协程随之干净退出并在 finally 里释放锁。
"""
from deerflow.config.runtime_paths import runtime_home
from deerflow.skills.curator import maybe_run_curator
lock_handle = _acquire_singleton_lock(runtime_home() / ".curator_scheduler.lock")
if lock_handle is None:
logger.info("Curator scheduler: 另一个 worker 已持有调度锁,本 worker 跳过")
return
logger.info("Curator scheduler: 取得调度锁,本 worker 负责 curator 调度")
try:
# 启动后延迟一会儿再首次检查,避开启动高峰
await asyncio.sleep(_CURATOR_STARTUP_DELAY_SECONDS)
while True:
try:
await maybe_run_curator()
except Exception:
logger.exception("Curator scheduler tick failed (non-fatal)")
await asyncio.sleep(_CURATOR_CHECK_INTERVAL_SECONDS)
finally:
_release_singleton_lock(lock_handle)
@asynccontextmanager
async def lifespan(app: FastAPI) -> AsyncGenerator[None, None]:
"""Application lifespan handler."""
@@ -302,12 +390,12 @@ async def lifespan(app: FastAPI) -> AsyncGenerator[None, None]:
except Exception:
logger.exception("Skill ownership migration failed (non-fatal)")
# Kick off skill curator in background if the run interval has elapsed
# 启动 skill curator 调度器:常驻后台任务,每小时检查一次需不需要进化
# (per-user interval_hours 仍是真正决定要不要跑进化的闸)。
try:
from deerflow.skills.curator import maybe_run_curator
asyncio.create_task(maybe_run_curator())
app.state.curator_scheduler_task = asyncio.create_task(_curator_scheduler_loop())
except Exception:
logger.exception("Skill curator startup task failed (non-fatal)")
logger.exception("Skill curator scheduler failed to start (non-fatal)")
# Start IM channel service if any channels are configured
try:
@@ -335,6 +423,17 @@ async def lifespan(app: FastAPI) -> AsyncGenerator[None, None]:
yield
# 停止 curator 调度器
curator_task = getattr(app.state, "curator_scheduler_task", None)
if curator_task is not None:
curator_task.cancel()
try:
await curator_task
except asyncio.CancelledError:
pass
except Exception:
logger.exception("Curator scheduler shutdown error")
try:
scheduler_service = getattr(app.state, "scheduled_task_service", None)
if scheduler_service is not None: