diff --git a/DESIGN.md b/DESIGN.md
index 6b71e1a..596da64 100644
--- a/DESIGN.md
+++ b/DESIGN.md
@@ -611,7 +611,7 @@ create index on usage_events (model_profile, created_at);
 **数据模型(新表 `scheduled_jobs`,独立加表不碰现有 schema → 公测兼容)**:
 `id, user_id, name, prompt, cron, tz(默 Asia/Shanghai), mode(isolated|persistent), bound_task_id(可空), notify(JSONB 可空), enabled, timeout_seconds, next_run_at, last_run_at, last_status, last_error, last_task_id, consecutive_failures, expires_at(可空), created_at, deleted_at`。Alembic 加表 migration;`usage_events` 复用现成记账(可加 `kind="scheduled"` 自由文本区分,无需 migration)。
 
-**守护循环(仿 §8.4 `_disk_scanner`,plain-asyncio)**:lifespan 起一个后台 task,每 ~30s 扫 `enabled AND next_run_at<=now()`;命中即 `asyncio.create_task(asyncio.to_thread(_run_agent_bg, ...))` 复用现成路径,登记到 `app.state.inflight`(随关停 drain 一起收尾)。与**单活 run 锁**(§7.x `run_status` + `SELECT FOR UPDATE`)交互:isolated 每次新 task 天然无冲突;persistent 若绑定 task 正忙 → 跳过本次 + 记 warn,下一个点再来(不排队堆积)。run 完回写 `last_*` + croniter 算 `next_run_at`。
+**守护循环(仿 §8.4 `_disk_scanner`,plain-asyncio)**:lifespan 起一个后台 task,每 ~10s(`ZCBOT_SCHEDULER_TICK_SECONDS`,只决定最坏延迟≤1tick、不决定会否漏 —— claim 取 `next_run<=now` 的全部)扫 `enabled AND next_run_at<=now()`;命中即 `asyncio.create_task(asyncio.to_thread(_run_agent_bg, ...))` 复用现成路径,登记到 `app.state.inflight`(随关停 drain 一起收尾)。与**单活 run 锁**(§7.x `run_status` + `SELECT FOR UPDATE`)交互:isolated 每次新 task 天然无冲突;persistent 若绑定 task 正忙 → 跳过本次 + 记 warn,下一个点再来(不排队堆积)。run 完回写 `last_*` + croniter 算 `next_run_at`。
 
 **croniter 选型**:存标准 5 段 cron 串 + 时区,`croniter` 算 `next_run_at`。理由:正确处理 dom/dow 同列的 vixie OR 语义和时区折算(手搓极易踩坑,四源都点名这个坑);纯 Python 小依赖。劣选:只支持"每天/每周 HH:MM"自己用 datetime 算 —— 零依赖但遇复杂周期要返工。
 
diff --git a/RUN.md b/RUN.md
index b51571e..4effd6e 100644
--- a/RUN.md
+++ b/RUN.md
@@ -53,7 +53,7 @@
   # SMTP_FROM=you@qq.com         # 可选,默认取 SMTP_USER
   # 定时任务守护循环(DESIGN §8.5,随 web 进程起,plain-asyncio 仿 _disk_scanner):
   # ZCBOT_DISABLE_SCHEDULER=1    # 可选,整体关掉调度(对照 Claude Code CLAUDE_CODE_DISABLE_CRON)
-  # ZCBOT_SCHEDULER_TICK_SECONDS=30   # 可选,扫描间隔,默 30s
+  # ZCBOT_SCHEDULER_TICK_SECONDS=10   # 可选,扫描间隔,默 10s(只决定最坏延迟≤1tick,不影响会否漏)
   # ZCBOT_SCHEDULER_CONCURRENCY=4     # 可选,并发跑的定时 run 上限,默 4
   ```
   > litellm 在 import 时副作用加载 .env;入口走 `main.py`,`.env` 自动生效。直跑 `python -c "from core.storage import ..."` 不经 litellm 链路时记得自己 `import litellm` 触发,或手动 `export ZCBOT_DB_URL=...`。
diff --git a/core/__init__.py b/core/__init__.py
index a0f7bb1..76f08ca 100644
--- a/core/__init__.py
+++ b/core/__init__.py
@@ -1,3 +1,3 @@
 # zcbot 版本号单一事实源:web/app.py 的 FastAPI version、/healthz 返回、前端展示都引这里。
 # 改版本只动这一行。
-__version__ = "0.20.0"
+__version__ = "0.20.1"
diff --git a/scripts/smoke_scheduler.py b/scripts/smoke_scheduler.py
new file mode 100644
index 0000000..e194f50
--- /dev/null
+++ b/scripts/smoke_scheduler.py
@@ -0,0 +1,156 @@
+"""Smoke: 定时任务守护循环端到端(DESIGN §8.5)。
+
+跑法(**需先在另一个终端起 web 服务** `.venv/Scripts/python.exe main.py web`):
+    .venv/Scripts/python.exe scripts/smoke_scheduler.py [--email a@b.com]
+
+干什么:
+  1. 给某用户(默认 DB 第一个 / --email 指定)插一条 isolated 定时任务,
+     prompt 是"回一句早安、不调工具",并把 next_run_at 改成现在 → 让守护循环下一 tick 就认领。
+  2. 轮询 scheduled_jobs.last_status 直到翻成 ok/error/skipped(超时 180s)。
+  3. ok 则打印它新建的 task_id + agent 实际回复片段,证明全链路(认领→建 task→
+     _run_agent_bg→LLM→回写 run_status→record_result)走通。
+  4. 收尾软删该 job(留下 task 供查看)。
+
+**会真的发起一次 LLM 调用**(一句短回复,费用可忽略)。不测邮件 —— notify 投递另行验。
+"""
+from __future__ import annotations
+
+import os
+import sys
+import time
+from datetime import datetime, timezone
+from pathlib import Path
+from uuid import UUID
+
+ROOT = Path(__file__).resolve().parent.parent
+sys.path.insert(0, str(ROOT))
+
+# Windows 控制台默认 GBK,打印中文会乱码 → 强制 stdout UTF-8(只影响本脚本打印)
+try:
+    sys.stdout.reconfigure(encoding="utf-8", errors="replace")  # type: ignore[attr-defined]
+except Exception:
+    pass
+
+# 读 .env
+env_file = ROOT / ".env"
+if env_file.exists():
+    for line in env_file.read_text(encoding="utf-8").splitlines():
+        line = line.strip()
+        if not line or line.startswith("#") or "=" not in line:
+            continue
+        k, _, v = line.partition("=")
+        os.environ.setdefault(k.strip(), v.strip())
+
+from sqlalchemy import select, update
+
+from core import scheduler
+from core.storage import session_scope
+from core.storage.models import Message, ScheduledJob, Task, User
+
+POLL_TIMEOUT = 180          # 秒
+POLL_INTERVAL = 3
+PROMPT = "请直接回复一句『早安,今天也加油!』。不要调用任何工具,不要创建文件,不要做其它事。"
+
+
+def _pick_user(email: str | None) -> UUID | None:
+    with session_scope() as s:
+        if email:
+            return s.execute(select(User.user_id).where(User.email == email)).scalar_one_or_none()
+        return s.execute(select(User.user_id).order_by(User.created_at).limit(1)).scalar_one_or_none()
+
+
+def _last_assistant_text(task_id: UUID) -> str:
+    """取该 task 最后一条 assistant 文本(payload JSONB)。"""
+    with session_scope() as s:
+        rows = s.execute(
+            select(Message.payload).where(Message.task_id == task_id)
+            .order_by(Message.idx.desc()).limit(20)
+        ).scalars().all()
+    for payload in rows:
+        if not isinstance(payload, dict):
+            continue
+        if payload.get("role") != "assistant":
+            continue
+        c = payload.get("content")
+        if isinstance(c, str) and c.strip():
+            return c.strip()
+        if isinstance(c, list):  # 富内容块
+            for blk in c:
+                if isinstance(blk, dict) and isinstance(blk.get("text"), str) and blk["text"].strip():
+                    return blk["text"].strip()
+    return "(未找到 assistant 文本)"
+
+
+def main() -> int:
+    email = None
+    if "--email" in sys.argv:
+        i = sys.argv.index("--email")
+        email = sys.argv[i + 1] if i + 1 < len(sys.argv) else None
+
+    uid = _pick_user(email)
+    if uid is None:
+        print("[FAIL] DB 里没有用户(或 --email 未匹配)。先 main.py user add。")
+        return 1
+    print(f"[..] 用户 {str(uid)[:8]}  prompt={PROMPT[:24]}...")
+
+    # 1) 建 job(cron 随便给个合法值,马上覆盖 next_run_at 为现在)
+    job = scheduler.create_job(
+        uid, name="[smoke] 早安测试", prompt=PROMPT, cron="*/5 * * * *", mode="isolated",
+    )
+    jid = UUID(job["job_id"])
+    with session_scope() as s:
+        s.execute(update(ScheduledJob).where(ScheduledJob.job_id == jid)
+                  .values(next_run_at=datetime.now(timezone.utc)))
+    print(f"[ok] 已插入 job {job['short_id']},next_run 置为现在,等守护循环认领...")
+    print("     (若卡住不动 → 确认 web 服务在跑、ZCBOT_DISABLE_SCHEDULER 未设、tick 已过)")
+
+    # 2) 轮询 last_status
+    deadline = time.time() + POLL_TIMEOUT
+    status = None
+    last_task_id = None
+    last_error = None
+    while time.time() < deadline:
+        time.sleep(POLL_INTERVAL)
+        with session_scope() as s:
+            row = s.execute(
+                select(ScheduledJob.last_status, ScheduledJob.last_task_id,
+                       ScheduledJob.last_error, ScheduledJob.next_run_at)
+                .where(ScheduledJob.job_id == jid)
+            ).first()
+        if row is None:
+            print("[FAIL] job 不见了(被并发删?)")
+            return 1
+        status, last_task_id, last_error = row.last_status, row.last_task_id, row.last_error
+        waited = int(time.time() - (deadline - POLL_TIMEOUT))
+        print(f"     [{waited:>3}s] last_status={status or '(待触发)'}")
+        if status in ("ok", "error", "skipped"):
+            break
+
+    # 3) 结果
+    print("-" * 50)
+    if status == "ok":
+        print(f"[PASS] 守护循环已触发并成功。task={str(last_task_id)[:8] if last_task_id else '?'}")
+        if last_task_id:
+            print(f"       agent 回复: {_last_assistant_text(last_task_id)[:120]}")
+        rc = 0
+    elif status == "error":
+        print(f"[FAIL] 触发了但 run 报错: {last_error}")
+        rc = 1
+    elif status == "skipped":
+        print(f"[WARN] 被跳过(目标 task 正忙?): {last_error}")
+        rc = 1
+    else:
+        print(f"[FAIL] {POLL_TIMEOUT}s 内未触发(last_status 仍为空)。web 服务/调度是否在跑?")
+        rc = 1
+
+    # 4) 收尾:软删 job(留 task)
+    try:
+        scheduler.cancel_job(uid, str(jid))
+        print(f"[..] 已清理 smoke job {job['short_id']}(task 保留可查看)")
+    except Exception as e:
+        print(f"[..] 清理 job 失败(可手动删): {e}")
+    return rc
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/web/app.py b/web/app.py
index 57d48ad..51f004a 100644
--- a/web/app.py
+++ b/web/app.py
@@ -700,11 +700,12 @@ def create_app() -> FastAPI:
         stats_logger_task = asyncio.create_task(_stats_logger(), name="stats-logger")
 
         # ── 定时任务守护循环(§8.5)── 仿 _disk_scanner 的 plain-asyncio 范式,不引
-        # APScheduler/Celery。每 ~30s 认领到点 job(claim+advance next_run 防重复触发),
-        # 复用 _run_agent_bg 起 run,跑完确定性兜底投递 + 回写 last_*。ZCBOT_DISABLE_SCHEDULER=1
+        # APScheduler/Celery。每 ~10s 认领到点 job(claim+advance next_run 防重复触发),
+        # 复用 _run_agent_bg 起 run,跑完确定性兜底投递 + 回写 last_*。间隔只决定最坏延迟
+        # (≤1 tick),不决定会不会漏(claim 取 next_run<=now 的全部)。ZCBOT_DISABLE_SCHEDULER=1
         # 整体关掉(对照 Claude Code CLAUDE_CODE_DISABLE_CRON)。
         scheduler_enabled = os.getenv("ZCBOT_DISABLE_SCHEDULER", "").strip() not in ("1", "true", "yes")
-        sched_tick = int(os.getenv("ZCBOT_SCHEDULER_TICK_SECONDS", "30") or "30")
+        sched_tick = int(os.getenv("ZCBOT_SCHEDULER_TICK_SECONDS", "10") or "10")
         sched_sema = asyncio.Semaphore(int(os.getenv("ZCBOT_SCHEDULER_CONCURRENCY", "4") or "4"))
 
         async def _execute_scheduled_job(snap: dict) -> None: