Heartbeat 的 runOnce 异常隔离与自愈（PR #14901）

• OpenClaw 版本已包含 PR #14901（或同等 heartbeat 调度修复）。
• 你可以访问 heartbeat 相关日志并手动触发一次任务执行。

1. Identify one heartbeat task that occasionally fails and capture its stack trace signature.
2. Upgrade to PR #14901 build and restart gateway in a low-risk window.
3. Inject a controlled failure in runOnce path (or replay known bad input) to confirm scheduler survives.
4. Observe next 3-5 heartbeat cycles and verify subsequent jobs still trigger on schedule.
5. Keep alerting on repeated runOnce errors, but avoid process-level restarts as first reaction.

• Error isolation prevents scheduler death, but does not fix the root cause of the failing task.
• If failures are side-effectful (e.g., partial writes), add idempotency guards separately（需验证）。

This tip is aggregated from community/public sources and preserved with attribution.