xiaotianfotos/homerail

I found a concrete and plausible cause for symptom 2.

Before [`71015a8`](https://github.com/xiaotianfotos/homerail/commit/71015a8676438a685c9cd3bb9a4e3a1f6aec446a), `validatePluginSkill()` only performed validation and returned `void`. Starting with that commit, the Manager stores its return value and reads `metadata.description`. Therefore, a new Manager loading an older `homerail-plugin-sdk/dist` would produce exactly:

```
Cannot read properties of undefined (reading 'description')
```

This mismatch can pass TypeScript compilation because the SDK exposes types from `src`, while Node loads runtime code from `dist`; the package version also remained `0.1.0`. The intermittent behavior may indicate that different processes, containers, or images are loading different build artifacts, or that only plugin-backed activities hit this path.

Please try the following:

1. Stop all Manager, Node, Worker, and old container processes.
2. Clean and rebuild in dependency order: `homerail_protoco...

I’ll investigate symptom 2 separately. Thanks for reporting it.

Cold recovery restores persisted active DAG runs after a Manager restart. Nodes that were RUNNING are marked FAILED to avoid unsafe duplicate execution. Therefore, failed=1 means one run was restored but ended in a failed state—not that the recovery process itself crashed. See [recoverAllActiveRuns()](https://github.com/xiaotianfotos/homerail/blob/97497bc/homerail_manager/src/runtime/active-runs.ts#L1187-L1214). The main improvement needed is logging the affected run IDs and reasons.

> > 感谢这条提示词，对新手朋友挺重要的，另外我刚刚开了讨论区，以后有什么想法也可以发在那里
>
> 讨论区在哪里，今天跑起来配置完模型发消息 就报错

是用语音发消息报错吗？

> 感谢这条提示词，对新手朋友挺重要的，另外我刚刚开了讨论区，以后有什么想法也可以发在那里

讨论区在哪里，今天跑起来配置完模型发消息 就报错

感谢这条提示词，对新手朋友挺重要的，另外我刚刚开了讨论区，以后有什么想法也可以发在那里