#Issues that need to be monitored

1 messages · Page 1 of 1 (latest)

clever flower
#

首先,外部环境与基础访问问题影响了正常使用。中国境内用户访问 GitHub 时持续出现不稳定状况,直接干扰代码拉取、版本更新及环境部署。这一网络相关障碍已成为用户安装、更新或维护 Hermes Agent 及相关组件的根本性阻碍。
其次,出现了多项与文件完整性和脚本安全相关的重大风险。多名用户反馈,Hermes Agent 偶发删除文件或错误修改脚本的情况,造成运行时文件不可逆损坏,命令行工具无法正常启动。此类事件不仅导致服务中断,还带来极高的数据丢失风险。此外,升级过程中若终端意外关闭,关键系统文件会损坏或丢失。虽建议用户使用 screen 或 tmux 工具规避此类问题,但这仅为临时解决方案,并非内置安全机制。
第三,配置不稳定及更新过程中的非预期覆盖是最为严重的问题类别之一。执行 openclaw doctor --fix 命令会重置飞书官方插件配置,迫使用户反复重新配置集成参数。同样,Hermes Agent 在升级时会重建并覆盖用户定制化的飞书插件逻辑,清除大量修改后的代码,使用户需重新定位集成节点并重新开发部分系统功能。无需 @提及的群聊触发规则等自定义配置在升级后也会被覆盖。尽管建议用户留存补丁记录,但官方并未提供保留自定义修改或支持可靠补丁回滚的机制。修改后的 Hermes 文件常被官方更新覆盖,原因是 hermes update 命令执行 git 拉取、依赖更新、配置迁移及网关重启时,未对用户自定义内容进行保护。官方文档虽提及冲突处理,却未提供保留或恢复自定义补丁的标准化流程。
第四,迁移与兼容性问题引发功能紊乱。从 OpenClaw 迁移至 Hermes Agent 后,大量用户出现技能混乱,尤其是嵌套式浏览器技能在新环境中不兼容。同时,OpenClaw 自动更新功能近期近乎完全失效,进一步增加运维压力。官方迁移文档仅说明基础导入逻辑及跳过、覆盖、重命名等简易冲突处理选项,未涉及嵌套技能兼容性或自动升级失败后的恢复方案。
第五,性能与资源占用问题亟待优化。Hermes Agent 产生的 API 调用量远高于 OpenClaw,导致令牌与配额消耗速度远超预期。此外,上下文压缩阈值默认设置过低,用户对相关配置的调整往往无法生效,造成对话质量不佳及内存使用效率低下。
第六,多项功能配置缺乏充分的指引与易用性支持。尽管原生支持流式输出,但文档不完善导致大量用户无法成功部署。同样,部分社区反馈的问题在官方仓库与文档中缺乏官方佐证或清晰说明,解决方案主要依赖用户经验,而非经过验证的可复用最佳实践。许多日常日志仅记录现象,未提供可执行的标准化解决方案,加大了故障排查难度。

crude tundraBOT
# clever flower 首先,外部环境与基础访问问题影响了正常使用。中国境内用户访问 GitHub 时持续出现不稳定状况,直接干扰代码拉取、版本更新及环境部署。这一网络相关障碍已成为用...

First, external environment and basic access issues affect normal use. Users in China continue to experience instability when accessing GitHub, directly interfering with code pulling, version updates, and environment deployment. This network-related obstacle has become a fundamental hindrance for users to install, update or maintain Hermes Agent and related components.
Second, several significant risks related to file integrity and script security emerged. Multiple users reported that Hermes Agent accidentally deleted files or incorrectly modified scripts, causing irreversible damage to files during runtime and causing the command line tool to fail to start normally. Such incidents not only lead to service interruptions, but also bring extremely high risks of data loss. In addition, if the terminal is closed unexpectedly during the upgrade process, key system files may be damaged or lost. Although it is recommended that users use screen or tmux tools to circumvent such problems, this is only a temporary solution and is not a built-in security mechanism.
Third, configuration instability and unexpected coverage during updates are one of the most serious problem categories. Executing the openclaw doctor --fix command will reset Feishu's official plug-in configuration, forcing users to repeatedly reconfigure integration parameters. Similarly, Hermes Agent will rebuild and overwrite user-customized Feishu plug-in logic when upgrading, clearing a large amount of modified code, requiring users to relocate integration nodes and redevelop some system functions. Custom configurations such as group chat triggering rules that do not need to be @mentioned will also be overwritten after the upgrade. Although users are recommended to keep patch records, the official does not provide a mechanism to retain custom modifications or support reliable patch rollback. Modified Hermes files are often overwritten by official updates because the hermes update command does not protect

#

user-defined content when performing git pull, dependency update, configuration migration, and gateway restart. Although the official documentation mentions conflict handling, it does not provide a standardized process for retaining or restoring custom patches.
Fourth, migration and compatibility issues cause dysfunction. After migrating from OpenClaw to Hermes Agent, a large number of users experienced skills confusion, especially nested browser skills that were not compatible in the new environment. At the same time, the automatic update function of OpenClaw has almost completely failed recently, further increasing the pressure on operation and maintenance. The official migration document only explains basic import logic and simple conflict handling options such as skipping, overwriting, and renaming, and does not cover nested skill compatibility or recovery plans after automatic upgrade failure.
Fifth, performance and resource usage issues need to be optimized urgently. Hermes Agent generates a much higher volume of API calls than OpenClaw, causing tokens and quotas to be consumed much faster than expected. In addition, the default setting of the context compression threshold is too low, and user adjustments to related configurations often fail to take effect, resulting in poor dialogue quality and inefficient memory usage.
Sixth, multiple functional configurations lack sufficient guidance and ease-of-use support. Although streaming output is natively supported, incomplete documentation prevents successful deployment for a large number of users. Similarly, some issues reported by the community lack official support or clear explanations in official repositories and documents, and solutions mainly rely on user experience rather than proven reusable best practices. Many daily logs only record phenomena and do not provide executable standardized solutions, making troubleshooting more difficult.

latent merlin
#

Well hey do you have any fixes? Are you comfortable writing up PRs?

clever flower
#

Well, these are some issues summarized by our Chinese community.

abstract flare
#

这个整理有价值,不过这里面混在一起的其实是几类完全不同的问题,不能作为一个单一故障一起追。

先说清楚两点。中国境内访问 GitHub 不稳定,这个确实会直接影响拉取、更新和部署,但它属于外部网络与分发链路问题,不是 Hermes 自身在本机上的会话、工具或网关逻辑能直接修掉的。另一类要单独切开的是 openclaw doctor --fix,这个是 OpenClaw 侧的修复路径,不等于 hermes update 自己的行为,不能直接并到同一个结论里。

就 Hermes 这边本身来说,hermes update 确实会做代码拉取、依赖更新、技能同步和网关重启,但它对“直接改 Hermes 仓库源码”这种本地定制并没有承诺长期兼容保护。现在的更新路径会先自动 stash 本地改动,再 pull,再尝试恢复;如果你改的是 Hermes 核心源码,长期稳定的做法也不是直接改 tracked 源文件,而是放到插件、hooks、dashboard 插件或者独立分支/补丁流程里。对应地,bundled skills 的同步逻辑本身是会保留 user-modified 技能副本的,所以如果你们遇到的是“更新后自定义技能或技能逻辑被覆盖”,这个需要拿具体技能目录和更新前后状态来复现,不能先按“全部会被覆盖”来定性。

像“偶发删文件”“脚本被错误修改”“升级中断后关键文件损坏”“上下文压缩配置不生效”“API 调用量明显异常偏高”“迁移后嵌套技能混乱”这些,单靠现象汇总还不够落到可修的问题。这里每一项都需要单独的复现条件和证据,不然最后会变成一个很大的监控清单,但没有哪个点能真正进到修复。

如果你希望把这批问题变成可持续跟踪的事项,下一步最有用的是把它们拆开报。每个问题单独给这几样:出问题时的 Hermes 版本、触发命令、影响到的具体文件或配置路径、更新前后发生了什么变化,以及一份对应的 /debug。涉及源码或插件被覆盖的,再补一下你改的是 Hermes 仓库内文件,还是 ~/.hermes/plugins/~/.hermes/skills/、hooks 这些扩展目录。这样我们才能把“外部环境问题”“OpenClaw 问题”“Hermes update 设计边界”“真实回归 bug”分开处理,而不是全都堆在一个大类里。

crude tundraBOT
# abstract flare 这个整理有价值,不过这里面混在一起的其实是几类完全不同的问题,不能作为一个单一故障一起追。 先说清楚两点。中国境内访问 GitHub 不稳定,这个确实会直接影...

This sorting is valuable, but what is mixed together here are actually several completely different types of problems, which cannot be pursued together as a single fault.

Let me make two points clear first. Access to GitHub in China is unstable. This will indeed directly affect pulling, updating, and deployment. However, it is an external network and distribution link problem and cannot be directly repaired by Hermes's own sessions, tools, or gateway logic on the local machine. Another category that needs to be separated separately is openclaw doctor --fix. This is the repair path on the OpenClaw side. It is not equal to hermes update's own behavior and cannot be directly merged into the same conclusion.

As far as Hermes itself is concerned, hermes update will indeed do code pulling, dependency updating, skill synchronization and gateway restart, but it does not promise long-term compatibility protection for local customization such as "directly changing the source code of the Hermes warehouse". The current update path will automatically stash local changes first, then pull, and then try to restore; if you change the Hermes core source code, the long-term stable method is not to directly change the tracked source file, but to put it into plug-ins, hooks, dashboard plug-ins, or independent branches/patch processes. Correspondingly, the synchronization logic of bundled skills itself will retain user-modified skill copies, so if you encounter "custom skills or skill logic are overwritten after update", this needs to be reproduced with the specific skill directory and the status before and after the update. You cannot first qualify it by "all will be overwritten".

For example, "accidentally deleted files", "scripts incorrectly modified", "key files damaged after upgrade interruption", "context compression configuration does not take effect", "API call volume is obviously abnormally high", "nested skills are confused after migration", etc., the summary of