ICDYCT / I Can Do, You Can Too

These days, insights from freeloading hermes stepfun/step-3.5-flash Not that smart, but highly proactive, Development tasks can't be completed correctly in one go. But if you set clear judgment criteria , it can automatically iterate through multiple rounds to complete the task. For example, developing a server-side for file uploads , developing the corresponding client , both of these tasks reported as completed actually had BUGs. But, I let it connect these two results together to test whether the functionality was correct, and it could identify issues on its own, then iterate through several rounds to clean up the BUGs. What? You say the token usage is high, bringing about high cost issues? I agree, but, isn't this freeloading? Group friends say stepfun/step-3.5-flash is talkative, wordy. Yes, so you need to define clear standards, and let it do its thing. Either define what format of file/report to output, and what data should be contained in it. I only look at the output f...

这几天 hermes 白嫖 stepfun/step-3.5-flash 的心得 不那么聪明, 但是能动性强,  开发任务不能一次性正确. 但是如果设置 明确的判断标准 , 可以自动多轮迭代完成任务. 比如, 开发一个上传文件的服务端 , 开发对应的客户端 , 这两个任务报告说完成了其实都有BUG. 但是, 我让TA把这两个结果自己对接起来测试功能是否正确, TA自己能发现问题, 然后多循环迭代几轮就能清理掉BUG. 什么? 你说token用量多, 带来的费用高的问题? 我同意, 但, 这不是白嫖着嘛. 群友说 stepfun/step-3.5-flash 话多, 啰嗦. 是的, 所以要定义明确的标准, 让TA自己动. 要么定义输出 什么格式的文件/报告, 里面要装什么数据. 我只看输出的文件.

Preface In the previous post we tested the capabilities of the Blogger API ,  so what can we do with it? In the world of technology, English-speaking users are also an important part. So let's translate tech articles into English and publish them. Practice Ask hermes Can I see how many blogs I have using the current Blogger API? Answer Ask hermes In https://zelikk.blogspot.com/   found https://zelikk.blogspot.com/2026/03/cpa-cliproxyapi.html Translate to English, publish on https://icdyct.blogspot.com/ Note, keep all HTML tags Get https://icdyct.blogspot.com/2026/04/self-built-cpa-cliproxyapi-aggregating.html ======== Postscript When completing this task, hermes was connected to the free stepfun/step-3.5-flash provided by Nous Portal

前言 上一篇我们测试了 Blogger API 的能力 ,  那么能拿TA干点什么呢? 技术的世界里, 英文使用者也是重要的组成部分. 那么, 把技术文章翻译成英文发布吧.

Preface Previously I played around with using the LLM-WIKI skill to analyze relationships between blog posts and generate related recommendations At that time I didn't implement an automated solution, so the final modifications were done manually by hand.

前言 之前玩了 用LLM-WIKI技能分析博客文章之间的关系生成相关推荐 当时没有实现自动方案, 最终的修改是人肉手动的. 思路 Blogger 有 API 可以用. 在 AI "手把手"的陪伴下, 我启用了Blogger API, 设置了 OAuth 权限, 下载了 OAuth 凭据. 然后就开始让 hermes 来操作 blog 先简单测试了一下发布blog. OK 接着测试了一下修改blog, 尾部添加 "相关推荐". OK 实践 添加一些测试文章 接下来, 和 hermes说 假设你处于不知道 https://icdyct.blogspot.com/  上面已经发表了什么内容 的状态, 你做这样一件事情 遍历这个博客的每篇文章 如果文章的url是 表格中的 page, 那么在文章末尾添加以下表格中的 link1, link2, link3 这个表格是 | page | link1 | link2 | link3 | | 页面 | 链接1 | 链接2 | 链接3 | | 页面 |---|------|------|-----| | https://icdyct.blogspot.com/2026/04/test-4.html | https://icdyct.blogspot.com/2026/04/test-5.html | https://icdyct.blogspot.com/2026/04/test-6.html | https://icdyct.blogspot.com/2026/04/test-7.html | | https://icdyct.blogspot.com/2026/04/test-5.html | https://icdyct.blogspot.com/2026/04/test-6.html | https://icdyct.blogspot.com/2026/04/test-7.html | https://icdyct.blogspot.com/2026/04/test-8.html | | https://icdyct.blogspot.com/2026/04/test-6.html | https://icdyct.blogspot.com/2026/04/test-7.html | ...

Using multi-agent in Hermes, want to share OAuth authentication   The two approaches I thought of before both require modifying Hermes's own code   https://zelikk.blogspot.com/2026/04/hermes-agent-auth-json-api-key-refresh.html The changes are quite substantial   I came up with another approach that doesn't modify Hermes code at all   Still copy the default agent's auth.json to each profile agent's workspace   But push back the time that prompts api key / token expiration in auth.json    This way, each profile agent won't update the api key / token Then write a small program and run it as a scheduled task. The content is the operation above Copy the default agent's auth.json to each profile agent's workspace   But push back the time that prompts api key / token expiration in auth.json    Actually, with the information above, it's already enough for you to have a conversation with Hermes to implement the feature. To prevent you from going astray, let me b...

在 Hermes 里面用多agent, 想共用 OAuth 认证   之前想的2个方案都需要修改 hermes 自己的代码   https://zelikk.blogspot.com/2026/04/hermes-agent-auth-json-api-key-refresh.html 改动有点大  

Preface Previously we implemented a SKILL that summarizes 233 scripts knowledge into a wiki, and then provides a Telegram bot as the user interface . In real usage scenarios, we found some detail issues.

前言 之前我们实现了一个 把 233脚本知识总结为wiki, 再提供telegram bot为用户使用界面 的SKILL 在实际使用场景中, 发现一些细节问题.

Preface I often see some blogs that have a "Related Recommendations" section at the end of each post. A few days ago I tried out the LLM-WIKI skill , so I thought about using the LLM-WIKI skill to analyze my blog and create related recommendations for each post. Approach Save my blog pages as WIKI pages, Use the LLM-WIKI skill to lint and analyze these pages, establishing links between them, Count the link counts of all pages, For each page, among the pages it links to, find the top 3 pages with the highest link counts. Practice Initialize the wiki I had a chat with hermes (MIMO v2 pro 2026-4-21) I plan to do this task, help me analyze the specific plan, don't implement it yet, let's discuss first There is a website https://zelikk.blogspot.com/ This website's RSS output contains full content https://zelikk.blogspot.com/rss.xml Use your llm-wiki skill to create a new WIKI Ingest every article from this website into this WIKI Lint and organize this WIKI, and output ...

前言 经常看到有些博客, 每篇文章尾部有个"相关推荐". 前几天 用到了一下LLM-WIKI技能 , 于是想用 LLM-WIKI 技能来分析我的博客, 也给每篇博文搞个相关推荐.

Hermes uses telegram react emoji to indicate work status and task progress Eyes 👀 means "seen" Thumbs up 👍   means "done" This is the same as when I collaborate with colleagues in Teams. To avoid the work group being flooded with low-information noise like "got it", I ask colleagues to use green check mark ✅ to indicate work completed,  and non-green check mark ☑️ to indicate "got it".  (Why say "non-green"? Because in different systems this emoji may appear in a different color -_,-  and when you search for "check mark" in the iOS emoji, another one ✔️ also shows up) React with emoji when receiving a message hermes config set telegram.reactions true Related Recommendations 《Using topics in telegram groups to limit which messages agents receive in Hermes multi-agent collaboration tasks》 《Binding different telegram bots to multiple agents in Hermes-agent》 《Hermes-agent has a strong ability to modify its own code》

Hermes 用 telegram react emoji 表示工作状态 任务进度 眼睛 👀 表示看到了 拇指 👍   表示完成了 这和我在 Teams 中与同事合作一样. 为了避免工作群中充斥"收到"这类没有信息含量的打扰, 我要求同事 用 绿色对勾 ✅ 表示工作完成,  用 非绿色的对勾 ☑️ 表示"收到".  (为什么说非绿色呢, 因为在不同的系统中, 这个emoji有可能是其它颜色 -_,-  而且ios系统emoji用 "对勾"搜索, 出来的还有另一个✔️ ) 收到消息时 react emoji hermes config set telegram.reactions true 相关推荐 《在 Hermes 多 Agent 协作任务时 telegram group 用 topic 限制 agent 接收消息》 《Hermes-agent 在Telegram 使用多Agent 绑定不同的telegram bot》 《Hermes-agent 修改自己代码的能动性挺强的》

Preface Based on the previous successful multi-agent team collaboration experiment , I invited group friends to give me some small tasks they thought of to try out. Someone provided one like this Build a premium interactive isometric 3D cozy room using Vite + React + Three.js (react-three-fiber/drei), with all objects modeled in code (no external assets) and subtle ambient animations. Users can click objects to smoothly focus the camera and reveal descriptions, delivering a polished, accessible, $50M-startup-quality WebGL experience. Problem During development, I noticed that when agent-test said "test report has been sent to agent-watch. Testing complete", agent-watch never received the report. (right-click open in new tab to view original image) Analysis I used the default agent hermes to analyze this event, to avoid context interference. The analysis conclusion is Conclusion: agent-test hit the maximum iteration count limit and was forcibly stopped by the system. It didn...

前言 基于前面 成功的多agent团队协作实验 , 我邀请群友给我一些TA们想到的小任务试试. 有人提供了一份这个 Build a premium interactive isometric 3D cozy room using Vite + React + Three.js (react-three-fiber/drei), with all objects modeled in code (no external assets) and subtle ambient animations. Users can click objects to smoothly focus the camera and reveal descriptions, delivering a polished, accessible, $50M-startup-quality WebGL experience.

Hermes is based on python, not only does it have higher agency in modifying its own code, but it is also more convenient to submit the modifications you have implemented on this code. Because the code you are running is the one you modified, and you have actually tested the effect of using it. I now need you to do a task with relatively long steps 1. Create a new directory and clone the latest code of your own github down 2. Apply the agent binding topic function obtained from the modifications we previously completed together to your own code onto the code from step 1 Note, do not synchronize all the modifications you have made now, only implement the agent binding topic function (Right-click to open in new tab to view the original image) Commit this modification to the github project main line TA prompted me to create a github token, check the repo permission I operated on github.com and sent the obtained token to TA I want to see how the fork is going Github https://github.com/crazy...

Hermes基于python 不仅是修改自身代码的能动性更高, 而且更方便提交你这份代码上实施的修改. 因为你正在运行的代码就是你修改过的, 而且你实际测试了使用效果.

Preface From Openclaw to Hermes, I've attempted multi-Agent team collaboration many times. https://zelikk.blogspot.com/search?q=%E5%A4%9Aagent Many times the cross-Agent communication has issues, and the failure behavior is non-deterministic. In other words, the same prompt / agent .md file might work fine this time, but fail the next time. The most recent attempt was using Hermes , and I had already had it modify itself to have a feature similar to openclaw that binds agents to telegram group topics . During use,  Once agent-design didn't send a message to agent-watch, but instead replied in its own topic. Twice when agent-design sent messages to agent-watch, it didn't follow the message specification text format. https://www.youtube.com/watch?v=TBmbAcV6Gis Approach From the descriptive level of SOUL.md, a sufficiently intelligent "person" or "model" should be able to understand the effect I want. * The SOUL.md before this experiment had been overwritte...

前言 从 Openclaw 到 Hermes, 我多次尝试过 多Agent团队协作. https://zelikk.blogspot.com/search?q=%E5%A4%9Aagent 很多次 跨Agent通信 出问题, 而且出错的行为是不确定的. 也就说, 同样的 prompt / agent .md文件, 可能这次没问题, 但再来一次就有问题.

Preface Unlike openclaw, hermes does not design a method for agents to communicate with each other (sending messages into the session) in the system framework. It happens that telegram now supports bots sending messages to each other . So let's have hermes's multi-agent team use telegram to communicate Approach Combined with the previously implemented multi-Agent binding to telegram bot We attempt to establish a multi-Agent team for collaboration Deploying the Hermes-agent System Completed installing Hermes-agent Configured the model Set up the telegram channel Set which telegram user IDs are allowed to use it The above steps are omitted. Create telegram group, bots, and agents Refer to this tutorial , set up the telegram group and bots, and test that bots can see each other's messages. Process omitted. Assume the id of this telegram group is -1003918964882 Assume these bots' usernames and tokens are @CrazyAgentWatchBot 1234567890:AAHkMpXv2nQrWsYd8bJtLfCeUo9GiN1KmZw @Cr...

前言 与 openclaw 不同, hermes 在系统框架中没有设计agent 之间交流(往session里发消息)的方法. 正好  telegram 支持了 bot 之间发消息 . 那我们就让 hermes 的 多agent 团队利用 telegram 来沟通

Preface During a Hermes multi-Agent collaboration task, I noticed that when an agent is talking to itself, it sometimes writes the username of another agent's bot, which triggers a response from that other agent. As shown in the image below Practice I tried using negative prompts to restrict it But, the effect wasn't good I remembered that openclaw binds the topics of a telegram group to agents. This way, when an agent talks to itself in its own topic, it won't interfere with other agents. So, I asked hermes to develop this feature for itself 我有一个 telegram group, 建立了topic,  我希望这个 telegram group中, 只有 某个 topic 里的消息 被某个 agent 接收 Through further discussion, some details were confirmed agent 与 topic 一对一 agent 的自言自语也输出到 指定的 topic 中 The final effect is Each agent's config.yaml has the following configuration item telegram:   topic_id: 100  # agent 只收发 topic 100 的消息 For those affected, add the following to the prompt of the agent team 发给其它agent的消息需要指定topic发送. ======== update Gi...

前言 在 Hermes 多 Agent 协作任务时, 发现agent有可能在自言自语时, 写了其它agent的bot的username, 触发了其它agent的响应. 如下图所示

Preface In Hermes multi-Agent collaborative tasks, I only have one Nous Portal account,  and the gateways of these agents are independent, so it is possible that API requests are concentrated in a short period of time, and Nous Portal returns a 429 Too Many Requests error code. Approach Start the next request only after one API request is completed? Use a separate process to control the rate of API requests? Hermes (via mimo-v2-pro model) gave me a better implementation plan Use flock file lock + status file: ~/.hermes/api_rate_limit  →  content: timestamp of the last request When Gateway wants to call the API:   flock(api_rate_limit.lock)     read(last_time)     if now - last_time >= interval:       write(now)           ← Update timestamp       unlock       → Send API request     else:       wait_time = interval - (now - last_time)       unlock       sleep(wait_time)       → Retry No additional process is needed; each gateway just checks before calling the API. All agents share the...

前言 在 Hermes 多 Agent 协作任务时, 我只有一个Nous Portal 账号,  而这些 agent的 gateway 是独立的, 所以有可能在短时间内集中请求API, 被 Nous Portal 返回  429 Too Many Requests 错误码.

Preface Hermes-agent uses the free Xiaomi model provided by Nous Portal, which is logged in via OAuth. When we configure multi-agent and want to use the same OAuth, we encounter the issue of updating the api key

前言 Hermes-agent 使用 Nous Portal 提供的免费 小米模型, 是用 OAuth 的方式登录的. 当我们 配置多agent , 又想使用同一份 OAuth 时, 会遇到 更新 api key 的问题

Basics openclaw uses a single configuration file config.json for the entire system, and multiple agents are written in this configuration file. Official documentation https://docs.openclaw.ai/concepts/multi-agent Hermes-agent's multi-agent approach is different from openclaw. Each agent has its own config.yaml Official documentation https://hermes-agent.nousresearch.com/docs/user-guide/profiles In openclaw's tui, you can use the /agent command to switch agents hermes generates a command-line "command" for a new agent, used just like the original hermes command. For example, you create a new agent/profile hermes profile create agent-watch Then: If you want to configure the model agent-watch model If you want to set YOLO mode agent-watch config set approvals.mode off Each agent (or profile) in hermes has its own complete runtime environment. They only share code logic. Each agent has its own .env, SOUL.md, config.yaml Therefore, each agent has its own telegram bot tok...

基础知识 openclaw 是整个系统用一份配置文件 config.json , 多 agent 是写在这份配置文件里面的. 官方文档 https://docs.openclaw.ai/concepts/multi-agent Hermes-agent 的多agent方案和 openclaw 不一样. 每个 agent 都有一份自己的 config.yaml 官方文档 https://hermes-agent.nousresearch.com/docs/user-guide/profiles

The idea Use the Hermes-agent LLM WIKI skill to build a WIKI about the 233boy script  Expose it to outside callers through the Hermes-agent API Server Drive it from a Telegram bot inside a group

思路 用 Hermes-agent 的 LLM WIKI 技能 做一个 233boy 脚本知识的 WIKI  用 Hermes-agent 的 API Server 功能提供外部调用 用 telegram bot 在群组中工作

Official information source https://core.telegram.org/bots/features#bot-to-bot-communication @BotFather Create a bot, omitted Command /setbot2bot Select bot, set to ENABLE Test Create a group Add all these bots to this group  Set all these bots as admins of the group  Ask AI to write a test program (the blue parts depend on your actual situation) Generate a test program to test whether telegram bots in a telegram group can see each other's messages. Test the two usages: sending messages directly and sending messages via @ username There are several telegram bots, with usernames and bot tokens as follows: @CrazyAgentWatchBot 3942075861:AAHkMpXv2nQrWsYd8bJtLfCeUo9GiN1KmZw @CrazyAgentDesignBot 7150283946:AABxRqNj5vTgKhEa3cWmYsPdFu7LiOz4Xbn @CrazyAgentCodeBot 5827364019:AAEvNuGz3tFwBqXk7yRmJcSh5OaDpI8LlTx @CrazyAgentTestBot 6038195274:AAHkMpXv2nQrWsYd8bJtLfCeUo9GiN1KmZw There is a telegram group with the following id: tg group id -1003918964882 If you used the web version of AI to wri...

官方信息源 https://core.telegram.org/bots/features#bot-to-bot-communication @BotFather 创建 bot , 略 命令   /setbot2bot   选择 bot, 设置为 ENABLE

openclaw has a built-in feature,  agent binds to telegram group topic hermes doesn't have this feature I asked whether it's possible to make the agent only handle messages from its corresponding topic, and hermes just modified its own code. When I ask hermes about some of its features, it's capable of finding the corresponding location in the source code. Indeed, having fewer lines of code does have its benefits. The top half of the image is openclaw https://api.codetabs.com/v1/loc?github=openclaw/openclaw The bottom half of the image is hermes-agent https://api.codetabs.com/v1/loc?github=nousresearch/hermes-agent Github https://github.com/crazypeace/hermes-agent/tree/feat/agent-topic-binding ======== Let's take a look at some other things 《Deploy Hermes-agent on Oracle VPS with Ubuntu system using a non-root user without sudo -i, you can use the official installation script》 《Minimal one-click script is to have an easy-to-learn and read base material, making it conveni...