Hermes Agent Is the One That Actually Gets Work Done¶
I have not written here for a good while. A lot happened in between. A series of events shifted my whole trajectory, and I moved back to Indonesia after almost a decade abroad. I decided to stay and work here, in my home country.
Anyway, back to what this post is about :)
For a long time, generative AI meant one thing to me: open a chat window, type a question, copy the answer, paste it somewhere else, repeat. ChatGPT answered questions well. Claude wrote code blocks nicely. But neither one ever closed the loop. They never ran the terminal command, never edited the file, never scheduled a job, never remembered what I was working on last Tuesday. They answered. I did the rest.
I first tried Hermes Agent after seeing it on X (Twitter). A couple of AI influencers I follow were posting screenshots of terminal sessions where the agent was running full research pipelines and editing code while they did something else. I installed it that evening. Two weeks later, I stopped opening my browser chat tabs entirely.
What Hermes Does That Chatbots Don't¶
Hermes runs in your terminal. It isn't a web interface with a send button. It's an agent that calls tools directly: it reads files, runs shell commands, searches the web, edits code, and pushes commits. The model sits in a loop. You give it a task, it decides which tools to use, it runs them, it reads the output, and it keeps going until the job is done.
The difference isn't flashy. The loop actually closes. When I ask Hermes to refactor a Python module, it doesn't print a plan. It opens the file, rewrites it, runs the tests, and tells me what fixed the linter error. When I ask it to check the status of my cron jobs, it lists them. When I ask it to write a blog post, it creates the file in the right directory, checks the frontmatter, and opens it for review.
The Top 10 Skills People Actually Use¶
Hermes comes with over a hundred ready-made skills. You can also create your own, install community ones, or let it auto-generate skills from your workflows. The Hermes Atlas ranks community skills by GitHub stars. Here are the ten that show up most often and what I think makes each one worth installing.
| Skill | Stars | Why It Matters |
|---|---|---|
| Anthropic-Cybersecurity-Skills | 5.9K | 754 structured cybersecurity skills mapped to frameworks like MITRE ATT&CK and NIST CSF 2.0. If you do any security work, this is your playbook. |
| avoid-ai-writing | 1.3K | Flags and rewrites content to strip AI-isms. A 109-entry word replacement table catches the obvious tells. |
| SkillClaw | 1.1K | Evolves skills collectively in the background. It deduplicates and improves your skill library across sessions. |
| drawio-skill | 901 | Generates draw.io diagrams from natural language. I used it to diagram an air traffic control workflow last week. |
| wondelai/skills | 812 | Modular frameworks for architecture, UX design, and product strategy with a dedicated CLI. |
| pydantic-ai-skills | 272 | Type-safe skill support for Pydantic AI with progressive disclosure that cuts token use. |
| hermes-skill-factory | 179 | Watches your workflows and turns repeated steps into reusable skills automatically. |
| super-hermes | 122 | Teaches Hermes to write its own analytical prompts and identify structural patterns in codebases. |
| litprog-skill | 118 | Literate programming: weaves source code and prose into synced documentation. |
| icarus-plugin | 109 | Extracts training data from your workflows to remember work across agent instances. |
The pattern is obvious: the top skills aren't about generating text. They're about structuring work. Cybersecurity playbooks, diagram generation, skill deduplication, literate programming. These are tools for building systems, not just having conversations.
The Commands I Use For Every Day¶
Hermes is a terminal-first tool, so everything goes through the CLI. These are the commands I use most often after a few weeks of daily use.
hermes¶
Just type this. It starts an interactive session. No flags, no setup. This is my default entry point.
hermes chat -q "task description"¶
Fire-and-forget mode. Good for one-off tasks like "check if port 8080 is open" or "list Docker containers."
hermes setup¶
The interactive wizard. I ran this once on install to pick my model and provider. You can rerun sections anytime if you want to switch providers.
hermes tools¶
Interactive tool manager. I use this to enable or disable toolsets. For example, if I don't want the agent touching the web for a task, I turn off the browser tool here.
hermes skills list and hermes skills install <id>¶
Browse and install skills. I usually search the hub first, then install by identifier. Community skills get quarantined and scanned before installation, which I appreciate.
hermes config set <key> <value>¶
Quick config tweaks without opening a text editor. I used this to enable secret redaction (security.redact_secrets true) and to bump my terminal timeout for longer builds.
hermes doctor¶
Checks dependencies and config. When a tool stopped working last week, this told me my API key was expired. Took ten seconds to fix.
hermes update¶
Pulls the latest version and updates the CLI. I run this whenever I see a new release on the repo.
hermes model¶
Interactive model picker. I switch between OpenRouter and local models depending on the task. This is faster than editing config.yaml manually.
hermes sessions list and hermes sessions browse¶
Persistent sessions mean I can go back. I use this to resume work I started yesterday or export a transcript for reference.
hermes cron create <schedule>¶
Scheduled agent runs. I have one set to search for new GNSS research papers every two hours and append findings to a markdown file. It runs even when I'm not at my desk.
What Changed in My Daily Workflow¶
I do a lot of terminal work: Python scripts, data processing, aviation standards research. Before Hermes, my workflow was split between three places: the terminal for commands, the browser for research and chat, and an IDE for code. Now it's mostly one screen.
The main change: I delegate now. Instead of reading Stack Overflow threads or asking a chatbot for code and pasting it myself, I say "find why this import fails and fix it." Hermes searches, edits, tests, and reports. I verify the change. That's it.
The self-improving part actually works. After I walked Hermes through my GNSS-SBAS research setup once, it created a skill file. Now I load that skill in new sessions and skip the context-building entirely. It knows my directory structure, my naming conventions, and which sources I trust.
Who This Is Actually For¶
Not everyone needs a terminal agent. If your work is mostly email and web apps, a web chat interface is fine. But if you spend time in a terminal, edit files, run scripts, manage servers, or track research workloads, the gap between "answering my question" and "doing the work" becomes obvious fast.
Hermes fits developers, researchers, sysadmins, data analysts, and anyone who hits the same workflow often enough to want it automated. The cron scheduling and skill persistence make it especially useful for recurring tasks that a chatbot forgets the moment you close the tab.
How to Try It¶
Installation is one command on macOS or Linux:
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
Then start the setup wizard:
hermes setup
Pick a model and provider. I use OpenRouter because it gives access to many models with one API key, but Anthropic, OpenAI, DeepSeek, and local models all work. After setup, type hermes and start a session.
If something breaks, run hermes doctor. It will check your config, tool availability, and credentials. I run this whenever a tool acts unexpectedly, and it usually points to the fix in one line.
Final Thoughts¶
Hermes isn't perfect. There is an actual learning curve to using Hermes Agent, because you are not just chatting with a bot. You are working with an agent that can edit your files, run shell commands, and push commits to GitHub. At first you will stare at the approval prompts wondering whether to say yes or no. You will learn by trial and error which tasks to delegate and which to handle yourself. That friction is real.
The fastest way to shorten that learning curve is by learning Hermes with Hermes. Just ask it: "how do I learn Hermes?" or "show me the most useful commands for a beginner." It will walk you through its own features, explain the approval system, and suggest skills to install. I did exactly that in my first session and it saved me hours of reading documentation.
I still use web chat interfaces occasionally for brainstorming or creative writing. But when I need work to actually happen, I open my terminal. Files edited, commands run, jobs scheduled, research compiled. That's what Hermes does.
That's the difference. Hermes gets work done.