Hermes Agent: Autonomous SEO engineer for 11ty

This is a submission for the Hermes Agent Challenge.

What I Built: The End of Manual SEO Era

Let's be honest: we all love the speed and security of the Jamstack architecture. Writing articles in Markdown for generators like Eleventy (11ty) is a pleasure. But there's one thing that every developer-blogger hates. It's routine. 😩

Writing an article is only 50% of the work. Then hell begins: you need to come up with the perfect title, squeeze the essence into 150 characters of description for the Google snippet, and write a short ai_summary. And what if you have hundreds of them?

That's why I decided to go beyond the usual ChatGPT assistants. I created an autonomous digital worker based on the Hermes framework. 🤖

My script doesn't just "generate text" in a browser window. It's deeply integrated into my workflow "in a meaningful way". This agent:

1. Scans the local directory with hundreds of Markdown files on its own.
2. Analyzes the content of each article.
3. Generates perfectly structured JSON with missing SEO metadata.
4. Directly overwrites the Front Matter in files, breaking nothing.

Demo: Autonomy in Action

Instead of building heavy and unnecessary graphical interfaces (GUI) that only consume resources, my demo project is a clean, lightning-fast console automation. It's an invisible back-office worker that manages content right in your terminal.

When you launch the agent, it scans the entire blog in milliseconds, detects "problematic" articles without meta tags, connects to cloud intelligence, and autonomously closes all SEO gaps. The work that previously took hours of manual copy-paste is now done in one click.

(Below in the article, we will detail the terminal logs and the final result, but for now, let's take a look under the hood of this system).

Code: Architecture and Technology Stack

The entire project is built on the principles of idempotence and fault tolerance. This means that the script can be run as many times in a row as you like: it will never create duplicate content, spoil existing tags, or simply continue working from where it left off.

Here is the complete source code and showcase repository for this challenge:
👉 Hermes Jamstack Showcase Repository

To better understand how a local file on your computer communicates with the cloud, take a look at this architectural scheme:

My Tech Stack: Why These Tools?

To create Hermes, I chose the lightest, but time-tested development technologies:

• Node.js (Runtime): A Scandinavian hammer for working with the file system and asynchronous requests. Provides instant directory scanning.
• Eleventy (11ty): Our favorite, super-fast static site generator. Since it works with pure Markdown, it became an ideal environment for automation.
• Gray-Matter (Parser): A special library that can break down a Markdown file into two parts: an object with metadata (Front Matter) and pure article text. Without it, the AI would simply get tangled up in markup.
• Gemini 2.5 Flash API (Cloud Brain): The new generation of models from Google. I chose the Flash version because it has phenomenal generation speed, strict adherence to prompt instructions, and a penny's worth of request cost.

How I Used Hermes Agent: Three Pillars of Autonomy

To integrate the agent into the workflow "in a meaningful way", I had to solve three main engineering problems. I didn't want the script to just chatter - I needed a predictable digital worker.

Here are the three pillars that my Hermes Agent stands on.

1. Structured Data (Taming Chaos)

The main problem with all LLMs is that they love to "talk". If you ask an AI to write an SEO description, it might respond: "Of course, here's your description: ...". For an automated system, such text is a failure and a broken site build.

So I used strict prompt engineering, forcing Gemini 2.5 Flash to return a strictly typed JSON object.

As you can see in the code above, we're not just asking for JSON, we're also additionally cleaning the response from markdown markup (`json), which neural networks love to add. This guarantees that the JSON.parse() method will work without errors. 🎯

2. Idempotence: The Main Rule of a Doctor - "Do No Harm"

Imagine that you accidentally ran the script twice. A bad script would generate duplicates or spoil your manual work. My autonomous agent adheres to the principle of idempotence. 🛡️

Before wasting API credits, Hermes uses gray-matter to parse the Markdown file and check for the presence of SEO fields:

The agent works like a smart filter. If an article already has a perfect description and generated ai_summary - it instantly skips it. If the fields are empty or too short (less than 50 characters) - it precisely fills them in, assembles the file back, and safely saves it to disk.

We also recommend reading: Pressable MCP i Gemini CLI: AI-managed hosting in Linux

3. Fault Tolerance: API Error Handling

During a massive audit of hundreds of files, I encountered the fact that even Google servers sometimes can't handle the load and return a 503 Service Unavailable error.

An ordinary script would "fall" at this point, and the developer would have to search for which file exactly broke. But I have an autonomous agent! It must survive. 🦾

I added a try...catch block directly inside the processing cycle. If Hermes receives a refusal from the servers, it doesn't stop the entire program. It outputs a red warning to the console, understands that the servers are overloaded, makes an emergency pause for 15 seconds (so the API can "breathe"), and calmly moves on to the next file.

This makes the architecture absolutely fault-tolerant. You can run the script on 1000 articles, go drink coffee, and be sure that it will complete the work. ☕

We also recommend reading: SMM Turbo: Hybrid AI editor for Instagram carousels

Final: Mass SEO Audit in Action

This is what the real triumph of automation looks like. When I run the node hermes.js command in the terminal, my autonomous SEO engineer comes to life.

It scans the entire blog directory in milliseconds. Pay attention to the log: it recognizes articles where SEO is already set up and outputs a [SKIP] message. But as soon as it finds a file with an empty description or missing ai_summary, it connects to cloud intelligence, generates JSON, and writes it.

The yellow letters M (Modified) in the left panel of VS Code are the best proof that the script actually physically overwrote files on disk. No manual copy-paste between browser tabs. No broken tags in Front Matter. Just pure efficiency. ✨

Conclusion: Delegate Processes, Not Tasks

Creating this project for the Hermes Agent Challenge completely changed my understanding of artificial intelligence. We are moving from the era of "AI as a chatbot" to the era of "AI as an autonomous employee". 🚀

You no longer need to spend hours filling in meta tags manually. You can delegate AI an entire process, creating such fault-tolerant local scripts. Your Jamstack blog can live its own life and autonomously optimize itself for search engines.

Ready to launch your first agent? 🤖🚀
Jump to the official Hermes Agent Framework page. Explore the documentation, download the source code, and join the autonomous open AI revolution!
👉 OPEN OFFICIAL HERMES WEBSITE

If you want to feel more automation magic, be sure to test our free SMM Turbo service. It's a hybrid AI editor that takes over the routine of creating viral Instagram carousels! 🔥

And how do you use AI in your Jamstack projects? Are you ready to give an autonomous agent access to your local files? Write your thoughts in the comments, it'll be cool to discuss this with the community! 👇