10 Hermes Agent Skills You NEED To Install Today transcript

I just made this in 1 minute using a single text file. Now my agent never forgets anything I ever tell it. And I have 10 of these running on my agent right now. Three of them cut my agent's costs down, five gave it abilities it did not have at all before, and the last two turned it from one chatbot into an entire team. So, in this video, I'm going to install every single one of them, show you exactly what each does, and show you which three I would install tonight.

By the end of this video, your agent will be doing things it could never do before. Let's get started with three skills that make your agent cheaper to run. The first one is the simplest thing on this list, and it is the one I would install tonight, absolutely tonight, if I were you. Here is the problem it solves. Every time your agent visits a webpage, it reads the whole thing. The navigation bar, the footer, the cookie banner, the sidebar ads, the newsletter pop-up, none of that is what you asked for, and all of it stays in your conversation for every turn after.

Defuddle strips the page down to a clean reader mode markdown before your agent ever touches it. I mean, watch the counter on both of these. This is the same article, same question, yet with Defuddle, we're able to spend a fraction of the tokens we would have otherwise. If your agent does any research at all, this skill is not optional. It pays for itself on the first page you read. And cutting what goes in is only half of it, because the expensive side is what your agent writes back.

This is my favorite thing on the list. Caveman is an ultra-compressed output mode. The repo's own claim is that it cuts output tokens by 65%. So, I ran it. Here is why that matters more than it sounds. Output is the expensive side on basically every model you use. Output costs you several times what input costs, and it does not stop costing you once it is written because that output becomes part of your conversation and gets sent back with every turn after that.

Basically, pay for the same question twice, three times, four times, and so many times over. Here is an example generation without Caveman and with Caveman on the right. The one on the right is not worse. It is the same information you're receiving without the preamble, the recap, the summary of what it just said, and the offer to help with anything else. Your agent has been writing you an essay when you asked for a simple answer.

And the third skill handles the thing that quietly eats more of your conversation than either of those. This is the skill that I started this video with. Quick note before I show you what it does. It is technically an MCP server rather than a skill file. So, do not go looking for a markdown file. I'm including it because the effect is essentially the same. Here is the problem it solves. When your agent needs to understand a code base, it reads files, then more files, then the files those files import.

Every one of those reads lands in your conversation permanently. A 3,000 token file your agent reads at turn four of a 40-turn session does not cost you 3,000 tokens. It costs you 3,000 tokens 37 more times. Codebase memory indexes the whole repo into that graph once. After that, your agent queries the map instead of reading files. Their claim is 99% fewer tokens for exploration across 158 languages. That 99% is their number, not mine, because I ran the same test and got to 95% fewer token usage.

Very close and very very good. And those three skills make your agent cheaper to run. But every skill after this one gives it an ability it did not have at all. All right, this one is the biggest file on the list and it is the only one I'm careful about. Humanizer is 29.6 kilobytes. Everything else here is either tiny or it is a server. This one is genuinely a large file. So, the question with this one is whether what it does is worth the room it takes up.

Because once it loads, the whole thing sits in your conversation for the rest of the session. But what does it buy you? Well, it strips the tells out of AI writing. Somebody scraped the Wikipedia page on signs of AI-generated writing, which is a real page and it is thorough, and turned the whole thing into a skill. It auto-updates when the page updates. You run your agent's output through it and it rewrites it in a natural voice.

The reason I keep it is simple. If you publish anything your agent wrote, this is the last thing between you and getting clocked. I know the second I see an M dash in a LinkedIn post that it is AI-generated. So, my rule with this one is different from the rest. Invoke it late. Invoke it once. And if you are deep in a long session, do the writing pass in a fresh session. And that is your agent sounding human. The next one is about it being able to see anything at all.

Your agent is smart. It can plan, reason, write code, and the moment you ask it to find out what people are actually saying about something, goes blind. X is a paid API. Reddit returns a 403 error the second you hit it from a cloud IP. So, if your agent runs anywhere except the laptop in front of you, ask it to research a competitor and you get nothing back. Agent Reach fixes all of it in one install. X, Reddit, GitHub, and so many more social networks with zero API fees.

And the part that makes it the best one here is the failover. When a platform changes its rules and blocks one path, which happens constantly, it already has a backup route map. Your agent does not even notice the outage. Every other skill so far makes your agent better at something it could already do. This one hands it a sense it did not have before. And that is your agent knowing things. The next one is about it knowing what to do with them.

Here is the one that is going to look out of place on this list, and it is the one closest to money. Marketing skills is 49 skills in one single pack. It has SEO, paid ads, conversion, copywriting, buyer psychology, and so many more. The reason this one earns the space in your arsenal is that these are not tools, they're judgment. Your agent already knows how to write. What it does not know is what a good landing page does differently from a bad one, or why one subject line beats another, or whether your three ad variants are actually testing anything.

This is what is in this pack. If you sell anything, this is the one I would install first, before any of the clever infrastructure stuff. And that is your agent knowing how to sell. The next one is about it being able to actually go ahead and do it. Okay, so your agent can research and it can write. It still cannot touch the tools your business actually runs on. And if you have ever tried to connect one yourself, you know why.

The console is intimidating, the off is confusing, and you end up lost in a settings page at midnight. Now multiply that by every tool you want to connect. Compozio does all of it in one install. Gmail, sheets, Slack, Notion, HubSpot, Salesforce, over a thousand tools that you can connect through Compozio. There's no API key juggling, there's no OAuth flow you have to reverse engineer. This one is a platform rather than a skill file, same as code-based memory that we shared earlier.

So do not go looking for a markdown file, but the effect is the same and it's removed the biggest bottleneck I had. Every new tool my agent needed used to mean a new connection I had to build. And the next one does not reach into anything. It produces. Open Montage is the first open source agentic video production system. 12 pipelines, over a hundred tools, 700 plus skill and knowledge files in one single install. Here is what it does.

You paste in a video you like, the agent reads the transcript, analyzes the pacing, the scene structure, the key frames, and the tone. Then, and this is the part I care about, before it generates a single frame, it comes back to you with two or three different concepts, the exact tool path it plans to take in order to achieve these concepts, a cost estimate, and a preview. You approve the direction before anything gets spent.

On screen right now you're seeing a demo of a video that it has edited itself. Look, I've used a lot of agentic video tools. Almost all of them start burning money and hand you something at the end. This one negotiates first, and everything I have shown you so far assumes one agent doing one job. These last two skills are about what happens when you stop running a chatbot. You ask your agent something, it gives you an answer.

You say that looks wrong and it immediately agrees with you. Then, you say, "You know what? You actually were right." And it agrees with that, too. Your agent can be the most agreeable liar you will ever work with. Oh my cloth fixes that by turning one session into a coordinated team. Decomposes the task into subtasks, assigns specialist agents or external workers, runs them in parallel or in staged pipelines, and then verifies the result instead of stopping at a half-done answer.

For anything irreversible, shipping a feature, a migration, a proposal that goes to a client, this is the quality gate your agent has been missing. And the last one is the biggest project in this entire space, and in this entire video, and it does not get talked about nearly enough. Superpowers has over 260,000 stars on GitHub. That is one of the biggest repos in AI right now. And I almost never see it mentioned anywhere.

It is not a single skill, either. It is a skills framework with an actual methodology behind it for how work gets broken down and executed. It is the closest thing to an operating system for your agent that I have found. Look, I'll be straight with you because this one takes an afternoon to properly understand, but it is worth that afternoon. And that is 10 skills. I do, however want to give you an honorable mention two more skills that I run that I did not include on this list well because they are mine.

On our honorable mentions I'll first mention mission control which is my Hermes agent Fleet dashboard. It has task dispatch agent health live cost tracking one screen for the whole stack. I did a full video on it so I will link it instead of repeating myself. And the second honorable mention is Nova my YouTube skill my YouTube employee. She researches ideas checks my performance data and drafts video ideas for me. This video you're watching right now in fact was suggested for me to film by her.

She's also the reason half this list exists. Every one of those got tested against real work before it made the cut. Anyway, you came here for 10 skills and you got 12. But the thing I want you to leave with is simpler than a list because the list changes every month. Skills are free to install. What costs you is burying the good ones under 40 that you never use. Take the first three skills tonight. They cost you nothing and you will absolutely feel the difference today.

And by the way, I've included every repo I mentioned in this video in the description with the real links. So you can check them all out tonight. Comment below with the first one you are going to install. And if you enjoyed this video, make sure to leave a like or if you're new to my channel then subscribe because I have a ton more content like this coming your way. Oh, and the algorithm gods are predicting that you're very likely to enjoy this video as well.

So click it and I'll see you there.