Hermes Agent Masterclass: 4. Skills

So, welcome back to Hermes agent master class. This is module four and this is is going to be on skills which are a critical part of any agent.

If you watched module 3, you know that Hermes remembers. It has built-in memory, session search, honcho, obsidian, other types of memory plugins as well. Memories is what Hermes knows. Skills are what Hermes does. And the agent already has skills. When you install Hermes, it already has around 80 of them in this Hermes/skills file directory, and those are just ready to go from the start. But there's also a lot of optional skills and skills you can actually build and develop yourself, as well as the auto skill function that Hermes is kind of known for.

So, this is going to be today's video. And today, we're going to be talking about the structure of a skill. I'm going to show you a couple of the kind of bundled skills in different charact categories, show you what they do, and then I'm going to show you where to find them or develop them on your own. And then we're going to talk a little bit about the skills curator, which is a newer function. And lastly, we're going to build using our enhanced uh quote project that I've been building on from episode to episode. We're going to use a skill that'll be helpful for that project.

And if you like this video, please subscribe to my free weekly newsletter where I give my honest thoughts about the week's AI news that I can't share here, as well as interesting research and papers that I found and what projects I've been building behind the scenes. There's a link in the description or go to onchain.com. Every word is written by me, mistakes and all. So, if you're sick of AI slop articles or all hype with no substance, subscribe and give AI Garage Weekly a shot. Now, back to the video.

Part one is the anatomy of a skill MD file. Because the thing is a skill is its skill MD. That is the base form of any skill. It's the one file you need to actually have a skill. And it's just a plain markdown file. There's no registration, no build step, no compiler. The agent just reads it like instructions. So you could see kind of a breakdown of how this looks like in your directories. It'll be under your slash Hermes. Hermes slashskills slash whatever category it happens to be. This one is research and then archive which is the name of this skill. So you can look inside this directory and see you're always going to see like I said a skill MD file. This is the skill itself and then there are other kind of optional directories. There's a reference directory for context that the agent can pull on demand. templates for boilerplate scripts for executable helpers and assets for binary stuff. So if you drop this skill directory into the Hermes/sklls directory, the skill exists the next time the agent boots or if you run reload skills, it's there and you can use it.

Looking a little closer on the skill MD file, this is the front matter that it has and there's two required fields. the name this is archive and then as description search archive papers by keyword author category or ID there's a bunch of other um kind of optional ones there's platforms version metadata tags a bunch of other stuff you can see if you want to see all of the bundled skills this is the uh new research GitHub Hermes agent where you install it to begin with but you can actually come here and look at all the skills and all the specific skillmd files. So this is the archive skill as well. So this is the front matter. You can see it has description, version, author, license, platforms, and then metadata. You can kind of take a look at what it has down here. This is all going to vary based on the skill itself. Only these first two items are required, but you'll see items like when to use, procedure, pitfalls, verification, and the agent will read this body on demand.

So the key thing about description here is that description is the picker hint. It's what the agent sees in the skills list. The when to use instruction in the body section is an expanded trigger context. The agent reads after it's already picking up the skill. So if you have a concrete description, the skill will fire whenever you want to use it. If you have a vague description, the skill never gets picked and the body section never matters. So very important to make sure that the description here is very clear and accurate to what the skill does so that the agent can actually pick it up.

Give you another quick example. This is the Google Workspace skill, the skill MD which comes bundled. So description very clear, right? Gmail, calendar, drive, docs, sheets via GWS CLI or Python. So it tells you exactly what platforms it's going to be useful for. And you can look down, you see it has references, scripts that help execute code and then first time setup and some triage. A lot of times skill MD files will have this kind of troubleshooting element as well. So you can see all the instructions there.

So here's the thing that makes the whole system work and it's called progressive disclosure. If you take a look at the red uh naive approach, which would be to load all 80 plus skill bodies into context every turn, which obviously creates token bloat, you can't ship 80 skills that way. So the Hermes approach here in the green, you can see the agent only sees the title and description in something called a skills list. When it picks a skill, only that skill's body load. Everything else stays on disk, which is necessary to keep the context reasonable. And once again, this is why the description is so important in your skillmd file. If it's too vague, it'll never get picked cuz that's the only thing your agent can see.

Now, let's start looking into more actual skills that come with your Hermes agent. They're always adding more. So, this is as of version 14, I believe we're up to, and these are all 24 categories. As you can see, it's around 80 different bundled skills. It's a little closer look, but some of these you'll probably never touch if it's not relevant to what you're doing. Uh others you'll be using every day. So, you can take a second, take a look at all these here. And like I said, you can check the GitHub. You see on the side here under the skills directory, you have all these different categories as well. So, you can look into these skills if you want to see if they can help be helpful for you. But for today's master class, we're just going to look at five of them. Uh creative, GitHub, MLOps, productivity, and research.

So this is research category. And there's a bunch of others, but research is a key function for almost everybody working with agents, right? We need to read papers, read markets, and just generally do research. A lot of my Chrome jobs are all based around research. Here are a couple of them. Archive, which I talked about before. This hits the free archive API. Archive is uh for research papers. LM wiki. And this is the Carpathy style wiki. It works as an obsidian vault. And uh blog watcher here. It tracks RSS summarize new posts. Poly market uh which pulls prediction market data and research paper writing. So, this does structured paper drafting with citations. And there are lots of other optional skills as well.

Okay. So, I'm going to show you this skill um here with my Hermes agent and I'm going to ask it uh research recent papers about recursive reasoning models. And you can see it automatically goes to a couple skills here. Skill one is this archive which I showed you before and then paper summary which is another skill uh to read some of these and then give me a summary here. And while it's doing that if you want to check out your skills that you have uh Hermes dashboard here you can see them all in one easy place here. This is the web dashboard that just runs locally. If you go to the side here in the skills, you can kind of see all the ski u skills you have on this specific Hermes agent. So if I went down to research here, I could see um that I have archive and then paper summary is also another research skill which summarizes academic papers from hugging face or archive. So, this is another way to kind of see what you have on your specific Hermes agent because sometimes you may not it might just create skills without you really realizing unless you're looking closely at all the tool calls or it may install some skills it needs based on whatever task you give it.

Okay, so it's producing my answer now just to show you that it actually worked. These are all the different uh papers that it came back. So those skills just help with this kind of research, doing it faster and more efficiently, finding the proper information, and making sure it doesn't fall into any kind of common problems or issues.

So the next section we're going to kind of look at is the GitHub and dev uh categories. And obviously, if you're doing any kind of development work, you're going to be using GitHub a lot. So having these kind of skills are very useful just to man manage your GitHub repos. So this is kind of combining I'm kind of cheating here but two categories the GitHub and software development which has a lot of different kind of development related uh skills you can use. Uh but for this category you can see here we have GitHub PR workflow and this is the headline full life cycle branch through merge in one skill really useful GitHub code review and GitHub repo management that kind of round out different uh GitHub skills to review diffs or issues or anything like that. Then uh on the software development sides we have systematic debugging which is hypothesisdriven diagnostic methodology. Writing plans which makes an agent emit a plan before code which is very big for kind of complicated largecale tasks or projects. Sub aentdriven development is for delegation patterns and all the GitHub ones use the GH CLI when it's available but they'll fall back to the git plus REST API if not so they work on any machine with git.

So I have here I have this GitHub repo uh when my older ones on TurboQuant but let's give it this. So I said, can you check um this the TurboQuant PyTorch repo and tell me any outstanding issues? And you can already see it's checking using these skills right here, the GitHub issues and then codebase inspection skills. So these are kind of the GitHub just an example of GitHub skills that you could use. And if we look here uh GitHub issues, we look here GitHub issues and skillmd sees what it does. create tri edge label assign GitHub issues. So that is because I specifically asked about GitHub issues and have this nice little management skill to view and create issues, manage issues. Yeah. So it has a lot of instructions for all of these. And there you go. It gave me all the results, all the open issues and um what needs to be done for all of them. So you can see recommended action list and this is all using the GitHub issues scale.

Next one I wanted to look at is one that perhaps most people may not need but I do a lot of kind of local LM operations. So MLOps, machine learning operations is an area lots of skills that I use. And if you're running local LMS, which I know a lot of people are with Hermes agent, uh you're going to be using some of these skills certainly. There's not a ton in terms of uh bundled skills. Most of them are going to be has very lean core bundled. Most of them are actually going to be kind of optional skills because there's so many different options you can go with machine learning ops. And if we look at the the MLOps directory on uh the new research GitHub, you can see they're categorized even further into evaluation, hugging face hub, inference models, research, training, vector databases. So a lot of when you see my machine learning uh videos, a lot of those are based on training training skills and we can see these are optional ones but uh DFT PFT accelerate pietorch simp um so you're going to use a lot of these if you're doing any kind of local training or fine-tuning or stuff like that. So, if you want to see full like a full training section, you can check out my machine learning episodes. I use lots of those skills. But just for this, I'm going to see if I do can you find three model options for running a local model on my RTX 3060 locally using llama.cpp.

Okay, you see it did use the llama-c skill, uh, which is one of the the machine learning op skills. You can see it's down here under inference. uh you go MLOps inference and then llama.cpp or slashcp uh which is one way perhaps the best way to run local models with a lot of different modifications and stuff like that. So this is the skill um it can be used for actual inference but in this case we're just doing repo discovery which was kind of my question asking for a model that would fit me well. So you can see here it came back with the answer fairly quickly. Uh so the best default quen 3 8B. So this is the model it recommends and gives me different kind of options of how to run it. See running it here with flash attention. So this was more kind of a researchbased question but like I said you can do training uh fine-tuning any kind of machine learning tasks locally uh with Hermes agent. It's really good for that. There's a lot of different skills for it.

So the next category we're going to look at is productivity and this is for business systems documents workflows. This is a lot of what you would probably use in your day-to-day work. A lot of times when we're talking about agents and agentic workflows, a lot of it is just about efficiency and productivity, right? So definitely look into what skills fall under this category. You can see a couple here. Google Workspace does Gmail, Calendar, Drive, Sheets, Docs through Python. Uh Notion here uh has a full overhaul in the most recent version for a new developer platform, but you can read and write Notion pages and databases. Very useful if you use that in your work teams. And then there's linear uh this is for issues project cycles via linear's graphql API. Uh nano PDF is for doing uh PDF manipulation locally. PowerPoint and OCR and documents are both for document related work.

So as our example, I think I'm going to try to use this PowerPoint tool. And you can see the full skill here, the PowerPoint skill. Once again, this is their GitHub. Uh you could see this is the preview of the the skillmd file. Tells you when to use it. Basically anytime you're trying to create a a PowerPoint file or not even create, just read it or analyze it, anything. You can see what it has. Design ideas, avoid common mistakes. These are really useful in skills because if the model or agent doesn't know this stuff, it's very easy to fall into these kind of common mistakes. And you can see everything here. Very detailed instructions for how to manipulate PowerPoint files.

So, let's go back into Hermes. And the prompt I'm giving it is, can you turn this into a PowerPoint presentation? And I'm giving it this markdown file. This was my original concept file for the infamous humans among AI um game experiment I did. If you look into um the the PowerPoint skill itself, you see the skillmd file, but this also has kind of this optional scripts uh directory which I talked about earlier. Not every skill is going to have this, but this will have some helper scripts, some executables. These are all Python Python files. um like this one adds a slide, this one cleans uh the material, the files. So having these kind of helper scripts as well is very helpful. And you can see back in Hermes here, the first thing it did is it used the skill PowerPoint, which is what we were talking about.

Okay, so it generated it. So let's take a look at it very quickly. See what we got. There we go. We got a um a PowerPoint presentation. Uh kind of basic. If I was going to actually use this for a presentation, I would tweak some of the visuals, but you can see it took that which was just a markdown file. Didn't have anything visual inside of it. And kind of it actually did try to create some visuals and stuff. So, that's a nice little skill. There's a couple different ways to make slides. I usually just do like an HTML slides, but if you specifically want PowerPoint, it's a nice skill to have.

So, the last category we're going to look at is creative, and this is animation, diagrams, generative, directly relevant to anyone doing visual content. It's kind of the most fun uh group of skills. If you want to take a look, I did a specific video about man video here, um p5 js and touch designer, mcp. So, check out my other on my YouTube channel if you want to take a look at those. They're all pretty good, but just to run down a couple of these. Manom video, um, if you looked at my other YouTube video, but it creates kind of video explainers out of just a text or a prompt. Uh, Scaladraw, which is programmatic handdrawn diagrams. I use that actually in the um LM Wiki Obsidian video as well. It's pretty nice. P5JS creates really nice like art from a single slide or anything you give it. Um, it created a really nice thumbnail for me. Comfy UI, which is stable diffusion via Comfy UI's official CI CLI in REST. This is one of the newer ones. Um, look up I might do a video on Comfy UI in the near future. So, look out for that. Touch Designer. Like I said, I did this in that other video and it creates kind of really cool uh video effects and it's kind of difficult to use. It's kind of an advanced tool, but if you learn how to use it, you can create really cool stuff. humanizer, which is strip AIS isms from generated text. Always necessary if you're doing a lot of AI written text if you don't want a thousand M dashes everywhere.

So for today, I'm going to try I think I'm going to try this pixel art one. I haven't done that. I want to try something I haven't done yet. So we're going to try to create pixel art from my PFP. So, I just the prompt here is can you turn this and that's the the path to the my PFP PNG file into pixel art. So, we'll see if it's able to do that. And yep, it called the pixel art skill. Interesting. There must have been other pixel art skills. Just a second. We can look at this cuz ambiguous skill name. That's pretty interesting because there were three skills probably in the description that referred to pixel art. Um, but then it specified creative, which is the name of the category in the official skills, and then pixel art. Um, cleaner do SNES classic. Just looking in the GitHub um of it, you can see pixel art here as one of the the skills can convert any image into a retro pixel art and then optionally animate it. It's interesting. So it has these this is another one that has scripts. This one actually has references as well. This references file which is pal named pallets. Interesting. So, this is kind of the the different pallets based on the type. We did Super NES. Um, but there's different artistic pallets it goes with. Interesting. So, the scripts as well. This has a bunch of different Python scripts that are going to be used uh to do different functions. And you see pixel art video as well. This is the main skill file though. When to use this also has a when to use. User wants retro pixel art. User asks for NES styling, stuff like that. And it has the full workflow. So, you can kind of see here through all these examples what's in the skillmd file is just instructions. And there's probably some kind of um pitfalls. Yeah. So, this one also has kind of like a pitfalls, troubleshooting, things to avoid, gotas, stuff like that, which seems pretty common in skill files.

Okay, it's done. It's pretty quick, actually, this one. We'll see if it looks decent. There we go. It is pixelated. It probably looks better um if it's not an illustration like this because mine was already kind of basic basic lines, but it definitely is pixelated. If you can zoom in here and see it's more of that kind of style, especially in the hair, you can kind of see it more. There you go. So, that is the creative.

And so, there's ops optional categories as well. And you can see this is how you install an optional skill. These are the ones that don't come with it. Some example are blockchain, uh, finance, health, security, and there's a bunch of them. And we're going to get more into how to find these, but you can look in this um slash skills list, and there's 17 optional categories total that aren't covered in the uh ones you'll find up here in the uh GitHub.

So part three of this video is going to be about the skills hub. And this is where skills come from. So we just walked through bundled and optional skills. And now I'm going to show you the full picture. Every place skill can land in your Hermes install, including one source that gets overlooked, which is the agent itself. So this is where skills come from, and the one we just saw was the bundled skills. There's also official optional. So these are opt-in extras maintained by the Hermes repo. So you can trust them more than you would just any random skill. There's skills.sh, which is Versel's public skill library, and you could search that from the CLI. There's GitHub taps, which is kind of interesting. In version 4, they added this hugging face skills which is really useful but they have default taps as well. Open AI skills, anthropic skills, a lot of the major providers and the hugging face skills is pretty big itself because the community skill index on hugging face surfaces automatically in your Hermes skill browse without any kind of config. Um, and you can own and you can add your own skill taps uh with just Hermes skill skills tap ad. There's also well-known marketplaces, sites like uh sites that publish well-known/sklls is a discoverable source. So you have clawhub uh lobehub clawed marketplace. Many of us when we were first starting out with OpenClaw uh before we found Hermes would use Claw. That was kind of one of the first skill marketplaces. But now there's a ton of options out there. And of course, you can just do a direct URL or just dropped a folder in. Like I said, these are all just directories. So if you have a skill directory, you can just drop it into your Hermes. But obviously, you want to be very careful. There's been a lot of concern and rightfully so about, you know, prompt injection and malicious skills. So, I try to I try to be more conservative and stick to the bundled and stuff that I know comes from a reasonably official source. If it's something that looks interesting but doesn't come from one of these sources, I try to make actually make the skill myself instead of just dropping it in or trying to trust something that I'm not exactly 100% sure about.

So, the install pattern is the same for every source. You do Hermes skill browse and then you can install and there is a security scan that'll run automatically but then the uh skill will be enabled and you have to do this uh slash reload skills which does a hot reload without having to restart your Hermes agent. You can just do it while you're still working there.

And of course there's one more s source. The agent writes its own and this is really a unique aspect of Hermes agent and what really attracted people to it uh when it was first released. So the agent writes skills too. So you probably heard it too about the kind of evolving nature of a Hermes agent and it's really cool but kind of still underappreciated I would say. So the agent can actually write skills and this is super useful. So this is not as a separate mode that you have to enable, not as a feature you have to turn out on, just how the agent works. So you can look here on the left side this how it works section. So the agent has a skill manage tool and it can it's just a tool that the agent can call like web search or file write and the actions are create, patch, edit or delete. Um and a couple file management actions for skills with multiple files. So after every turn, Hermes spawns a background review fork. And this got substantially better in version 12. Uh it's rubric graded and it's scoped strictly to memory and skills tools. So it can't sprawl into spell shell access or web. And that inherits the parents provider model. So it's not using some cheap fallback. So the fork reads the trajectory of what just happened and asks, was this reusable? Is there a skill here? So if yes, it writes a new skillmd to the Hermes uh/skills file directory. And as of version three, it gets marked agent created, which matters because the curator can prune unpinned agent created skills more aggressively than the ones you wrote yourself. So the right panel here in purple is when the when this is triggered. So when it fires and the docs are pretty specific about when it fires. So complex tasks, five or more tool calls completed successfully. Recovery, agent hits errors and figured out a working path. Correction, the user pushed back and the agent updated its approach uh or just a novel workflow it figured out that's worth saving. And this is key. The docs call this the agent's procedural memory. So, both skills that come from the hubs we talked about or from the agents self-created skills, they both end up in the Hermes/skills folder and are both managed by the curator who we're going to talk to talk about next.

So, if you've watched my videos, you've seen my Hermes agent create a bunch of different skills, but let's take a a closer look at some of them. Um, this is the skill I created called Hermes Dashboard Plug-in Development. And if you watch my video about me creating Hermes dashboard plugins, this is the skill it made. You can see I didn't tell it to write it. It just did it automatically as we were building that project. Um, and this was for building a Hermes web dashboard plugin. You can see the description here. very specific build and verify Hermes web uh dashboard plugins including front-end tab bundles fast API plug-in APIs install scripts MCPbacked readonly data sources and dashboard smoke tests so this was in the process when to use and you see it's fairly specific uh use this skill when building reviewing or debugging a Hermes web dashboard plugin so it has all the instructions that you would uh recommended workflows, open- source release checklist. So, a lot of this stuff is very specific to that one uh project. But if I were to build a new plugin, it could use this skill um when it's doing that. it. It's really helpful because especially if you're doing very similar tasks or projects like this to actually go and learn all these instructions um to do it over and over again is just going to take a ton of time and a ton of tokens. If you're paying per token, it's going to cost you a lot of money. But having these skills already built in um just makes it a lot more efficient and a lot quicker.

So, in part four, we're going to talk briefly on plugins. There's some kind of confusion what plugins are versus skills. Sometimes there's been a lot of changes in plugins recently as well. So, I think this will evolve over time, but plugins are just another way to kind of level up your Hermes agent, give it more capabilities. So, there are three types and many different backends that you can use. to plugin types based on the docs and like I said this can change but there's general memory providers which we talked about in the last video and then context engines and these are the specialized backend slots uh image gen model providers platforms video gen a lot of these were in the the config if you remember us setting up like what my image gen was going to be though skills are a separate subsystem plugins can bundle them but skills are their own thing So I asked Hermes itself what's the difference between skills and plugins and there is some crossover here but the short version is that skills teach the agent how to do something. Plugins add new software UI API capabilities to Hermes itself. So just going back to what I said at the beginning skill is mainly just instructions. So here's the procedure commands caveats and verification steps. There are some scripts and stuff involved like we saw earlier. Uh but it's mainly at its core just instructions whereas a plug-in is an executable product surface. So here's new code that Hermes loads and it loads whatever it is UI backend routes other kind of integrations. So it's there's a little bit more to it. So if it changes an agent's knowledge, make a skill. If it changes Hermes actual capabilities and interface, make a plugin. So I hope that clarified. This has been a question myself as well. I'll have a good idea for a project or some kind of tool for Hermes agent and I won't know should this be a skill, should this be a plug-in, should it be something else completely. So, it's important to keep that kind of distinction in mind.

Okay, so part five is the curator and this is a new feature that landed in version 12 of Hermes agent and they enhanced it even more in version 13. It's an autonomous skill maintenance. it the ma agent maintains its own skill library on a schedule and it's pretty cool. So these are the things that it does. It grades, prunes, and consolidates. Uh you can see here it grades every skill in your library against a rubric. It then prunes the dead skills uh anything that hasn't been called recently and isn't pinned. And then it consolidates related skills that have drifted towards duplicating each other. Especially if you use your Hermes agent a lot, it'll sometimes create very similar skills. So, this consolidate is really nice to kind of merge those so you don't have like a bunch of custom skills that all kind of do similar stuff. So, the default uh frequency is a 7day cycle. Um, and this runs on the gateways cron ticker. So, per run reports lands in the uh logs/curator and it's configured under auxiliary.curator. curator in the configuration file. So it's the same way you do configure any auxiliary slot. You can pick a model in the Hermes model uh for the curator. And the version three additions were that it made the manual run synchronous which matters because before that if you ran Hermes curator run you'd wait for the next tick to see the results. So now Hermes curator run returns when it's done. Plus the operator subcomands archive prune list archive status pin. you can drive everything from the command line. So they really upgraded it in the 13 version. Uh so look over here in the green box, you can see pin what you care about and this is important. It's important for safety. Bundled skills and hub installed skills are gated. So the curator can't touch them and that's kind of a defense mechanism. Custom skills, including the one the agent writes itself, can be pinned with Hermes curator pin. And this will lock them away from any kind of modification by the curator. If it's something important you don't want the curator to touch, you can just pin it there and then it'll be just ignored or left alone by the curator. If you do Hermes curator status, you can kind of see what it's done. Um, I've only had two runs on this cuz I don't usually have this agent running 24/7, but here you can actually see a summary of what changed. Um, my Hermes didn't change anything, but you can see what it's done. So that's all about the curator and the other skills in Hermes agent.

So in the last part here, we're going to actually try to build a skill. So the skill is going to be called unit lookup and it's going to consume the HVAC vault we built in the previous module. This is going to be the spec we're working on. The name is going to be unit lookup. The inputs are going to be the model number, uh, contractor name, contractor quoted price. It's going to search the Obsidian vault. If it's not found in the vault, we're going to use web lookup specs. Writes a new vault note image search for the unit photo and compose markdown cards with specs brand notes price image and the output is going to be the markdown card. And this is going to be important because this is going to be part of the enhanced quote itself. So this skill because we're going to be doing this task over and over again, having this as a skill will be really useful. So this is the actual written out spec file. So this is what I'm going to prompt it so you can see the whole thing. It's kind of long. I'm just saying to create a new skill. It's going to be under the business category unit lookup for the HVAC unit research. And it's going to take in like I said these different things. And this is the procedure based on the vault. Uh so this is just the quote based on the spec that I just showed you. So there it is. I put in the prompt. So now Hermes is going to work on creating the skill. You can see it's actually using a skill to create the skill. Hermes agent skill authoring which is another another skill it has.

Okay, so it's done. You can see it created the skill and this is what it looks like an actual skill MD format. You see name, unit lookup, description used when the user provides an HVAC unit model number, contractor name and quoted install price and wants a strict markdown quote card uh with the unit specs etc. So this is the full skill. You see also has a when to use section and the procedure, the instructions. So let's actually try to use this and see how well it works. This also has a pitfall section like a lot of the skills do.

There we go. So, let's try it out. So, I'm just going to give it the prompt. Please check the model carrier this number and create a unit card for it. So, let's see how well the skill you is to do this. And you can see right away use the skill business/unit lookup which is the one we just created. Also using the skill obsidian. And that's for the obsidian the knowledge base that we created in the last episode. Okay, so this model wasn't in the the vault. So it created it. It did the research it needed to do and created the data and actually put it into the vault so that we can reuse it. It also created this kind of markdown card file that has all the information about this specific unit. Here's the card itself. Here's a better look at it. Um, so this has all the data about the brand sear tonnage technology brand notes and everything uh that you would want to see from it with the sources as well. So the idea would be to process a bunch of these different models basically that anything that the company I'm working with carries and then automatically create these cards that have all the information and data so that you can create those kind of enhanced quotes. So this would be a really useful usable skill for this kind of business idea. And you can see if I want to make sure that the curator doesn't touch this, I just do Hermes curator pin and then the category business/unitlookup curator responds pinned the skill will bypass auto transitions. So I can make sure that this skill is not touched.

So two quick notes before we wrap up here. So something about the scope is that um a skill can be limited by platform by the operating system. So it'll hide the skill on incompatible operating systems. And another thing you can enable disable each skill per CLI, Telegram, Discord, etc. So say you don't want to use any of the heavy machine learning skills in telegram uh you can make that specific uh to just the CLI and you can just do that through Hermes skill the interactive TUI and other thing uh agents skills.io Hermes skillmd format is compatible with this standard and the readme states this explicitly and the skill tool code calls it out. So the unit lookup skill that we just created, the same file can run in claude code or cursor codecs, other agents on that standard. So your skills are kind of a portable IP. If you build a really great workflow in Hermes agent and you switch agents or want to try it out in a different type of agent harness, your skill can go with you and that's really helpful.

So that's going to be my module on skills. um a very important part of any agent and Hermes agent has a lot of really cool and unique features around skills that I hopefully explained well enough. Uh we talked about what a skill MD file is. We looked at some of the bundled skills as well as some of the optional ones. Took a quick look at plugins. Uh talked about custom skills that your Hermes agent will make automatically. talked about the curator and also we created our own skill here that we're going to keep using in future videos. So this has been the module on skills and for module 5 it's going to be models and providers. Until now we've been running on whatever the main model I set up back in module one. So module 5 is going to take that apart. We're going to really look into the 20 plus providers, smart routing for cross tiered work, fallback chains, and plus a lot of the big stuff that came in the most recent version. Super Grock Oath with 1 million token context for Grock 4.3 and the OpenAI compatible local proxy that turns your Claude Pro or Claude or Chat GBD Pro subscription into an endpoint for codecs to hit. Plus, we're going to quickly touch on the auxiliary models as well for anyone who didn't see my dedicated video on that. So, a lot of interesting elements and features around models and providers we're going to talk about in the next episode. So, I hope you enjoyed this module on skills. If you're watching this on YouTube, please subscribe, leave a like, leave a comment, and I will see you in module five of my Hermes Agent Master Class.