Hermes Agent Skills That Make It 10x More Powerful — Transcript
Full transcript
- 0:00Ever since Hermes' Agent was released,
- 0:02people have been setting up all kinds of
- 0:03workflows on it. And to make an already
- 0:06powerful Agent even better, people have
- 0:08also built a lot of skills for it. Some
- 0:10of them have even become some of the
- 0:12most trending projects on GitHub. Now,
- 0:14you might wonder why you'd need to
- 0:15install more skills in the first place
- 0:17because Hermes already comes with a lot
- 0:19of them, and it can even create new ones
- 0:21from your workflows. But, the skills
- 0:22Hermes creates only come from the
- 0:24processes it learns from you. These
- 0:26other skills are built by people who
- 0:28have already solved problems Hermes
- 0:30can't handle on its own. And they've
- 0:31packaged those solutions into skills you
- 0:33can use. Now, if you're new to the
- 0:34channel, then welcome. We're a software
- 0:36company, and this is AI Labs. And in
- 0:38this video, we're going to go over the
- 0:40skills that actually make using Hermes'
- 0:42Agent so much better. But, before we
- 0:44start with the first one, you need to
- 0:45know this. Even though we're using these
- 0:47skills with Hermes, most of them can be
- 0:49used with any AI coding agent. Now, if
- 0:51you've been working with coding agents,
- 0:53even if it's Hermes or literally any
- 0:55other agent, they all have the same
- 0:57problem. If you're having a long
- 0:58conversation, your agent can forget the
- 1:00details you mentioned earlier. So, you
- 1:02have to give it the same instruction a
- 1:03second time. And this happens more often
- 1:06when you're using Opus models with
- 1:07Hermes because they tend to forget
- 1:09things more quickly. This happens
- 1:11because there are too many messages in
- 1:12your context window, and the model has
- 1:14to pay attention to all of them. When
- 1:16that happens, it can lose track of the
- 1:18smaller details, and you get worse
- 1:20answers. To solve that problem, there is
- 1:22a skill called Planning with Files. It
- 1:24keeps the plan where the agent can't
- 1:26ignore it. Instead of keeping the plan
- 1:28in the chat and taking up space in the
- 1:30context window, this skill converts the
- 1:32plan into three files in your project
- 1:34folder. The first one is task plan.md,
- 1:37which basically splits your main task
- 1:39into smaller tasks so the agent can do
- 1:41them one at a time. Then, there's
- 1:42findings.md, which holds every problem
- 1:45the agent faced and how it solved it. If
- 1:47the same problem happens again, the
- 1:49agent can use those learnings to avoid
- 1:51it. And the last one is progress.md,
- 1:53which records how far the agent has
- 1:55gotten. Now, you might think keeping
- 1:57your plan in files isn't new, and most
- 1:59of you might already be doing it. What
- 2:01sets this skill apart is that the agent
- 2:03can't ignore the plan. That's because it
- 2:05uses a hook, which forcefully inserts
- 2:07the three files into the context window,
- 2:09which is why the model can't choose to
- 2:11ignore it. Now, most agents don't
- 2:13support hooks, but Hermes is one of the
- 2:15few that does. And according to their
- 2:17own testing, the agent performed much
- 2:19better with the hook running because it
- 2:20didn't drift away from the task that it
- 2:22was assigned. The install commands for
- 2:24all the agents are listed in the repo,
- 2:26which will be listed in the description
- 2:28below. Once you have it installed, it'll
- 2:30be added alongside all the other skills.
- 2:32And the hooks are going to be
- 2:33automatically installed in your agent.
- 2:35The skill file explains what the three
- 2:37files are, where they should be, and
- 2:39what each one does. It also includes all
- 2:41the rules the agent needs to follow, so
- 2:43it knows what it's working with. So,
- 2:45once you have this installed, you can
- 2:46run the skill and give it a prompt for
- 2:48what you want the agent to do. Here you
- 2:50can see that the agent started by
- 2:52creating the task plan, the findings,
- 2:54and then the progress files as well.
- 2:56Then it asked us questions, and we
- 2:57answered them as we went. After that
- 2:59long session, it finalized the plan and
- 3:01gave us the three files. But before we
- 3:03go into the next skill, it would be
- 3:05great if you subscribe to the channel
- 3:07and hit the hype button. This small
- 3:09gesture of support goes a long way for
- 3:11us. You might already know that you can
- 3:13use multiple coding tools with Hermes.
- 3:15For example, you can connect your code
- 3:17subscription to Hermes agent and use its
- 3:19models. You might also know that Hermes
- 3:21already has a Claude code skill that
- 3:23sends tasks directly to it. And it has
- 3:25the same kind of skill for Codex as
- 3:26well. But the problem with these skills
- 3:28is that Hermes won't use them unless you
- 3:30ask it to. You also have to say which
- 3:32tasks should go to each agent in your
- 3:34prompt. But there is another skill that
- 3:36sends a task to any agent that is
- 3:38already installed on your system. And
- 3:40this skill acts as an orchestrator
- 3:42between all of them. It also figures out
- 3:44which of your installed coding tools is
- 3:46best for which task and gives the work
- 3:48to that tool. This is actually a set of
- 3:50skills. The main one is called delegate
- 3:52setup, which coordinates the other
- 3:54agents. It also comes with dedicated
- 3:56delegation skills for different coding
- 3:57agents, including Claude Code, Codex,
- 4:00and Cursor. These agents run in a mode
- 4:02where they don't ask permission before
- 4:04running any command. So, the coding
- 4:05agent doesn't stop and wait for you to
- 4:07approve every change. To install this
- 4:09skill, run the install command. It asks
- 4:11you to choose from 18 skills, so you can
- 4:14select the tools installed on your
- 4:15system. Once installed, you will see
- 4:17that Hermes has the delegate setup skill
- 4:19ready for use. To use it, you just have
- 4:21to run the skill with the slash command,
- 4:23type in the prompt, and give it to
- 4:25Hermes agent. First, it identifies which
- 4:27coding tools are installed. Then, it
- 4:29gives you a complete map of what area
- 4:31each agent will handle, along with the
- 4:33exact model name and effort level. Once
- 4:35you approve the plan, it starts running
- 4:37the skills. It then sends each task to
- 4:39the agent picked for it. After every
- 4:41agent reports its findings, Hermes gives
- 4:43you one collective report and marks the
- 4:45task as complete. Now, you might already
- 4:47know that your Hermes agent runs
- 4:49terminal commands. Each terminal command
- 4:51is run, it returns an output that the
- 4:53agent has to read. 90% of the time, you
- 4:56don't even need all of the output that
- 4:58command produces. You only need a
- 4:59certain part of it. The rest just bloats
- 5:01the context for no reason. And just like
- 5:03we mentioned before, having too much
- 5:05information in the context window makes
- 5:07it harder for the agent to focus. That's
- 5:09where RTK comes in. It cuts the output
- 5:12of the terminal commands the agent runs
- 5:14and keeps only the parts the agent
- 5:15needs. So, the output gets filtered
- 5:17before it ever reaches the context
- 5:19window. It has a database of the
- 5:21terminal commands agents use most often,
- 5:23so it knows how to filter their output.
- 5:25For example, if you run tests, the agent
- 5:28only needs to see the failed test
- 5:29results and not the passing ones. RTK
- 5:32does that by default and keeps only the
- 5:34failures. But the thing is, it's not
- 5:36actually a skill. It's a separate tool
- 5:38you install on your device and run from
- 5:40the terminal. And it's on this list
- 5:42because it makes sure your agent gets
- 5:43only the right context. If you want, you
- 5:46can also install it directly and tell
- 5:48the agent to use it. But, since agents
- 5:50tend to forget what we ask, we actually
- 5:52created a skill that tells the agent to
- 5:54route commands through this tool instead
- 5:56of running them the usual way. This
- 5:58skill also contains the details the
- 5:59agent needs along with a reference for
- 6:01each command, so it knows which commands
- 6:03it can use and how to use them. You can
- 6:05add this skill to your Hermes setup.
- 6:07Once you've done that, you can give
- 6:09Hermes any task and it will run every
- 6:11command through RTK instead of using the
- 6:13usual method. Only the output the agent
- 6:16needs reaches the context window, so
- 6:18Hermes can give you more accurate
- 6:19answers. But, before we move on to the
- 6:21next one, let's have a word by our
- 6:23sponsor, Plural. Running production
- 6:25infrastructure usually comes down to one
- 6:27thing, the 3:00 a.m. alert and digging
- 6:29through Kubernetes logs to find what
- 6:31broke. Plural fixes that with
- 6:33workbenches. They're AI agents that
- 6:35actually operate your infrastructure for
- 6:37you and they run inside your existing
- 6:38access controls, so nothing happens that
- 6:41you haven't already allowed. Here's what
- 6:42sold us. We wired it into our
- 6:44observability and when an alert fired,
- 6:46the agent didn't just ping us. It
- 6:48investigated on its own, dug through the
- 6:50logs and metrics, and came back with a
- 6:52real diagnosis and clear next steps. And
- 6:54workbench hooks into the tools you
- 6:56already run, like Datadog and Slack, so
- 6:58it can kick off on its own the moment an
- 7:00alert fires or on a schedule you set.
- 7:02And instead of writing prompts from
- 7:04scratch, you hand them ready-made skills
- 7:06from Plural's library. It's basically an
- 7:08on-call engineer that never sleeps. Try
- 7:10Plural at the link in the description.
- 7:12Now, the [snorts] next skill is by
- 7:13Google and it's called Mantis. Mantis is
- 7:16a set of security review skills that
- 7:18your agents can run. It contains a
- 7:20variety of skills, each focusing on a
- 7:21different area of security review. This
- 7:24setup works no matter what app you're
- 7:25building, be it a mobile app or a web
- 7:27app. You can install it and use it for
- 7:29whatever you're working on. The
- 7:30installation command is listed in the
- 7:32GitHub repo. When you run the install
- 7:34command, you will be asked to choose
- 7:35between the many specialized skills it
- 7:37has. It's better to install all of them
- 7:39because they are separate pieces of the
- 7:41whole review process. Once you've
- 7:43installed it in your setup, you'll see
- 7:45all of the installed listed. For
- 7:46example, there is a skill called Mantis
- 7:48Plan, and it plans out your security
- 7:50review process before you actually
- 7:52begin. It also contains a skill called
- 7:54Mantis Reflect that goes back over all
- 7:56the findings and make sure the failures
- 7:58are documented properly. Now, you might
- 7:59think if Hermes is an always running
- 8:01agent and we don't usually code with it,
- 8:03why would we even need to add Mantis
- 8:06there? But, the reason is that once your
- 8:08app is hosted, it needs security reviews
- 8:10again and again, and that isn't
- 8:12something you want to remember to do
- 8:13yourself. And that's what makes Hermes
- 8:15the best place to run them. It has cron
- 8:17jobs, which are basically tasks that run
- 8:20on their own on a schedule. You can set
- 8:22one up that runs this review set against
- 8:24the app you've hosted. And because
- 8:25Hermes is running all the time, it can
- 8:27alert you on Slack or any other
- 8:29connected platform if it finds a
- 8:30problem. We also have a similar setup
- 8:32for our own project. Here, you can see
- 8:34that we gave Hermes our project in the
- 8:36prompt and told it to run the full set
- 8:38over the code. It keeps finding issues
- 8:40and applying fixes, which keeps the app
- 8:42healthy for us and ensures that it
- 8:43doesn't have any bugs. If you've ever
- 8:46tried researching with agents, you might
- 8:47have run into a huge problem. Your agent
- 8:50is denied access to certain content.
- 8:52This happens because a lot of sites like
- 8:54Reddit don't want models using their
- 8:56content. But, those sites are really
- 8:57valuable since they have real users
- 8:59talking about problems and their
- 9:01opinions and experiences, which makes
- 9:03the information useful for research. To
- 9:05give agents access to that content,
- 9:07there is a skill called Agent Reach. It
- 9:09gives Hermes access to the sites that
- 9:11are usually restricted, including
- 9:13Reddit, GitHub, and many others. It uses
- 9:15free methods, so it doesn't cost you
- 9:17anything. It was also the number one
- 9:19trending project on GitHub recently.
- 9:21Now, you can install this in your Hermes
- 9:23setup, and it'll give you access to more
- 9:25than 15 platforms. It can search
- 9:27Instagram, Facebook, Twitter, and
- 9:29Reddit. Once you have it installed, you
- 9:31can use it directly from Hermes and ask
- 9:33it to search for any topic on any
- 9:35platform. It runs its own scripts, finds
- 9:37the relevant information, and then gives
- 9:39you the answers that you need. Now, as
- 9:41[snorts] you already know, Hermes comes
- 9:43with more than 90 skills built in. And
- 9:45aside from that, when Hermes sees a
- 9:47workflow in your chat that could be
- 9:49useful in the future, it actually
- 9:50creates a skill out of it, and then it
- 9:52can use that skill in future runs. It
- 9:54sounds good because it's adapting to you
- 9:56automatically, but there is a problem
- 9:58with this. The only part of a skill that
- 10:00actually goes into the context window is
- 10:02its name and description. They get sent
- 10:04with each of your messages so that the
- 10:06agent knows these skills exist. Now, you
- 10:08might say that the name and description
- 10:10make up a small chunk and wouldn't
- 10:12affect the context window at all. But if
- 10:14you have over 100 skills and you've been
- 10:16using Hermes for a long time, all those
- 10:19names and descriptions start consuming a
- 10:21huge portion of the context window. So,
- 10:23to solve this issue, there's a whole set
- 10:25of skills built to make the agent more
- 10:27productive. But the one we're most
- 10:29interested in is the skill retriever
- 10:31skill. It replaces the full list of
- 10:33skill prompts with only the most
- 10:35relevant skill descriptions, which
- 10:36causes the context size to drop. So,
- 10:38basically, it runs another step before
- 10:41Hermes calls the model. It splits your
- 10:42message into small chunks that a model
- 10:45can understand, does the same to every
- 10:46skill you have, and then matches your
- 10:48message to the skills list. The skills
- 10:50that score highest are the ones it
- 10:52sends. Their docs put numbers on it.
- 10:54They say that with the whole list going
- 10:56in, it used around 11,000 tokens. After
- 10:58this, it uses only 2.3 thousand, so they
- 11:01claim it saves more than 9,000 tokens
- 11:03every turn. Now, this helps you in two
- 11:05ways. First, there's a lot less going
- 11:07into the context window with every
- 11:08message you send, so it costs you less
- 11:10to run. And second, the agent isn't
- 11:12distracted by skills that have nothing
- 11:14to do with what you asked, so you get
- 11:16better answers. This comes as a plugin,
- 11:18and skill retriever is one of the skills
- 11:20inside it. Once you've installed the
- 11:22plugin, you can enable skill retriever
- 11:24from the plugins area. Then the skill
- 11:26list going into the context will be much
- 11:28shorter than it used to be. Now, the RTK
- 11:30skill that we created in this video can
- 11:32be found in AI Labs Pro, which is our
- 11:34community. So, if you found value in
- 11:36what we do and want to support the
- 11:37channel, this is the best way to do it.
- 11:39The link's in the description. That
- 11:41brings us to the end of this video. If
- 11:43you'd like to support the channel and
- 11:44help us keep making videos like this,
- 11:46you can do so by using the Super Thanks
- 11:48button below. As always, thank you for
- 11:50watching and I'll see you in the next
- 11:52one.
About this transcript
This page contains the full transcript of Hermes Agent Skills That Make It 10x More Powerful by AI LABS, generated from the public captions YouTube serves with the video. The transcript has 2,587 words across 369 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.