Hermes /learn: The Learn Anything Engine Explained
Hermes /learn is the learn-anything engine — show the agent something once and it keeps the skill forever. Here’s what it actually is, the five steps behind it, the beliefs that stop people using it, and what to teach it first.
Short answer
One word, any source: /learn distils URLs, PDFs, folders, notes or finished jobs into kept skills.
Five steps: learns from anything, keeps forever, writes its own recipe, loads only when needed, works everywhere.
It captures how YOU work — the part no downloaded skill or MCP server can supply.
Sibling guides: the /learn command (usage) and /learn skills (the file it writes).
What Hermes /learn actually is
Hermes /learn is the learn-anything engine: show the agent something once — a blog post, a PDF, a folder of code, pasted notes, even a job it just finished — and it distils that source into a verifiable, reusable skill it keeps forever.
Before this, agents had a goldfish problem. You’d walk Hermes through a workflow, it would do the job beautifully, and then forget the entire recipe the moment you were done. Next session: explain it all over again. You were the only one who remembered how your own workflows ran.
/learn is the fix. One word, one source, and the recipe stops living in your head and starts living in the agent.
The five steps of the Hermes /learn engine
It learns from anything. URL, document, PDF, folder, notes, or the task it just completed — if you can point at it, it can learn from it.
You teach it once; it keeps it forever. The lesson survives sessions, restarts and everything else.
It writes its own recipe. The output is a clean skill.md the agent authors itself — you don’t write code or config.
It loads only when needed. Skills wake on demand or on trigger terms, so they never bloat your everyday context.
It works everywhere. Dashboard, terminal, Jarvis voice, messaging — one learned skill, every surface.
In my first test I fed it a guide from my own website and it worked first time — learned the system, saved the skill, loaded it on command. The mechanics of running it are in the /learn command guide, and what the saved file looks like inside is in the /learn skill breakdown.
🔥 Want this set up without the guesswork? Teaching your agent its first skill takes five minutes tonight — the roadmap covers which workflows to teach first. Inside the AI Profit Boardroom you get the Agent OS as a ready-to-install file, a 30-day roadmap, daily tutorials the same day new tools ship, and four live coaching calls a week where you share your screen and get unstuck. → Get access here
Why Hermes /learn changes the maths
Count the cost of the old way honestly: a workflow you repeat weekly, re-explained every session, a few minutes each time, hundreds of times a year. /learn makes that five minutes once.
And it’s different from bolting on someone else’s tools. Downloaded skills and MCP servers are generic; /learn captures how you do things — your steps, your rules, the quality checks you’d normally hold in your head. You can’t download that anywhere; your agent can only learn it from you.
That’s also why it compounds. Every skill taught makes the agent more capable for good — the goldfish becomes muscle memory, and the gap between your agent and a fresh install grows every week you use it.
The three beliefs that stop people
“Teaching an AI means writing code.” No — one word and plain English. Hermes authors the skill itself; I watched a free local model do it in about 25 seconds.
“My agent forgets everything, so why bother?” That’s precisely what this fixes. The learned skill is kept — forgetting is the problem /learn exists to end.
“I’ll wait until the tools are perfect.” An agent that learns gets further ahead every week it’s used. Waiting just widens the gap between you and the people teaching their agents today.
Practical guardrails still apply: you see the skill file before it saves (keep the approval gate on until you trust the loop), a clearer source produces a better skill, and anything wrong is a one-line markdown edit — the audit habit is in memory management.
What to teach it first
The workflow you re-explain most often — highest payback, first night.
A guide you want implemented, not just read — point /learn at the article and the method becomes a skill.
A job it just did well — capture the win while the session is fresh.
Your quality checklist — the rules you normally enforce by hand.
Skills are shareable too — one markdown file others can install in a single command — which is how a personal workflow becomes a team asset. The wider skill strategy sits in Hermes SEO skills.
FAQ
What is Hermes /learn?
The learn-anything engine: point Hermes at a source — URL, PDF, folder, notes or a finished job — and it distils it into a reusable skill it writes itself and keeps forever.
How is /learn different from the /learn command page and skill page?
Same feature, three angles: this is the overview, the command guide covers running it step by step, and the skill page covers the skill.md file it produces.
Do I need to know how to code?
No. You describe things in plain English and Hermes writes the skill itself — even a free local model manages it in under half a minute.
Does it really remember forever?
Yes — the skill is saved as a markdown file that persists across sessions and loads on demand or on trigger terms.
What should I teach it first?
The workflow you re-explain most often. Five minutes once replaces a few minutes every session, forever.
What if it learns something wrong?
You see the file before it saves, and afterwards it’s just markdown — fix one line, delete it, or rerun /learn with a clearer source.
The bottom line on Hermes /learn
Hermes /learn turns re-explaining into teaching: one word, any source, a skill the agent writes itself and keeps for good — loading only when needed, working on every surface, compounding every week. Stop being the only one who remembers how your workflows run, and teach your agent its first skill tonight.
