The Hermes /learn skill is exactly what it sounds like: Hermes can now learn from any guide you give it — a URL, a PDF, a folder, your notes — and turn what it read into a skill it keeps forever, saved as a readable skill.md file. Here’s what gets saved, when it loads, how to fix a bad one, and how to share them.
Short answer
- The /learn skill = Hermes learning from any guide you give it — URL, PDF, folder or notes — kept forever as a skill.md it writes itself.
- Loads three ways: /skill by name, by asking for the task, or on trigger terms.
- Bad skill? You see the file before it saves — fix one line, or rerun with a clearer source.
- One markdown file = shareable: others install your recipe in one command.
What a Hermes /learn skill looks like inside
When /learn runs, the output is a skill.md — a plain markdown file, not a black box. The real one from my demo (I didn’t touch a line of it) contained:
- Name and description — what this skill is and when it applies
- Version — so updates are trackable
- What it does and when to use it
- The procedure — the clean step-by-step recipe
- Pitfalls to avoid — the mistakes baked in as warnings
That structure is why these are trustworthy: a skill is a briefing document you can read, not behaviour you have to infer.
When a /learn skill loads
Three ways, from most to least explicit:
- By name:
/skill agent-operating-systemloads it on demand. - By request: ask for the task by name and the agent pulls the recipe itself.
- By trigger: skills auto-load when related terms come up — my agent-OS skill wakes on mentions of agent OS, mission control, memory galaxy or the Boardroom stack. No command at all.
And crucially, skills load only when needed. They don’t ride along in every session bloating context — the recipe lives in the agent, not in your head, and not in your token bill either.
🔥 Want this set up without the guesswork? A library of skills that load themselves is the closest thing to compounding you can install. 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
Fixing a bad /learn skill
What if it writes a bad one? You see the file before it saves — with the approval gate on, nothing is kept until you say yes. (Mine works instantly because I’ve turned the gate off; keep yours on until you trust the loop.)
- A skill that’s slightly off: fix one line — it’s just a markdown note.
- A skill that’s badly off: delete it like any other note, or run /learn again with a clearer source.
- Nothing is locked in, and a wrong skill can’t break anything else — it sits inert until loaded.
This is the same audit discipline as the rest of memory: read what your agent learned, because self-improvement compounds whatever’s written — the routine is in memory management.
Sharing /learn skills
Because a skill is one markdown file, it travels. Publish it, hand it to a teammate, share it with a community — they install it in one command and get your recipe with their own keys and context. I share mine inside the community exactly this way.
That’s the quiet significance: /learn doesn’t just teach your agent — it turns your working knowledge into an asset other people can run. The command mechanics live in the /learn command guide; how skills fit the wider architecture in the Hermes architecture breakdown.
FAQ
What is a Hermes /learn skill?
The artefact /learn produces: a skill.md markdown file holding a name, description, version, when-to-use guidance, the step-by-step procedure and pitfalls to avoid.
How do skills load?
By name with /skill, by asking for the task, or automatically when related trigger terms come up in conversation — and only when needed, so they don’t bloat context.
Can I edit a learned skill?
Yes — it’s a markdown note. Fix a line, rewrite a step, or delete it entirely. Nothing is locked in.
What if /learn writes a bad skill?
With the approval gate on you see the file before it saves. Approve, edit, or rerun /learn with a clearer source — a wrong skill can’t break anything else.
Can I share skills with other people?
Yes — publish the file and others install it in one command. Your workflow becomes a portable asset.
Do skills work across chat, voice and messaging?
One skill follows the agent everywhere it lives — dashboard, terminal, Jarvis voice and messaging apps.
The bottom line on Hermes /learn skills
A Hermes /learn skill is a readable recipe — name, procedure, pitfalls — that loads by name, by request or by trigger, costs nothing when idle, fixes with a one-line edit, and shares as a single file. Read what your agent writes, keep the gate on until you trust it, and your skill library becomes the asset a fresh install can’t copy.
About Julian Goldie
I run Goldie Agency, a 7-figure SEO agency, and teach this daily on a 400K+ subscriber YouTube channel. 240+ client projects on Upwork at a 100% job-success score, 10+ years through every major Google update. My systems are in the AI Profit Boardroom; my link building book is free here.
Related reading
Last updated August 2026. This is the living guide to Hermes /learn skill — it gets updated as the tools change.