Skills
Teaching New Tricks: give your agent abilities it calls on when needed.
What a skill is
A skill ↗ is a reusable instruction package (SKILL.md). Front matter names and describes it; the body teaches the procedure. Hermes sees names and descriptions first — a vague description means the right skill may never load.
Skills use progressive disclosure: the catalog stays cheap; only the selected body enters the prompt. That is how a large library avoids burning the context window every turn.
In the Think → Act → Observe loop, a skill mostly shapes Think (and which tools to pick). It is not itself an Act unless it triggers tools. Tools execute; skills teach.
/learnauthors a skill for you from a folder, a URL, a PDF, or a workflow you just ran — you rarely need to writeSKILL.mdby hand.~/.hermes/skills/is the source of truth. Skills follow the agentskills.io ↗ open standard, so a skill written for another harness can work here — but Hermes only reads it once you list its folder underskills.external_dirsin~/.hermes/config.yaml. Dropping files elsewhere and hoping is the usual reason a skill "never loads".- Install narrowly; run security scans on hub skills before trust.
- Pin or curator ↗: critical skills should not vanish to automated cleanup when you outgrow defaults.
Where skills come from
Four sources, in the order most people meet them:
| Source | What you get | Trust |
|---|---|---|
| Bundled — the Skills panel in Desktop | What shipped with Hermes, ready to run | Highest — it came with the app |
Official hub — hermes skills install official/… | Maintained hub skills — useful, still read them first | High — but still read what it does first |
| `/learn` — you point it at a source | A skill written from your own book, docs or workflow | Yours; nothing arrives from a stranger |
| Anywhere else — GitHub, other harnesses | The agentskills.io ↗ standard makes these portable | Unknown — scan before you trust |
Which ones are actually worth it
We deliberately do not hand you a top-ten list — the catalog moves, and a skill you do not use is pure cost. Use the test instead: install it when you have already done the task by hand at least once and want it done the same way every time.
- It has a real job this week. Not “useful someday” — an actual repeat task.
- Its description is specific. Hermes picks skills by name and description; vague ones never load at the right moment.
- You could describe what it does to someone else in one sentence. If not, you have not read it.
- It survives the budget test — see the next section. A shelf of unused skills costs tokens in every session.
unbroker-ge and unbroker-eu from lesson 06 — pick the lane that matches where you live. Official unbroker is the US people-search tool; only add it if you also have a US footprint.- Open Skills in Hermes Desktop and read the list you already have.
- Copy the names into a note in your vault, one line each on when you would use it.
- Run or inspect one bundled skill, following the work-with-skills guide ↗.
- Install at most one new skill — and write down why you needed it.
Skills tax the fixed budget
The skills index and every enabled tool schema load into the empty-session budget (see module 06 — context budget). A large “install for later” catalog costs tokens before any task starts.
- Open the context / usage view in Desktop. Note the skills / tools share.
- Disable or uninstall one skill you will not use this week. Re-run the audit.
- Record before/after in
skills-notes.md(one line each is enough).
Proof
You install twenty skills and use none of them this week. What does that cost?
Pick one — the answer is revealed straight away.