AI Guides › Skills & Agents
What A "Skill" Actually Is — And Why The Word Undersells It
By Nigel Guy · 2 min read
"Skill" sounds like a party trick — something Claude can now do that it
couldn't before, like a new app icon. That framing makes people either
under-use skills entirely (treating them as optional extras) or over-trust
them (assuming a skill means "handled, don't check it").
The rule: a skill is a written procedure with a stated point and stated
limits — closer to a checklist you'd hand a new employee than a feature you
switch on.
The mechanism
- A skill states its point before it states anything else. The good
ones open with a single sentence — the "one rule" — that the rest of the
document exists to unpack. If you can't find that sentence in a skill
you're about to use, that's worth noticing.
- A skill has steps, not vibes. A real skill reads like a numbered
procedure: do this, then check this, then do that. If it reads like
inspirational advice, it's not really a skill yet.
- A skill names its guardrails. What it won't do, what it can't promise,
where a human needs to step back in. This section is where you can tell a
carefully built skill from a rushed one.
- A skill has a trigger. A description of exactly when it should kick
in — vague triggers ("use for productivity") mean it either never fires or
fires when you didn't want it to.
- A skill is reusable, not a one-off. If you're only ever going to do
the thing once, you want a good prompt, not a skill — see "The Difference
Between A Skill And A Saved Prompt" for that line.
What to skip
Skip treating "it's a skill" as a reason to trust the output more than you
would a plain answer. A skill is a better-organised request, not a
certification. And skip installing a skill because the description sounds
impressive — read what it actually does, in plain terms, before you rely on
it for anything real.
Guardrails
- A skill without a guardrails section is a skill you should read carefully
before you use, not one you should assume is safe by default.
- Skills rot. The tool or platform they were written against changes, and a
skill that worked perfectly six months ago can silently start doing the
wrong thing. Re-check periodically, don't "install and forget."
- The word "skill" doesn't change what's actually true of any AI output:
review anything that matters before you act on it.
All 751 AI guides · JulieMango plans from £17/mo